Transcript Summaries

Corporate transcripts are packed with critical information. But they are also verbose and full of ‘fluff’. A good summary should extract the vital information and discard said fluff. They should do this one task well, and they don’t require interactivity.

When embarking on this project, I observed how peers tackled this. There were a lot of examples to choose from (hence, table stakes). I noticed a lack of vision and understanding of what a user wants a summary to be. Most often, the summaries would whittle down a transcript into a handful of bullet points, which read more like a high-level earnings news highlight— “American Express reported $X.XX in EPS… Management suggested they are optimistic about YYYY”.

On the depth scale, the dial was turned all the way to ‘As little information as possible without the page being empty’. A typical earnings call transcript will take an avid reader ~20 minutes to read, while most ‘summaries’ I’ve seen out there take ~20 seconds. They optimise for extreme brevity. We took a slightly different approach, aiming for ~3 minutes to read.

We didn’t want them to be ‘as concise as possible’ at the expense of, you know, informing the reader. Instead, we focused on reorganising the important information into easier-to-consume chunks and removing the filler.

Every unique type of event (earnings call, M&A, special call, etc) dynamically takes on its own structure and formatting to organise the summary into segments that matter and are relevant to that particular event. In what is arguably the most diverse event type, earnings calls, the summary will adjust to centre around the individual qualities of a company. Sections may include KPI trend tables, geographic commentary, specific business segment results and commentary, risks, guidance, and a concise Q&A breakdown.

As a result, I believe these summaries are higher quality, balancing the verbosity of a full transcript vs. the lack of depth in providing a few bullets. I implore you to try a few and let me know what you think.

The reason we optimised for organisation over brevity is that we felt this is more aligned with the actual workflow. The problem users have is not that they want to consume an entire transcript in 20 seconds2 (which is not actually possible). The problem is that transcripts are conversations, and the critical information is buried amongst a large corpus of word salad. Extracting that critical information and presenting it legibly was the goal. And if that takes a few more minutes to read front to back, that’s fine.

While a seemingly simple3 user experience, transcript summaries are one example of how LLMs can be used to enhance an existing workflow. The user doesn’t necessarily care that this summary was created using an LLM. The important thing is that an LLM made this workflow possible and seamless. This is the theme I want to pick at a little more. Over time, I am gravitating towards the idea that incorporating AI into a product is less about slapping AI labels onto things and more about enhancing or creating new workflows.

LLM Workflows are Everywhere

The wide-scale adoption of LLMs has captured society’s attention. There is no doubt about that. As they endure their ‘hockey stick’ moment, part of what is so exciting is the vast potential in both consumer and enterprise applications. A huge portion of the world’s inhabitants have now tasted one of the most mundane applications of modern LLM use cases, the chat-based interface, through the likes of ChatGPT, Gemini, and others.

While it feels criminal to call this mundane when it still (to me at least) feels like magic, the reality is that people have become accustomed to this offering. We are getting past the phase of accepting suboptimal responses, and we expect them to be high quality. In November 2022, ChatGPT 3.5 was born, and the genie escaped the bottle. Five days later, it reached 1 million users. A couple of months later, it hit 100 million users. Today, the likes of OpenAI’s ChatGPT and Alphabet’s Gemini have monthly actives (MAUs) in the hundreds of millions. I am sure this moment in history will be documented for decades to come.

Thanks to the desire of gargantuan enterprises like Nvidia, Microsoft, Meta, Alphabet (and more) to pump capital into this space for fear of being a victim of the innovator’s dilemma, this product has already become somewhat of a commodity. It’s not incredibly expensive to the consumer, it’s widely available, and there are several institutions that develop and offer these frontier models in a consumer-friendly package (ChatGPT, Gemini, etc). There are thousands of enterprises building products that are essentially ‘wrappers’ on top of these models, too.

The latter is particularly exciting. As with most commodities, there is an oligopoly of firms supplying the raw materials, but the exciting part is the layer above in which those goods are transformed to create an unfathomable number of ‘new’ product offerings. It is those offerings which are so exciting.

We’ve already seen an abundance of ‘AI Helper’ style tools pop up in all of our favourite software. We’ve also witnessed early innings of how these models can be blended with hardware applications in cases like Meta’s ‘Ray-Ban Display’ glasses, which pair with a neural wristband, exploring the potential of the ‘next screen’ to displace smartphones.

I won’t pretend to have any insight into what this space looks like in ten years, but I am excited about what the world’s greatest minds come up with.

LLMs are Eating the World

Some time ago (2011), the phrase “software is eating the world” was coined by Marc Andreessen in a Wall Street Journal article. There, Marc was highlighting a trend that had already been underway for some time.

“We are in the middle of a dramatic and broad technological and economic shift in which software companies are poised to take over large swathes of the economy.”

In this article, he cited Netflix replacing Blockbuster, Amazon transforming retail, and software transforming mechanical vehicles into computers on wheels as early examples. It became the defining mantra of the 2010s. It feels like we are now seeing a similar theme, this time orchestrated by LLMs.

Today, I am going to be focusing on the financial space, but I suspect the ramifications on labour demand and consumption behaviours will reverberate throughout the global economy for years to come.

Customer support is a well-documented use case. Companies across all sectors and industries have already implemented LLM-powered support agents that essentially hook up to their internal knowledge base to solve common and simple user support cases. We did this at Koyfin over a year ago now. Here are a couple of examples, but there are many more.

1. Intuit (INTU) claimed they’ve seen a “15% reduction in contact rate in the past year” for their customer support desk after implementing GenAI4.

2. London Stock Exchange Group (LSEG) told investors that “More than 80% of customer queries are now resolved using AI customer support tools” in July5.

3. Lyft (LYFT) issued a press release in February 2025 suggesting that their “customer care AI assistant has reduced the average customer service resolution time by 87% and is resolving thousands of customer requests each day”.

Outside of directly replacing labour units in a one-for-one manner, you have to consider the argument that LLMs make a single unit of labour more efficient. Suppose it helps a developer ship 25% more code. Does that mean you now only need four developers instead of five? If product requirements and PRDs can be spun up in seconds with the help of co-pilots, does your organisation need so many product managers?

In a 2025 paper6, funded by Microsoft, the authors suggested that job titles such as interpreters and translators, historians, sales representatives, writers, authors, telephone operators, concierges, data scientists, models, and market research analysts (among others) were most at risk of displacement by AI.

I have no idea what ultimately comes of this, but I find it likely that both arguments will be true. Namely, that LLMs will make certain labour units more efficient, and that this will result in fewer units of labour being required for particular roles or functions. Similarly, I feel that efficiency gains are context-dependent. This idea of labour being replaced by technology is not exactly a new phenomenon.

I was watching a documentary about Cadbury’s (the chocolate maker) recently. In that documentary, they spoke with veterans of their Bournville production facility based in the UK. In 1878, the site employed ~200 workers. By 1899, the site had more than 2,600 employees. By the early 1900s, a village was formed7 to house employees.

“George Cadbury was a housing reformer interested in improving the living conditions of working people in addition to advancing working practises. Having built some houses for key workers, in 1895 he bought 120 acres near the works and began to build Bournville village”.

Bournville village was born. From thereon, the village developed into a community with its own shops, schools, playgrounds, colleges, and various other recreational facilities. It is thought that by the mid-20th century, the Bournville facility employed between 7,000 to 9,000 people. The veterans from the documentary reminisced about this period and talked about how this gradually came to an end as more machinery was introduced, displacing the need for thousands of workers. An older gentleman, being interviewed on the facility floor, looked around the room, noting that hundreds of workers would clock in and out of the building every day and that now it only requires a handful of bodies to ensure production runs smoothly.

This trend was accelerated in the 2010s after Cadbury was acquired by Mondelēz International, which sought to industrialise the plant further. Today, the plant employs some 1,000 people, and the village ceased to be ‘for employees only’ decades ago. Such is the cost of progress and efficiency.

I have no doubt the age of LLMs will cause mass disruption to the workforce as we know it today, but I am equally optimistic that we humans will figure out some other use for those idle hands in the long term. We have a knack for this, and history confirms as such. Nowhere will this workflow transformation be more visible or fiercely defended than in financial data.

LLM Workflows in the Financial Space

In the financial world, there are a small group of kingpins specialising in the realm of financial data. Companies such as London Stock Exchange Group (LSEG), Standard & Poor’s (SPGI), Factset (FDS), Bloomberg, and Morningstar (MORN). I will lazily refer to these companies as the ‘Big 5’. Between them, they carry a significant competitive advantage in the fact that they “own” and are the licensors of datasets covering, but not limited to, global ETFs, Mutual Funds, Equities, SMAs, Credit Ratings, Fixed Income Data, News and more8.

To use this data commercially, as a business, you likely have to go through at least one of these companies to license their data.

Focusing on the financial data & analytics market for a second, let’s say there are two levels. The base layer is the aforementioned companies. They own the data which powers the analytics. The layer above is the consumer or enterprise-facing platforms, which organise this data and present it in an appealing way for practical use. So that professional investors, advisors, and even retail investors can analyse markets, track portfolios, and so on.

The aforementioned companies also play in this space. Morningstar has Morningstar Direct, S&P has Capital IQ, Bloomberg has the Terminal, Factset has its version of the terminal, and LSEG has LSEG Workspace. This layer is also home to an abundance of other platforms that use the kingpin’s data. Examples include Quartr, Koyfin, TradingView, Ycharts, Fiscal AI, and many more. Some, like TIKR, are almost exclusively wrappers.

Others, such as Quartr, Koyfin, and Ycharts, have carved out niches in addition to serving a wider audience or have stitched together such a wide array of data from multiple vendors that they have less concentration risk. These companies offer tooling that does not exclusively resell vendor data at a lower price point. They provide value-added functionality such as portfolio analytics, integrations, reporting, and so on.

Fiscal AI is an interesting one because, on the face of it, they are a retail-focussed platform providing data on equities and funds, likely from ~3 big vendors. However, this particular company has taken steps to shed the handcuffs of the Big 5 and is now creating its own dataset. The first example is their innovative approach to Segment & KPI data. Instead of licensing, they collect it themselves. Which means they can now license it to others (Ycharts being one example, Perplexity being another).

More recently, they have made efforts to start collecting standard financial data from filings, displacing their need to license data from Capital IQ. A filing comes in, they scrape it using LLMs, they populate the data in their database, and this flows onto their platform. Fiscal AI is, in effect, an API business.

To simplify, think of it this way:

• Morningstar is in the Vendor business: They collect, aggregate, and clean data, store it in databases, and license it to other businesses.

• Koyfin is in the Value-Added Distributor business: They license data from several vendors and distinguish themselves by offering value-added services such as advanced visualisation, data transformation, analytics, and unique tooling.

• TIKR is in the Reseller business: They license data from vendors without much added value and resell it.

The important distinction here is that distributors and resellers don’t own the data. The equally important caveat, however, is that the industry isn’t quite so simple. Morningstar, Capital IQ, LSEG, Bloomberg, and the other vendors are rarely ever pure vendors.

Capital IQ, for example, is a vertically integrated data provider. They own the underlying data and the software layer which commercialises it. On top of that, it also relies (albeit to a significantly lesser extent than non-vendors) on third-party vendors for some of the data it licenses and includes in its platform. The same is true for Morningstar and LSEG.

What is interesting about Fiscal AI is that they, more than anyone, pose a direct threat to the likes of Standard and Poor’s Capital IQ. Fiscal AI is attempting to compete at both layers of the market, as a vertically-integrated vendor.

Competitive Advantage Narrative

The rise of LLMs has people questioning if this competitive advantage will hold for the Big 5, given that it is now easier to extract this information independently, both for consumers and for enterprises. Thus, bypassing the expensive licensing costs. In practical terms, this suggests the barriers to entry have declined. Fiscal AI is evidence of that trend. Initially starting life as a reseller, the company has since become a vendor.

As a product manager who is often in contact with data vendors, I have noticed a spike in start-ups aspiring to replicate this model. Many of which are using LLMs to extract data from filings to create a database of global financial metrics, which they can either use to create their own platform or sell to other platforms.

This theme appears to be puncturing the Big 4’s returns (excluding Bloomberg, which is not publicly traded). Since 2024, the companies have declined between 28% (in LSEG’s case) and 42% for Factset.

I suspect this is a great thing for the consumer. As Peter Thiel once said9, “competition is for losers”. There are most certainly other factors at play here. You can paint any narrative about historical price action that fits. Whatever the case, I find the argument to be slightly flawed.

Take LSEG as an example. It’s an incumbent. One of the criticisms of products like LSEG Workspace (and other ‘terminals’) is that they are slow to innovate and have UI that looks like it was etched into stone during the 1990s. They are often built on older technical infrastructure, making evolution slower. Contrary to that argument, Workspace actually looks modern with a modular interface. The same can’t be said for all platforms.

Anyway, LSEG has been talking about AI workflows for years. In their H1 2025 earnings call, they highlighted several initiatives ranging from agentic analytics, co-pilots, natural language search and data prompts, and AI-generated news.

LSEG’s CEO, David Schwimmer, has been vocal about how important these advancements are and takes a view of welcoming the technology shift. In his view, AI enhances, not replaces— “the future is AI integrated into a desktop, not AI replacing a desktop”.

He is referring to the idea that because AI lowers the barriers, it means the competitive advantages of LSEG (and other vendors) are at risk. In reality, cheaper alternatives have been available for decades. Nothing personifies this better than the ‘Bloomberg Killer’ idea.

Now and then, a tweet will go viral that either;

• A: Questions why Bloomberg hasn’t been replaced yet,

• B: Suggests [product X] will replace Bloomberg

• C: Claims it would be easy to build a Bloomberg killer

• D: Some mixture of the above

It almost always captures a lot of attention. The author is almost always entirely ignorant. Here’s the reality of the situation:

1. People don’t understand why Bloomberg is an institution: Bloomberg is like the Swiss Army Knife of markets on viagra and steroids. It does everything. It has a professional network effect. It provides niche datasets that nobody else provides. It is, in effect, not replaceable by some users because it’s a mission-critical software. Commonly, people look at Bloomberg and then compare it to Yahoo Finance and say something naive like “Yahoo has live stock quotes for free”. They are not serious people.

2. What is not happening: Bloomberg is not in the business of competing with retail platforms for mass market adoption. The price point alone is the greatest signal of that. Most cited ‘Bloomberg Killers’ are retail platforms. It’s apples to oranges.

3. What is happening: There is genuine displacement occurring for investors who use Bloomberg, but only use a fraction of the more basic capabilities of the Terminal. To give just one example, Financial Advisors. They can live without Bloomberg and substitute their needs with platforms like Factset, Koyfin, and Morningstar. Another example is Investor Relations departments, which could replace a terminal with Quartr. However, this is not a new trend.

Back to LSEG for a moment. David Schwimmer’s confidence is warranted. There are undoubtedly going to be a lot of great innovations in the retail space. LSEG are selling to the same people that the rest of the Big 5 are. Those companies are all incorporating LLMs into their workflows. So while it feels like groundbreaking stuff is happening, I sense that it’s more ‘business as usual’ for the most part. At least in the medium term.

During the H1 call, they were also asked about this theme of new entrants, to which Schwimmer expressed an opinion I largely agree with. It’s less about having a cool UI; the advantage still lies in the quality of data.

LSEG are not selling to 20-somethings with a $25k portfolio. That’s not their market. Any professional or serious investor cares about data quality, and that is the Big 5’s manifesto. They own it, they have large quality assurance teams to maintain it, and they provide estimates and other proprietary data. The question from the call reads as follows:

“There’s been some worry in the market about that as a sort of disruption risks in workflow to the industry as a whole from sort of new fintech AI platform”.

Schwimmer’s response, which I will largely leave un-edited, reinforces the idea that professionals care most about data quality, availability and, importantly, auditability. These less price-sensitive customers want the best.

“You can have a really cool user interface, but the quality of the product depends on the quality of the data. And we have the broadest, the deepest, the highest quality data sets. The second point I would make is that users don’t want just a sophisticated -- I used this phrase before, sophisticated chatbot. They want that great AI functionality, and they want that AI functionality embedded in all of the other workflows. And that can be news curation, charting, order and execution management, analytics, risk systems. We provide all of that in our user interface, in Workspace. And we are building AI into all of that.

And so I think that’s a really important aspect of this in terms of -- again, you can have, as I said, sort of a cool user interface, but that on its own doesn’t get you there in terms of what our industry is looking for and what our industry expects. Our users also want that workflow and that AI capability to be interoperable with their enterprise workflow. And then the last point, I touched on this earlier, but the last point, I think you cannot emphasize enough is that users want data that they can trust and they can rely on. And so with respect to some of the new offerings that are out there, and we’ve seen and we’ve played with some of the flashy demos and other things. But in one of those in particular, that has gotten a lot of attention, if you go into their own press release, there’s -- that press release points to their -- there’s a link in that points to what they’re relying on and their accuracy in finance is at 51%. And their accuracy in market analysis, they don’t define what market analysis is, but I think we do a lot of market analysis. Their accuracy and market analysis is 14%.

So I think coming back to your question in terms of some of the fintech offerings, they may be cool and they may be fun to try out, but they cannot meet the demands of this sector, the demands of our customers today. And I think that’s really important to keep in mind.

We are embedding AI functionality in our offerings, and we’re certainly not sitting still. So I think -- and there’s more to come. So as I said before, I like our positioning. We welcome the tech innovation. We welcome the tech change.

Two of the biggest takeaways after reading this for me were the comment on the sophisticated chatbot and the importance of audibility, which is essentially the element of trust in the output.

When he said “users don’t want just a sophisticated chatbot”, this resonated with me. At Koyfin, we’ve been patient with how we incorporate LLMs into our platform, and only now do we have an exciting roadmap of features. I am grateful we were patient because we’ve had the advantage of learning from the mistakes of peers.

I will preface this next part with some admiration first.

Fiscal AI’s Evolution

I have enjoyed watching the trajectory of Fiscal AI (formerly Finchat, formerly Stratosphere). I have huge respect for Braden (Co-founder) and their team. They do a lot of things right, and it’s refreshing to see a business that ships fast, ships hard, and spends time on features they know users will love. While there is some intersection between users (i.e some users will directly compare Koyfin and Fiscal AI in a purchasing decision), I believe the goals of each company are quite removed. I won’t comment too much on that, but the intersection exists at the retail investor level. Where it differs is that Fiscal AI are an API business, and Koyfin have a growing market share in serving Advisors (an area Fiscal AI is not currently in).

To put it another way, the two companies are as likely to be competitors as they are partners10. Koyfin licenses high-quality data. Fiscal AI sell high-quality data to enterprises.

Admiration segment complete, one learning I have passively absorbed from Fiscal AI is that they learned that users don’t want a “sophisticated chatbot” the hard way.

Several years ago, Fiscal AI was called Stratosphere. Stratosphere was a reseller platform. Then one day, the same minds that built the Stratosphere created a beta application called Finchat.io, which was essentially an LLM wrapper come chatbot which plugged into Stratosphere’s data. This was incredibly novel at the time, I must say. While the two products were connected on the backend, they were two separate offerings. One day, around 2023, the team decided to merge the two and create Finchat.

Finchat would essentially rebrand the Stratosphere platform, make it look a lot cleaner and more terminal-like. Only now, the Co-Pilot (a rebranded version of Finchat.io) was the central feature on the platform. The co-pilot was a ‘do everything’ chat interface powered by LLMs and Fiscal AI’s data. It was an impressive feature, but I got the sense it was a jack of all trades, master of none. You could try to run a screener, generate a report, ask it questions about filings, and tell it to build tables. Yet, it was never clear what its superpower was.

Around a year later, the co-pilot, which was once the primary driver of the platform, was relegated to a left-hand-side navigation element. It became more of a ‘feature’ than a flagship staple of the platform. Then, in 2025, another rebrand came along. Finchat became Fiscal AI. In a press release statement, the company claimed:

“As we’ve grown, so have our ambitions — beyond the Chat interface. Fiscal.ai represents a bold new vision: to power modern financial data infrastructure through our Terminal and APIs, transforming how the world accesses and uses information in capital markets”.

Behind the scenes, the company was quietly planning to shelve the co-pilot. On August 19th, 2025, Braden would reply to a user on Twitter that “copilot is being retired soon”. As of today, the co-pilot no longer exists on the platform.

Braden shared the story on a podcast I listened to a few months back. There, he explained that they started to license the API for this co-pilot, it grew traction, and then stuttered as the churn rates were high. To their credit, they bit the bullet and moved forward quickly. I think this was a smart move on their part.

Where Fiscal’s co-pilot failed, I believe, is due to generalisation. It lacked a workflow-specific focus and was never quite good enough at any particular flow to warrant that desired stickiness. It was arguably better than ChatGPT at some flows, but not by a significant enough margin.

There needs to be some kind of super-intuitive and workflow-driven element to an LLM feature; otherwise, it just becomes a feature that tries to do too much at once. A sophisticated chatbot, as Schwimmer puts it.

Workflow Enhancement

Let’s use Quartr as an example, another player in this space I have a lot of admiration for. They introduced a chat interface feature (I know I just poo-poo’d this, but hear me out). The important difference, however, is how this product is marketed. From the get-go, it’s incredibly clear what this feature is for. It queries information from documents (filings, transcripts, press releases).

You ask questions, it searches the database and spits out a response. In the mind of the user, you have the expectation that it does this workflow well. Unlike Fiscal’s co-pilot, which had less of a defined use case, users won’t feel slighted by the fact that it doesn’t function as a charting tool or a screener, because it was never marketed as such. I have used Quartr’s document query engine, and it was a remarkably better experience.

These days, many investors are downloading transcripts and filings and dropping them into generic LLMs like ChatGPT or Notebook LLM. The obvious problem here is that (a) these LLMs are not tailored to this workflow and (b) they don’t sit on top of a structured database of the required data. In other words, you need to give them the data, then they analyse it. Quartr’s product is built on top of the data, meaning you don’t need to carry your digital briefcase of filings around with you. The irony here is that Fiscal’s co-pilot could perform the same kind of use case, and it was launched almost 2 years prior. I have no idea about the commercial success of Quartr’s product, but it felt like a solid feature when I used it. There was more of an intuitive moment of “oh, this is what this thing does”.

Takeaway

Over time, I am gravitating to the idea that LLM-enabled tools need to be built to enhance a workflow. Not starting from the LLM and saying, “How many different things can we make this do at once?”. Alternatively, they should create brand new workflows that were previously not possible. Genuine novelty, unlocking new patterns while avoiding the trap of confusing novelty for utility.

I am incorporating this belief as we work on the exciting roadmap we have at Koyfin. I’m dying to share what we are working on, but given that some are truly unique in nature, I will be keeping the wraps on it for now. But what I will say is that it plays into this idea that you should start with the workflow first, and imagine how LLMs enhance that… with a little extra.

Like any good product/feature, it should be intuitive and obvious what it does, and not require a ton of instructions and articles.

Thanks for reading,

Conor

This is the longest I have gone without sending a newsletter since I started this publication over five years ago— the last entry being in May. During this absence, I have been busy.

In my personal life, I spent a month in Asia, where I visited Thailand for the first time. When in Ko Samui, I decided it was about time I proposed to my partner. I also had the pleasure of visiting Northern Ireland for the first time. I can now say I have visited every country in the United Kingdom.

At Koyfin, the business and the team are growing, and i’ve never been more excited about our trajectory and the opportunity we have ahead of ourselves. It’s been eight months since I last reflected on what we have been up to, so today’s newsletter will centre around that.

Below is a variety of the things we have released at Koyfin in 2025, as well as some commentary on what is coming down the line.

1. Shipped ✅

Features and improvements we have shipped to users.

I never tire of the satisfaction that comes from marking a highly upvoted feature request from our community feature request forum as ‘Complete’. The ones that garner more than 100 upvotes are particularly pleasing. This year, we have knocked a few of those off the list— including desktop alerts (300 upvotes) and adding historic data to our Watchlist & Screener (200 upvotes). We are also weeks away from unveiling another, the Interactive Brokers integration (150 upvotes).

1.1 Desktop Alerts

Desktop Alerts had been highly requested for years. We initially launched alerts on Mobile in 2023, with limited functionality of only price alerts for US stocks and Press Releases. The wait is finally over, and I believe that we went above and beyond to deliver a robust v1 of desktop alerts.

When we decided to work on desktop alerts, we wanted to ensure that this would be a dramatic improvement to what we supported in mobile, and in a way that complemented our desktop platform. In our first release of desktop alerts functionality, we:

1. Expanded coverage to include global equities, ETFs, and Mutual Funds.

2. Broadened alert types to include price, technicals, valuation, and documents (press releases, news, filings, transcripts).

3. Added the ability to set notes on alerts.

4. Supported setting alerts on Watchlists and Portfolios.

5. Introduced a notifications side panel to see notifications triggered live throughout the day.

   1. This panel is also interactive and will direct you to the source of the notification, whether that be a transcript notification or a price graph.

6. Created an alerts management dashboard that will allow users to edit, pause, and delete their alerts, as well as search and sort through their existing alerts, or manage their notification settings.

Alerts can be created or accessed through the ‘My Alerts’ button that will now feature throughout the desktop platform. Notifications delivery is also supported via Email, Mobile, and In-App for Desktop. There is more we can do with alerts, but this has proved to be a great foundation.

1.2 Historic Data in Watchlist & Screener

The concept is a simple one. Previously, in Koyfin’s watchlists, screeners, market and scatter plot graphs, the biggest limitation was a lack of time series analysis. Despite having thousands of metrics to choose from, you were limited to a point-in-time perspective. As an example, you could add EBIT margin to a watchlist, but the choices you had were the latest Fiscal Year (FY), Fiscal Quarter (FQ) or Last Twelve Months (LTM).

This update changed all of that. We created a new structure for our data in the column selection, meaning that metrics now have subcategories for FQ and FY history, as well as Historic Averages.

In a Watchlist, this made it possible to present the historical years or quarters of EBIT margins, to see how the metric has evolved. These historic metrics are also available for use in custom formulas— as demonstrated below, I created a custom formula to show the difference, in basis points, between today’s margin and the 5Y average.

It doesn’t take much imagination to consider how greatly this improves the screener's usability, either. If I wanted to screen for companies that had reported negative EBIT margins for the last 5 fiscal years but have flipped to profitability in the last twelve months, this is now possible, whereas before it was not.

In areas like Market Scatter, users can now more readily visualise data on a time series basis. In the above example, I have plotted the current vs 5Y average EBIT margins of the Semiconductor industry and overlayed a linear regression line to observe which companies are reporting higher margins relative to history.

One of the joys of being a PM at Koyfin is getting to help build features that I, as a user, am passionate about. This was certainly one of those occasions.

We are not done with historic data yet, however. We are going to follow up by adding more, such as:

1. Historic Trailing Growth Rates: For example, the growth of revenue over the last 3 years, 5 years, 10 years, etc.

2. Historic YoY & QoQ Growth Rates: For example, the YoY growth of each quarter going back the last 3 years. This will be great for visualising acceleration and deceleration.

3. Historic Actuals, Estimates, and Surprises: See how consistently companies beat estimates.

1.3 Portfolios & Integrations

Koyfin is loved by individuals and bottom-up fundamental analysts. But we also specialise in solving problems for Advisors. Back in December 2024, I wrote about how focused we have been on this demographic and how far we had come. At that time, we unveiled a new Advisor tier in our pricing plans. We now have two plans, Advisor Core and Advisor Pro.

The pace has not slowed. Below is a hand-picked selection of evidence that we are shipping features advisors have asked us for.

1.3.1 Integrations

In December last year, we had just launched our maiden integration, Schwab Advisor Centre.

Since then, we’ve added support for Altruist, Black Diamond, TradePMR, Interactive Brokers, and Orion. Our next conquest will be Fidelity, which we hope to make available later this year.

1.3.2 Client Portfolios v2

Client Portfolios was a feature we developed to push forward Koyfin’s status as a Portfolio Management System (PMS)— designed to support thousands of portfolios and integrate with an advisor’s custodian or other PMS tools. This gives advisors an efficient and fast way to visualise exposures, the P/L of their client’s assets, and seamlessly generate reports.

In our first version of Client Portfolios, the structure was flat— meaning that all portfolios were on the same level. We’ve since introduced a three-tiered structure:

1. Households: Highest level, contains Portfolios and Accounts.

2. Portfolios: Mid-level, contains Accounts.

3. Account: The atomic level.

The flexibility this provides with respect to household management has been immensely positive. Advisors can now organise client investments logically, move accounts across portfolios and households, generate reports at any level, and view data rolled up into an aggregate or zoom in to a particular account.

We also introduced the concept of views to Client Portfolios, meaning that all tables now feature customizable columns and the ability to create a multitude of unique views.

1.3.3 Custom Report Pages

In keeping with the theme of enhancing our core spine of advisor functionality, Reports was another feature we announced late last year that we have continued to improve on. We’ve added a host of new exhibits to reports, in addition to transforming several of the exhibits from static to customizable.

We also took this theme of customisation a step further by developing what we call ‘Custom Report Pages’.

Custom Report Pages allow users to create branded content within their reports and are designed to help tell the firm's story, explain the investment approach, and add context or commentary to proposals.

Download a Sample Report

This includes a new library of customizable pages that includes team bios, the ability to add charts with commentary, feature cards for outlining investment principles or service offerings, and editorial pages to meet every other need you might have when creating a report.

1.3.4 Short/Leverage Model Portfolios

If you are an advisor who employs leverage or short strategies, then model portfolios that support this are a mission-critical need. This was a common request from advisors, and considering established players like Ycharts don’t offer this, we filled that void.

Behind the scenes, this was no small feat. It required altering the fabric of our model portfolio infrastructure and a lot of testing, but on the front end, it works seamlessly. Users can now enter allocation weights as negative values and/or express >100% gross exposure at the portfolio level.

The UI dynamically adjusts depending on whether the model is long-only or contains shorts/leverage. Performance will factor in shorts and leverage, even to the extent that a model can become ‘bankrupt’ if the conditions allow for it. Reports will also dynamically adjust to support the correct analysis of short and leveraged exposures in models.

1.3.5 PDF Statement Upload

Imagine you are an advisor and you want to prepare a client proposal report for a prospective client later in the week. This report will compare the proposed model you are offering them vs. their current portfolio. They send you a PDF brokerage statement of their current holdings.

At this point, what are your options? You have to replicate the client’s portfolio in order to create this comparative report. This includes the lots, quantities, and cost basis, and a lot of other tedious manual work. While this manual workflow is available in Koyfin, it’s made a lot easier with PDF Statement Upload.

Simply drop the PDF statement into the reader, give the Portfolio a name, sit back, and wait. The portfolio should be processed and created within 24 hours, and Koyfin will keep you notified of the progress (in-app and via email) every step of the way. When it’s ready, you can spin up a proposal report in seconds.

1.3.6 Portfolio Capacity Upgrades

Quite simply, we bumped the maximum capacity (number of holdings) of our model portfolios to 1,000 holdings and the capacity of My Portfolio and Client Portfolios from 400 to 2,000 holdings.

Naturally, there have been a plethora of smaller enhancements made this year, but these are the ones I am particularly proud of.

Thanks for reading,

Conor

In short, this allows users to upload a time series of data (only requiring dates and units) and that data series will exist as a useable asset within their Koyfin account. The custom data series can be shared, added to Watchlists & Portfolios and can be pulled up in charts. The engine we use to create these custom series also afforded benefits to the rest of the system and other workflows, which I will discuss shortly.

What was the catalyst?

When you release a new feature or workflow, there follows a new timeline of improvements. The first version of a product is never the final one. For example, we launched Model Portfolios in mid-2023. We’ve since launched an abundance of improvements and new features related to that workflow that include but are not limited to, portfolio sleeves, reporting tools, richer exposure exhibits, a holdings matrix to identify exposure overlap, x-ray fund exposure, and more.

The idea for custom data series was born from the fact that Advisors, who use Models, often have clients who own assets that are not publicly traded. By virtue, they are not supported as instruments in Koyfin and can’t be included in their Models or reports. The initial idea was more of a ‘Custom Security’ engine that would allow Advisors to add non-supported or private assets to their portfolios such as corporate or government bonds, private equity, art, annuities, real estate, private funds, and more.

This would have been a great feature nonetheless, but with a few minor tweaks, it could be a great deal more flexible and appeal to an additional segment of consumers.

Custom Security ➝ Custom Data

The term ‘Custom Data Series’ differs in that it’s less definitive. In theory, this engine could be used to create anything, not just tradable securities. Things like company segment data or KPIs, economic data series, or just about anything you’d be interested in tracking on a time series. For example, below is a chart displaying Nielsen data on streaming as a % of US TV time. This data isn’t provided by Koyfin, but it exists publicly, for free, in a time series.

This data set now lives inside of my Koyfin account, I can update it easily and monitor it over time. Similarly, I now input all of the KPIs from the companies in my portfolio to Koyfin and update them quarterly. I can now also add the private equity and options that I own into my portfolio too.

The primary requirement of our first iteration of Custom Data Series was centred around allowing users to create these data sets, supporting them as assets which can be used around the platform, and most importantly enabling them to be added to portfolios. I am confident there are more great improvements we can make to this workflow in the years ahead, and I am equally confident we will gather many great suggestions from users, each of who have different ideas about how this feature can be used. Like other new keystone functions of Koyfin, the first version is never the final one.

Creation

The custom data series live in a user’s ‘My Data’ section, located in their account settings alongside other personal assets like Graph and FA templates or Watchlists. Here, they can be created, edited, managed, shared, deleted, and copied over to a watchlist.

Initial creation requires the series to have a Name, Ticker, Asset Allocation, Instrument Type, and be designated as a monetary (currency) or non-monetary (no currency) item. To upload data, a CSV with Dates and Values is required.

Each item of metadata will serve different purposes throughout the system. Name and Ticker are obvious, these will be used to identify and retrieve the security. These data series will then appear in ticker selection windows for Graphs, Watchlists, Portfolios, and Search.

Instrument Type and Asset Allocation are important as they will now surface in portfolio exposure exhibits. For instance, if you owned a portfolio of 50% SPY and 50% in a private corporate bond fund, the instrument type (corporate bond) and asset allocation (fixed income) would now surface in exposures.

Choosing whether or not a series has a currency is important when distinguishing between tradable assets with a monetary value (e.g. a private hedge fund asset) or a series with a numerical value (e.g. the number of Tesla vehicles sold in 2023).

Appending

Updating the data series can be handled in the My Data section, where users have the choice of uploading a CSV which will overwrite the previous data (+New) or maintain the existing data but add new data on top of it (Append).

Two Functions in One

The engine that we use to create these data series is now being used for other workflows too. Previously, Advisors who connected their custodians via our integration service would be able to pull in clients’ accounts and their positions but were limited by the fact that only securities we have in the system would be ingested.

For instance, if the client owned a Coca-Cola corporate bond, that security would be skipped from the ingestion. Thanks to custom data series, we now automatically ingest these securities, using a mapping from the custodian, and automatically assign their instrument type and asset allocation as well as loading their price, cost, quantity, and P/L data. This means that these securities will now provide a more accurate depiction of the client’s account and populate those allocations in exposure exhibits which then feed through to reporting tools.

I am excited about the custom data series because it feels like we have created something that serves our variety of user personas from Advisors to researchers to data fanatics. I look forward to seeing how users adopt the feature and how we develop this in the coming years.

This functionality is now live for Pro and Advisor Pro users who can create unlimited custom data series. There is also a limit of five custom data series in the Plus plan. If you’d like to test it out (or watch a demo), feel free to reach out to me with your account email and I can set up a trial for you. Equally, if this is the feature you were waiting for the make the jump to Pro, here is a 20% discount code also.

Thanks for reading,

Conor

We are also on the cusp of releasing something huge (with a capital Y). I’d say it’s firmly in my top two for features i’ve worked on in my two years at Koyfin. I’ll write more about that when the time comes, but for now, I wanted to catch up since the last Product Work update I shared back in August.

Below is a variety of the things we have released at Koyfin over the last three months, with some commentary.

1. Advisors

A central theme at Koyfin over the past 18 months has been Advisor workflows. We have incredible word of mouth at Koyfin and are blessed with an abundance of different user personas from the arms of an umbrella of two larger buckets I’ll just refer to as individual researcher types and professionals/teams. Within the professional bucket, we had a strong following among Financial Advisors despite lacking some of the core workflows they needed. We’ve undergone a journey to incorporate more of those workflows into the platform.

Two years ago, Koyfin lacked any portfolio functionality. Today, we have a library of them from My Portfolio for individuals to aggregate their brokerage accounts and track P/L and exposures, Model Portfolios to create custom baskets and proposals, and Client Portfolios to help bring the proposals into the real world with more emphasis on P/L and lot management. On top of that foundation, we have expanded into custodial integration and report generation, and that’s just getting started.

Creating a platform that an individual like me can use at the same time as an Advisor with more nuanced and advanced needs doesn’t come without its dilemmas. We launched an Advisor Pro tier this September to create a clearer distinction between functionality and guidance towards the right plan for a user.

1.1 Client Portfolios

The Model Portfolio functionality began development ~18 months ago and is a core advisor workflow. Advisors use models for all manner of reasons. As far as acquiring clients go, the model acts as the proposed portfolio. These portfolios are based on allocations, and dynamic weightings, and are flexible enough to be used for various purposes. It’s a tool for comparing the client’s current portfolio and fee structure vs. what the Advisor can offer as an alternative.

When the potential client becomes a client, the Client Portfolios function switches focus from simulation to reality. These portfolios are based on the actual structure of the client’s real-world portfolio, with a focus on quantities, lot management, purchase dates, P/L, and total return.

While the UI appears similar to My Portfolio, a feature built for the individual, the Client Portfolios hub allows for multiple portfolios to be created as opposed to a solitary portfolio. Advisors can create, upload, or import, hundreds or thousands of their client portfolios into a central hub and are equipped with the portfolio analytics they need.

Client Portfolios also introduce the concept of a workspace within Koyfin. For teams of Advisors, the Client Portfolios hub can be shared across a multitude of users at once, making portfolio and client management more collaborative.

1.2 Custodial Integration

We introduced our first integration by allowing Advisors to connect to the Schwab Advisor Center to import their client accounts directly from Schwab.

We are early in the integration journey, with greater functionality and a broader range of integrations on our roadmap, but it’s an exciting path to be on which removes friction for Advisors who want to use Koyfin as their primary analytics platform.

1.3 Reports

Report generation tools are a crucial workflow in the acquisition of potential clients and for internal purposes. From our feedback sessions, we understood that Advisors were frustrated at the lack of flexibility in most offerings. They wanted discretion in what gets included in the report and greater customizability.

We took this on board as we developed our first iteration of Koyfin reports, which allow users to generate reports on funds, models, and client portfolios that cover allocations, performance, risk, and portfolio breakdowns. These reports can be used on a single-entity or comparative basis and allow users to add logos, brand colours, and custom text pages for things like investing policy statements or disclosures. The order and inclusion of individual exhibits can be adjusted too.

The initial feedback has been great, and we are already working on the next version of reports to upgrade the functionality further.

1.4 Portfolio Sleeves

If you are not familiar with the concept of Portfolio Sleeves, I suggest watching Christopher Nolan’s Inception; it’s a portfolio within a portfolio.

Sleeves let users combine multiple models into one larger, more complex strategy, helping users build custom asset allocation models by pulling together different components (models), making it easier to tailor investment strategies to their clients’ needs. The underlying holdings of the sleeves will be surfaced in all exposure and report exhibits too.

2. Other Great Stuff

Besides building out Advisor workflows, we have been busy adding functionality that benefits every user. Below are a handful of some of my favourites from the past three months.

2.1 Comparison Graph

At Koyfin we have several core charting functions. The Historical Graph (G) is the Swiss army knife. It does everything and is the most customizable, powerful, and versatile chart tool we have. To make certain charting flows faster and more efficient, we created the Intraday (GIP) and Performance (GM) graphs which are designed to do one job. Intraday proves intraday prices and % changes up to a range of 10 days. The Performance graph provides the total return and CAGR for any security added to the chart. Both are simplified and only contain a menu to add securities instead of a full library of metrics. It gets you from A to B much faster than doing it on a Historical Graph.

The Comparison Graph (GC) is designed as another one of those simplified charting workflows. This chart is designed specifically for comparing metrics across companies and includes features like colour-syncing, auto-syncing of panels, and a different UI for ticker and metric selection. While more efficient than the Historical Graph for this particular workflow, it’s not as flexible concerning merging panels, hiding individual metrics from select tickers, and some of the formatting options for chart styles.

2.2 Quick Actions in Watchlist

Quick Actions are a way to overlay functions on top of a watchlist whereby the watchlist acts as the ‘universe’ on which the function is based. For instance, a customizable feed for filings, news, press releases, and transcripts.

We also added a scatter plot to the Quick Actions menu. In the near term, we have a lot of other functions we could surface here, such as the earnings calendar and the Lots of Charts dashboard. While it’s possible today to use a Watchlist as the universe for these functions, the Quick Actions menu makes this possible without navigating away from the Watchlist section. It also opens the door to creating new, unique, functions such as histograms, bar charts, and more.

2.3 Koyfin Assistant

In beta at the time of writing, we released our first consumer-facing artificial intelligence feature this month, the Koyfin Assistant.

The goal of the Assistant was to assist in support queries rather than being a fully-fledged co-pilot. The knowledge base will help users navigate the platform, inform them how to use features, and provide context on data coverage, functionality, and more. It’s not there to provide stock tips, and it doesn’t currently have access to our data (e.g., asking what Apple’s revenue was in FY22 compared to FY21).

This will be particularly useful for the thousands of new users we onboard to Koyfin each week, most of whom have simple questions the Assistant can handle. While it may appear a small step, it excites me when I consider the other use cases we can develop to improve the platform and reduce the learning hurdle.

2.4 Two-Factor Authentication (2FA)

Following the launch of custodial integrations, and with plans to integrate with more service providers, 2FA was necessary to maintain the level of security throughout the chain. This helps keep you or your client’s sensitive data secure.

Users can now enable 2FA from the profile settings with an authenticator app.

2.5 Fund Percentiles

Koyfin’s percentile ranks are one of the unique datasets we have on the platform. Previously reserved for equities, we created a set of percentiles for US and Canadian ETFs and Mutual Funds.

Users can now see percentiles across performance, characteristics, and quantitative categories. For US ETFs, there are valuation multiples and yield percentiles too. The data is available as a column selection in tables, the screener and the visualized snapshot format.

2.6 3Y and 5Y Equity Percecentiles

In addition to adding fund percentiles, we bolstered our equity percentile offering, adding 3Y and 5Y variants to the existing 10Y and 20Y periods.

This allows for a more holistic, well-rounded, view of an equity’s percentile scores and provides context for both long-term and short-term trends.

2.7 Linear Regression and R2

Linear regression is a statistical method that models the relationship between two variables by fitting a straight line (regression line) through the data points in a scatter chart. It helps to identify trends and predict outcomes while the R² (coefficient of determination) identifies the proportion of variance. In other words, how strong the relationship is.

We added a simple linear regression and R² annotation to our scatter charts which are used throughout the platform on the back of user requests. One improvement we can make here is allowing users to omit outliers from the chart, thus removing them from the equation in tandem.

2.8 Benchmarked Exposures in MYP and CP

The development of the three portfolio features began at different periods. They cater to different needs, so shouldn’t be identical in substance or style, but there were a few gaps in feature parity. Namely, the exposure exhibits of the portfolio.

In My Portfolio (MYP) and Client Portfolios (CP) we introduced richer exposure exhibits (that Model Portfolios (MP) already had) that came with several great upgrades:

• X-ray exposure for funds: We now look underneath ETFs and Mutual Funds to provide exposure breakdowns of the underlying holdings. For instance, a portfolio comprised of 100% S&P 500 ETF represents the exposure of those 500 or so companies.

• Benchmarked exposure: Users can toggle between showing exposures at the portfolio level (via pie charts) or vs. a chosen benchmark (via bar charts).

• 11 new exhibits: From 5 to 16 unique exhibits, we broadened the type of exposure analysis users can conduct on their portfolios.

• Interactive holdings table: Previous exposure pages were aggregated to the top level. For instance, in a sectoral exposure table, we would show you the weight (%) of each sector in the table. Now we break down the underlying holdings of each category and allow users to filter by subcategory (e.g., by consumer discretionary)

2.9 Crypto Support in Portfolios

Plain and simple, we added support for cryptocurrencies in portfolio features.

2.10 Enhanced Currency Support in Portfolios

Across the portfolio features, the portfolio currency is important because all positions, P/L, and returns are converted back to that currency. We increased the number of supported currencies from 11 ➝ 90 to ensure that everyone can make use of the features.

While the bulk of our users reside in either the United States, Canada, or Europe, we have users from across the world and this would have been a welcome update.

I hope to be back discussing another huge product update shortly but until then.

Thanks for reading,

Conor

1) Relative Valuation Charts

Relative charting has existed on Koyfin for years. Relative Ratio (A/B) is the ratio of two prices and is rebalanced daily. Relative Performance (A% - B%) is the performance spread of two assets and assumes an investor is long one dollar of A vs short one dollar of B at the start of the chart period with no rebalancing. Relative Spread (A - B) is simply the difference between A and B and is particularly useful for yields.

A workflow that has been missing from this family of relative data series is relative valuation. Relative Valuation (Multiple A / Multiple B) is like the relative ratio (A/B) but for valuation multiples. It allows an investor to quickly establish where a security’s valuation stands in relation to another asset, sector, industry, or another grouping of securities.

Suppose you wanted to see how Lululemon’s forward PE compares to the Consumer Discretionary sector. The relative PE series shows that Lululemon has historically traded at a premium to the sector, until mid-2024 (indicated by the fact the ratio dropped below 1.0).

Both Lululemon and the sectoral ETF have seen multiple compression over the last 12 months, but the magnitude of decline at Lululemon has been larger than the sector as a whole. Notice how the PE of XLY (consumer discretionary ETF, blue line) overlaps Lululemon’s PE (light purple line) in 2024.

The chart below shows Apple’s PE relative to the S&P 500 (blue line) in the upper panel and the PE of Apple (black line) and the S&P 500 (orange line) in the bottom panel. The thick, vertical, black line in 2019 indicates the period when Apple’s PE began to exceed that of the S&P. Above, you will note that the relative ratio surpasses 1.0 at this period too.

This data series is created differently from other relative series; which exist in the data series picker as individual items. Relative valuation exists in the indicator settings panel as a transformation option (similar to how we designed the Multiplier data series).

Because this was created as a transformation series, it means that the ability to apply statistical bands (means, standard deviations, lows, highs, etc) is included too.

This transformation is now present under all of Koyfin’s valuation data series and is available across the global library of equities, as well as US ETFs.

2) Fibonacci Customisation

Fibonacci annotations in Koyfin have existed in some form for years. However, it lacked the level of functionality that would satisfy the needs of an investor who uses this workflow. No extension lines and no level customisation. Just standard retracements. It was a bit like offering someone a car that drives, but only in first gear. It can get you to where you need to go, but serious drivers need more control and speed to enjoy the experience. With all that said, I dived into our feature request backlog and aggregated the feedback specific to Fibonacci.

Parsing through the requests revealed that folks wanted; (1) a retracement trend line, (2) extension lines, (3) the ability to show/hide levels, (4) the ability to customise levels via user inputs and, (5) some colour formatting.

The initial wireframes sought to bolt most of this additional functionality onto the pre-existing annotation settings window (from line type to delete). The new ‘values’ window would include checkboxes to show and hide levels, numerical boxes for customising levels, a colour picker for individual lines, and a reset button in case things needed to be turned back to factory settings.

What we ended up with was a lot more elegant than my wireframes, but captured the spirit of what we wanted to achieve, and then some. While there are still more improvements we have yet to make to Fibonacci, I believe this update made the annotation truly usable for a point-in-time analysis.

3) CAGR for % Change Annotations

Small design or function improvements can feel like thankless tasks. They are too small to warrant mass communication and generally get pushed to production without so much as a whisper; sneaking in and waiting patiently to be discovered. We ship so many of these it would be redundant to cover them all, but one that I enjoyed recently is the addition of a CAGR and period to our % Change annotation.

This one is particularly useful when dealing with fundamental data series, like in the above example where I plot Alphabet’s 20Y free cash flow CAGR. In many instances, these CAGRs exist as separate data series, but the ease of just attaching an annotation instead is nice too.

Thanks for reading,

Conor

A yield curve graph is a way to visualise the relationship between the interest rates (or yields) of bonds with different maturities (tenors). They are typically used for government bonds of maturities between 1 month and 30 years. We also added a 0-month tenor, which is represented by the current central bank policy rate of the respective country. The tenors are commonly displayed along the x-axis while the y-axis is used for the yields.

It’s an alternative perspective to viewing government bond yields on a time series. To illustrate, in the chart below, I have plotted a time series of yields for US government bond tenors between 1 month and 30 years. As you see, this can be visually cumbersome to study. We see that yields are going up, but seeing where one tenor trades in comparison to other tenors is difficult.

Now imagine selecting a date along that x-axis and striking a vertical line straight through the chart. That line, for that specific date, will meet a single data point for each government bond tenor. The Yield Curve essentially transposes this data and presents it across an x-axis. It can be used to compare the yield curves of countries, or the historic yield curves of a single country through different environments.

There are several ways to interpret a yield curve. The three most common are those which are ‘normal’, ‘inverted’, and ‘flat’. The normal yield curve is upward-sloping, indicating that longer-term bonds have higher yields than shorter-term bonds. The inverted curve is the opposite, a downward-sloping curve where shorter-term bonds exhibit the highest yields.

While not always the case, the rule of thumb is that inverted yield curves evoke caution from investors expecting a recessionary environment. Meanwhile, normal curves indicate the expectation of economic growth and rising interest rates. Flat curves, which look as they sound, are commonly seen during transitionary or uncertain periods.

There are other ways to illustrate these relationships. Below is a chart showing the United States 10Y2Y curve.

The inversion of this particular curve has had a 100% hit rate for forecasting recessions. This success rate did come into question during the inversion in the summer of 2019, however. Preceding the pandemic by ~8 months, some argue that the recession which followed, as ephemeral as it may have been, was an anomaly. That there was no way a yield curve could have forecast a pandemic; that without the pandemic, we have no way of knowing if a recession would have occurred. That’s just the reality of it, we will never know, and I imagine the 2019 inversion will forever be marked with an asterisk. As for the current inversion, which has been ongoing since the first quarter of 2022 with no recession yet to be unearthed, only time will tell.

I digress. The Yield Curve is a new way of looking at the data we already presented inside of Koyfin. While it sounds simple on the surface, this required building a new page to support this workflow. The chart on which the Yield Curve is positioned looks similar to existing charts in the platform, but besides a similar UI, it’s entirely different.

The make-up of the page itself was another factor to consider. At first, it looked like it should be a new addition to our Market Dashboards section. However, the required UI for such a tool was not consistent with the rest of the dashboards. While it shares the symbiotic relationship between tables and charts, traditional market dashboard tables house fixed data. In other words, the user has a predefined selection of columns. The data, selected by a checkbox in the table, would then appear on the charts. The Yield Curve required date selection, country selection, as well as the ability to duplicate each of these selections at will. With so many possible variations, it wouldn’t make sense to build this inside of a static table.

The workflow was more compatible with the Analytics tools we provide, and that is where we put it. The table, which dictates the data presented on the chart, allows a user to select a maximum of ten yield curves at one time, across a combination of countries and dates. It also presents the yield curve in tabular format for the date chosen.

Users can alter the colour, thickness, and style of the chart lines, as well as the rest of the usual alterations. We introduced a show/hide data label for the first time, which accents the data points with labels in one click. This is something I’d like to implement more across the platform. The ability to download the data, as well as share customised PNGs of the chart, is also available.

It just went live in production this afternoon and should prove to be yet another workflow Koyfin users can now perform in-house.

Thanks for reading,

Conor

Every time we ideate a new feature, a lot of thought and comprehension go into that. To ensure that we understand what we are building and that we communicate its value and purpose to engineers and, ultimately, users. A byproduct of this process is that I constantly learn new subject matter and strengthen my understanding of things I knew a little about already.

I am going to start occasionally documenting some of the work I do in that part of my life (like this post in April) and share it in a new section called Product Work. This way, you can unsubscribe from that section if you are not interested in that type of content. Without further ado, here are a few things I worked on that we shipped in the last week or so as we looked through our backlog of feature requests for charts.

1) Peter Lynch Charts

In several of Peter Lynch's old books, he shared a charting technique later dubbed the Peter Lynch Chart. "A quick way to tell if a stock is overpriced is to compare the price line to the earnings line", he commented in One Up on Wall Street. Lynch was watching to see when the share price made contact with that earnings line.

“If you bought familiar growth companies – such as Shoney’s, The Limited, or Marriott – when the stock price fell well below the earnings line, and sold them when the stock price rose dramatically above it, the chances are you’d do pretty well”.

Another variation of this workflow is something that investors refer to as Fair Value Bands. Suppose you believe that when JP Morgan trades at 2x total book value, it’s fair value. You want to plot that price level on a chart, like below.

We recreated this workflow by allowing users to multiply or divide data series. In doing so, someone can take the underlying data series in a multiple, such as revenue per share or consensus average NTM EPS and create constant multiples. The result is a time series which replicates the share price as though it were pinned at a constant multiple. Suppose Nike trades at 23x earnings and you want to see how that compares to historic averages.

You could open up a chart and display the PE in addition to some statistical bands such as mean to show the average over a period. Alternatively, you can now plot a sequence of constant multiples to see how often Nike trades in that value range. To highlight another example, Apple currently trades at ~9.1x trailing revenue and has averaged 4.3x over the last two decades and 4.9x over the last decade. The S&P 500 currently trades at ~3.1x revenue. Here’s what that would look like with each of those four bands plotted relative to Apple’s share price over time.

As illustrated, Apple’s price to sales has reached unprecedented levels in recent years. I am confident people will find other ingenious ways to use this feature that we didn’t consider. Whatever the nomenclature, this feature was the case of us taking some initiative to build something we thought would create a new and unique workflow for users. The number of explicit requests for this functionality was lower, relative to some of the other things I’ll discuss today. But sometimes you have to build things that are just plain awesome.

2) Shareable Screeners

At Koyfin, we’ve been thinking a lot about collaborative work and shareable assets. There are many tangents in this lattice of potential use cases. For one, how individuals share vs how teams share can be distinctly different. Sometimes the sharing flow depends on the asset. When sharing a chart template, it’s most often the case that you want to share a copy of that chart so that the recipient can have one of their own. In essence, every recipient gets their own replica. It’s like a scanned document.

In a Watchlist this duplication is desired. But there is also a demand for collaborative work inside of the same asset. For instance, Dave invites Jenny to Watchlist 1, where they build tables and views and can see each other’s changes in real-time. In another instance, the owner of the Watchlist may wish to share the live changes of the Watchlist but not grant viewers edit rights.

With screeners, we opted to go down the path most similar to the chart template. The core value in sharing a screener is the inputs; the selection of criteria the user has used to generate the results. Users can now enable public access to their screener which allows a recipient to (a) view the screener output with read-only rights and (b) save a copy of the screener and its inputs to their terminal if they wish to edit it.

The cool thing about this sharing flow, enabled by a URL, is that the screener can be viewed by anyone; regardless of whether or not they have a Koyfin account. If you’d like to check it out, check out this screener I created which scans for profitable companies trading at historically low valuations with moderately high margins relative to their history.

3) Volume by Price Profile

I enjoy assembling fundamental and economic charts, but I have never been much of a technical analyst. Nonetheless, I enjoy opportunities to learn more about these kinds of indicators and data series, and we had a backlog of feature requests relating to some essential technical indicators that we wanted to make disappear.

The Volume by Price profile is an overlay, usually pinned to a y-axis, that displays the total volume traded at a range of equally distributed price ranges over a specific period. It was a fun challenge, and after ensuring I could reproduce a few in Excel, the underlying maths and logic were easy enough to grasp.

Price intervals on the y-axis will be updated according to the time period selected on the chart, as well as the user’s preference for the number of periods (bars) they wish to show. The Point of Control (POC) line, which indicates the period with the highest traded volume, can be toggled on/off. The calculation will adjust when the user toggles between displaying adjusted vs unadjusted prices too.

4) Anchored VWAP

A technical indicator I had never heard of before until this year, but one that had a considerable number of upvotes on our feature request board. A VWAP (Volume Weighted Average Price) indicator calculates the average price of a security based on volume and price within a specified time frame. The regular VWAP typically begins at the start of the specified period. The anchored VWAP is just a VWAP where you can specify the start date (the anchor). They are used by technicians to identify trends, support and resistance, entry and exit points, sentiment, and so forth.

The math for this one was, again, relatively simple. However, it was an interesting feature because of the UX of applying the VWAP to a chart. There are a few components you’d want to specify; the anchor date and calculation method. While it’s not an annotation tool, we figured it would be nice to allow a user to click on the candle to specify a start date, but at the same time change the date if they wanted to. We settled on a hybrid of solutions. First, we enabled AVWAPs to be added using a right-click on the chart screen when hovering over a candle. This is the first time we are experimenting with this type of flow.

Doing so will populate the AVWAP as a data series in the left side panel under that security, where you can change the anchor date, prices used in the calculation, and chart stylings.

Alternatively, you can bypass the right-click and simply add VWAPs to the chart by using the side panel data series picker.

5) Historical Price Transformations

This update, which allows a user to transform the historic price of a security, was mostly a matter of UX and improving the visibility of functionality which already exists. Previously, a user would have to search for a secondary data series to display performance, drawdowns, YoY % and QoQ % change, YoY and QoQ change and cumulative change.

Possible, but perhaps not the most intuitive. For example, to display the drawdown history of Nike, a user would be required to first add Nike to the chart, and then search for a second data series called “Drawdown from peak”. Now, the user can transform the original Nike price chart there and then. This also enables users to add statistical bands to these transformations to show means, medians, highs, lows, and percentiles over a time series.

This is particularly useful for macroeconomic data series which are presented in notional terms. The Federal Reserve Balance sheet, for example. To quickly illustrate how far it has fallen from its peak now takes a couple of clicks. One of Koyfin’s advantages is that we pack an incredible amount of functionality into a modern UI. However, this also means that it can be difficult to realise the full extent of capabilities at your fingertips. With subtle improvements like these, we hope to alleviate some of that.

Thanks for reading,

Conor

“I have multiple brokerage accounts, and no place to aggregate them”.

Account aggregation is a common dilemma for investors; our portfolio is often a collection of multiple accounts. You might have a retirement account, a taxable one, a tax-advantaged one, etc. A few months ago I was tasked with leading a project to tackle this dilemma at Koyfin. I was excited because it resonated with me. I too, have this problem. I now had the reigns to ideate something. Something I’d be excited about as a user and investor. When I started to scope this project it confirmed my hunch that this was a common problem.

Regardless of sophistication, investors still rely on tools that are either not built for purpose, unnecessarily expensive, or antiquated. Most people I spoke to rely on Microsoft Excel to track a portfolio. Don’t get me wrong, Excel is fantastic. But it’s not a portfolio tracking tool. Koyfin has a lot of ambitions in this space, but decided to tackle the most fundamental issue first; “I have multiple accounts and no place to accurately aggregate and monitor them”.

This month, we will release this effort's first version (v1). I wanted to walk through this process, what v1 looks like, some of the issues we faced, why it’s differentiated, and some of the small “delight” features we worked hard on to create utility for investors. All credit goes to the fantastic roster of engineering talent we have at Koyfin and the folks who tirelessly tested the feature, the product team, and the design team. Once we ideated, planned, and wrote up this project, they were the ones who brought it to life and did all the heavy lifting.

If you’d like to get access to the early release of this project before it publicly announced, we’d love your feedback. If you are interested in doing so, reply to this email, shoot me a DM on Twitter or DM me directly on Substack.

If you don’t have a Koyfin account, you can create one here.

Tackling the biggest issue first

What v1 is and future functionality

The first version of My Portfolio is the beginning of a multi-stage development process. At the beginning of this project, we had a lot of ideas and had to focus on solving the most fundamental requirement first, and well. I will opine on what future developments might include also. The fundamental issue we sought to address in v1 was giving investors a tool to track and aggregate their existing portfolio, break down profit & loss, and understand their exposures.

What v1 is: My Portfolio v1 will allow you to track current holdings. These holdings can be aggregated across multiple accounts and lots. We will show you the breakdown of your P/L by lot and parent, converted into a single portfolio currency and include the impact of FX rate changes over time. Investors will be able to analyse the P/L of their portfolio, view exposure exhibits, and create views to fundamentally analyse their positions.

Future functionality: The concept of historic transactions (sells) and a time series of performance will come later. While this was part of our planning, we wanted to ensure we limited the scope of v1. Both of these projects would require considerable time to implement, but we plan to include performance and historic positions in future versions. My Portfolio is not something which plugs into existing brokerages via API; positions are updated manually or via CSV.

Starting BIG and narrowing the focus

Prioritising and the key features of v1

Most projects start similarly. You have a grand vision of what you want to build, where a product will stand a year from today, and what future versions should look like. The ideation phase often starts from the finishing line and works backwards. Once you have a picture of the end state, you then have to narrow that focus down to something achievable. When I say achievable, I don’t mean easy. It’s the difference between releasing something within a 3-month window and having the opportunity to put it in users’ hands, get feedback, and iterate vs. spending twice as long building something twice as technical, but without a user ever touching it. Long before I ever started working in a product role, a mentor gave me a scrap of advice that I often remember. You can ideate and plan but you have no idea what users will do with a product once they get their hands on it. Sometimes, the best ideas come from observing how users interact with a product in ways you hadn’t originally imagined.

When Koyfin launched the Model Portfolio feature late last year, the same attitude was applied. Just 3-4 months after launch, the feature has come so far that it’s unrecognisable compared to the version we initially launched. This was aided by listening to what users wanted and building it. Both as a user and observer, I have already witnessed interesting ways Koyfin users have engaged with Model Portfolios. Some have used it as a back-testing engine. Others have created custom benchmarks and ETFs using the tool; each model acts like a security within the terminal which facilitates this use case. I even realised that you can use a model portfolio to proxy the USD returns of GBP-denominated security to compare against other US assets. Model Portfolios wasn’t explicitly designed for any of these use cases, but the way users have adopted it provides us with valuable insight into (a) what additional tools/features investors want and (b) how we can modify the existing product.

Our focus for v1 in My Portfolio was delivering a great current portfolio tracker. Once that foundation has been set, we move on to the next leg.

Four key elements

Creation, Profit / Loss, Exposures, Analysis

Please note that screenshots are taken from a beta version of My Portfolios and don’t fully reflect the final product.

My Portfolios introduces a lot of new functionality to Koyfin. Things like FX adjusted returns, P/L, and lot entry which were brand-new features created specifically for this feature (which I will comment on shortly). There are plenty of doo-dads and whistles to comment on, but for now, I’ll cover the four key elements of My Portfolio; beginning with the creation flow.

1. Creation & Ongoing Management

Because of this feature’s novelty, we had to design a new creation flow. This flow had to incorporate manual lot entry, multiple accounts, separate cash balances for accounts, an overarching portfolio currency, and the ability to upload accounts via CSV.

We settled on a sleek dialogue box that facilitates all of these needs. At long last, Koyfin users can now enter individual lots from their portfolio, reflect multiple cash balances, and edit the positions within individual accounts. This can be done manually, or by uploading a CSV to each account with information on the ticker, quantity, cost per share, and purchase date.

For Koyfin users who currently use Watchlist to track their positions, you can download that Watchlist as a CSV in the platform and use this as your CSV upload.

Once completed and saved, your data will be aggregated into a unified portfolio and the feature will open up.

2. Profit / Loss

The vision for the Profit / Loss tab was inspired by brokerages; who are often the source of truth for P/L data. While brokerages generally offer the greatest level of detail about our P/L, they are seldom the place where investors conduct fundamental analysis because they are ill-equipped to handle it. We realised that investors are used to performing these activities as two separate workflows.

Therefore, the Profit / Loss tab is where we present lot-level data, context on P/L, account aggregation and grouping, as well as security-level and performance-related data.

Here, users can build a table to monitor their P/L as a collective portfolio or group by account, sector, industry or asset class. When grouping by account, the use of summary rows (like “sum”) will allow you to identify the weighting, market value, and P/L of separate accounts relative to your aggregated portfolio.

3. Exposure

The exposure page will break down your portfolio in pie and table form across securities, sectors, industries, asset classes, and countries. We created a custom, interactive, pie chart for this feature.

The v1 of the exposure handles the basics well. Some quick follow-up improvements to this feature may include the ability to show account-level exposures, exposure by P/L, currency exposure, and I am sure we will get some great suggestions for users too. Those familiar with the Model Portfolios feature will know how powerful that feature’s exposure exhibits are.

As the development of My Portfolio progresses, I am hoping to borrow functionality from the Model Portfolio exposure exhibits. Here, we present data in bar chart form, against a benchmark (the concept of a benchmark doesn’t exist in v1 of My Portfolio). We also present more choices for exposures including fixed-income, credit quality, and maturity exposure. I am also excited at the prospect of adopting some of the logic from Model Portfolios. In the image above, for example, clicking on a sector will populate the table to the right of the chart, breaking down the constituents of that sector in the portfolio, their weight, exposure to that sector (for funds), and their portfolio contribution. Another aspect I like in Model Portfolios is that we have x-ray fund logic. In simple terms, you could have a 100% SPY portfolio and Koyfin breaks out the exposures as per the constituents of that fund. The holdings matrix tab will also break down the constituents of that fund and show overlapping exposures between directly held positions and other funds.

I digress. The point is, that there is a lot more we plan to introduce with exposure.

4. Analysis

The analysis section will be the most familiar to regular Koyfin users. It has a likeness to the Watchlist, with the concept of views, currency presentation, grouping, summary rows, and the full selection of columns. This table will be pre-populated with your portfolio companies, where you can use summary rows like “portfolio weighted average” to show the weighted average multiples or margins of your portfolio.

Relative to the Profit / Loss column selection, Analysis is considerably richer (having the full spectrum of columns available) and is the place where investors will conduct more fundamental and valuation-centric analysis. Analysis also allows users to utilise a library of different views, which they can cycle between seamlessly. For users who have spent time creating views inside of Watchlists, fear not. Because views from Watchlists can be imported to the Analysis table.

Next, I want to highlight some of the “little delights” of this project. Many of these are new features to Koyfin, and while seemingly simple required a lot of thought to implement successfully.

Little delights

Features that I want to praise

Before having a dedicated portfolio tracking tool, Koyfin users would commonly perform this workflow in a Watchlist; a great feature but one that wasn’t purpose-built for this activity. The Watchlist, with its variety of portfolio tools, would allow a user to populate aggregate-level positions, add average cost, and purchase date, and see returns and P/L. There were some flaws, however.

To emphasise the value of adopting My Portfolio in place of Watchlists, we had to introduce a variety of new features to Koyfin.

1. Lot Level Transactions & Duplicate Tickers in Tables

Investors have asked about lots for some time. Watchlists are not built to house duplicates of the same ticker inside of a single table. Suppose you owned Starbucks in two separate accounts or that you owned 40 shares of Starbucks that were purchased in 10-share increments. Expressing either of these situations was not possible in a Watchlist and users would have to populate columns with aggregate-level portfolio data. For example, showing the 40 shares of Starbucks that they own in one line item, with a single average cost.

Now users can populate each lot and see both the aggregate and lot-level data, including P/L. They can also populate lots across multiple accounts; meaning that if you own the same asset across multiple accounts we aggregate this for you under a single parent-level position (see NVDA below).

Alternatively, if you want to group your portfolio by account, you can now see the same ticker appear multiple times in the same table. E.g, you can see Nvidia in the ISA, Retirement, and Taxable account.

2. Currency Conversion and the Concept of Portfolio Currency

A niggling issue in the Watchlist was that the market values of a position in a table could all be converted into a single security. Meaning that if I wanted to convert all of my US investments into GBP, I could. The problem, however, was that all profit/loss items would be shown in the security’s trading currency. While the conversion of market value was useful, the unreconciled nature of the P/L series meant that summary rows were not.

With My Portfolio, we introduced the concept of a portfolio currency. This means that any securities you hold in the portfolio will automatically be converted into the portfolio currency. You can add the cost basis in USD, for example, and while we present the cost basis in USD, we convert everything else into GBP (or whatever currency your portfolio is denominated in). This includes P/L.

3. FX-adjusted P/L

Showing an investor their P/L converted to portfolio currency at current FX rates wasn’t going to be enough. If I, as a UK-based investor, purchased META (USD stock) 5 years ago, I likely made a nice profit on that trade. Over those 5 years, however, the fluctuation between the FX rate of GBP and USD will have impacted my returns too. We know that a lot of Koyfin users invest outside of their domestic market and want to account for this.

That’s why we introduced three new data series to Koyfin relating to P/L which reflect these FX impacts. Each of these three data series is reflected as both notional and percentage values.

• P/L (excl. FX): The notional profit or loss from an investment, converted to the portfolio currency at the most recent FX rate.

  • You can think of this as solely the “security-level” P/L.

• P/L (from FX): The notional gain or loss from the change in currency conversion since the position was opened.

  • This reflects the impact of currency movements throughout your holding period.

• P/L: The notional profit or loss from an investment, including the P/L from FX. I.e. after the impact of currency conversion fluctuation.

  • You can think of this as the “true” P/L. The amount of cash you would receive (excluding costs) for selling the position in your domestic currency.

When the security trades in the same currency as the portfolio, P/L (excl. FX) and P/L will be the same, while P/L from FX will read zero. To produce these data outputs, add a purchase date for each position so we know the FX rate on the purchase date.

4. Total Return

At the position level, a Koyfin Watchlist can show you the price return of your investment since purchase. Showing the total return was a commonly requested feature from users; which factors in the influence of dividends on the return profile. My Portfolio introduces this and calculates total return which includes the impact of FX changes over time.

Koyfin shows users the price and total return of securities throughout the platform and in charts at the security level. However, the infrastructure of a watchlist is slightly different and wasn’t built to be able to retrieve specific dates in history; required to access the adjusted prices necessary for the total return calculation. Different databases, different logic. We overcame that challenge with My Portfolio and now show both price and total return. This adds more granularity to the return analysis an investor can conduct on their investments.

In the below example see Greggs, a UK company. The portfolio is GBP-based, so the FX impact is null; hence why P/L (excl. FX) and P/L are identical. The company pays dividends, however, and this is now reflected, as per the purchase date, in the Total Return field.

Something to note, however, is that total return assumes the reinvestment of every dividend, while price return ignores them completely. If the company has never paid a dividend, the price and total return will be identical. If the company has paid dividends, the total return should be greater. As an investor, it’s unlikely your true return % reflects either of these metrics perfectly. You might not reinvest every dividend into that same company, so your true return is likely somewhere between price and total return if the company pays dividends.

Nonetheless, having both data series at your disposal provides you with more context and understanding.

5. Quote box

The quote box at the top of the screen displays the market value of the portfolio, the 1D change, number of holdings, cash, cash weighting, and P/L of the portfolio and will tick away during the day. In the future, we plan to add portfolio valuation data here too. The quote box also contains portfolio news and a button to edit holdings.

6. Portfolio News

One of my favourite features in the Koyfin Watchlist is Watchlist News which, at the click of a button, generates a feed populated with news, filings, transcripts, and press releases tailored to the basket of securities inside of that watchlist. You can use the settings to filter various types of news and filings too. I remove all news and ownership filings from mine, due to personal preferences.

This feature was particularly useful in the context of portfolio monitoring, so naturally, we migrated this over to My Portfolio too.

7. Import Watchlist Views

Koyfin users spend hours creating unique views in their watchlists. Asking them to recreate those views in the Analysis section would create friction.

We like to eliminate friction where we can, so we ensured that users could import the views they have created directly into the Analysis section.

Looking ahead

This project tested us in many ways. It’s easy to conceptualise what we want from a product perspective. But while a product like Koyfin is dense with data, it doesn’t act and behave like an Excel spreadsheet. For instance, understanding the formula and logic to calculate FX-adjusted P/L and returns is easy. Historically, tables inside Koyfin have not been built to retrieve data in such a way. This was something we had to solve for. Showing the same security in a table multiple times as well as the concept of aggregating lots; we had to build an entirely new table to handle this. Believe it or not, the idea of having cash as a standalone asset which could then interact with other securities and influence the value of a portfolio was new and required a solution too.

What excites me is how these learnings and new features may eventually be adopted across the platform more broadly. We have a lot more work to do with My Portfolio. First, by enhancing what we have already built. I have no doubt the user feedback we get early on will fuel this. Second, introducing new functionality like performance, historic transactions, the concept of having multiple portfolios, and potentially custodial integration. I feel immensely proud of what we built in this first version, and I can’t wait to get it into the hands of users. To see an idea that was once a collection of documents and some ugly wireframes transform into what we have built, is a magical feeling.

I hope you will take me up on the offer and check out the early release, your feedback will be warmly welcomed.

Thanks for reading,

Conor