The chatbot answered. Finn did the work.
A PM sent us a WSJ story on the whey shortage and three questions about Indian dairy stocks. We put the same brief to Claude Opus 4.8 and to Finn. Both answered well. Only one of them was still working the next quarter.
A portfolio manager sent over a Wall Street Journal story from 22 June about the global whey-protein shortage. He attached three questions. Is this priced into Indian dairy stocks yet? What are the knock-on effects? And are the non-integrated brands more exposed than the market seems to think?
We ran the same article and the same three questions through two systems. The first was Claude Opus 4.8 on a Max plan, about as good as general-purpose AI gets right now. The second was Finn.
The chatbot wrote a good answer. Finn did the work an analyst actually needs done around it.
Credit where it’s due
Credit first: the chatbot was good. It knew that India buys most of its supplement-grade whey abroad. It noticed that Parag Milk Foods, the obvious winner here, had actually derated, and it put that well: known, but not yet rewarded. It read Milky Mist trimming its pre-IPO valuation as a sign of discipline rather than panic. Good calls, in about forty seconds.
So if you want one smart answer in a browser tab, that’s exactly what a chatbot is for. Finn runs on the same class of model, so none of what follows is about raw intelligence. It’s about what you build around it.
Chat vs Finn, at a glance
Dimension | Frontier chat | Finn |
|---|---|---|
Knows your book | Answers the question asked | Scopes to your portfolio & watchlist |
Coverage | The 5–6 names that come to mind | Full 9-name exposure table |
The math | Asserts the economics shifted | Computes it, shown in a chart |
Output | A thread that vanishes with the tab | A dated, referenced, archivable note |
After delivery | Stateless — brilliant, then amnesiac | A standing monitoring list that follows up |
It knew the book
Finn’s note opens by placing the idea. Neither the portfolio nor the watchlist holds an Indian dairy name, so this reads as a discovery note, not an update on something already owned. Sizing and conviction stay with the PM, where they belong.
The chatbot can’t do that, and it’s no fault of the model. It doesn’t know the book, the house style, or what’s already been looked at. It answers the question in front of it. Finn answers the same question for this particular investor, which quietly changes what the answer is.
It covered the field
The chatbot wrote about the five or six companies that surfaced as it typed. Finn built a table across the whole listed Indian dairy universe: nine names, market caps, whey exposure, how integrated each one is, and where the WSJ story actually touches them. That count includes the six where the honest answer is that it doesn’t.
One is a conversation. The other is coverage. An analyst has to account for the whole list, dull names included. A chatbot only has to produce a convincing paragraph.
It did the math
The chatbot said import economics had shifted. Finn worked out the number. US spot whey, converted to rupees a kilo, has landed above domestic Indian prices for the first time. The import-substitution case has flipped, and there’s a chart to check it against.
Then it valued the pieces. Parag’s protein and premium segment, at the multiples similar nutrition businesses trade on, comes out worth somewhere between 40 and 100 percent of the entire company. The market still prices the whole thing like commodity dairy. You can disagree with that. But it’s a number you can argue with, which is more than directional prose gives you.
It left something you can file
The chatbot’s answer sits in a thread, built for a screen, and it’s gone once the tab closes. Finn’s is a dated note with numbered sources, an exhibit, a risks section, and a straight answer to each of the three questions. You can forward it to a colleague, defend it in an IC meeting, and it lands in the one place a regulated fund already keeps everything: email.
There’s a compliance angle too. The note carries no recommendation by design, and it travels through channels that are already supervised. A chat transcript does neither.
It didn’t stop when the note was sent
This is the part that matters most. Finn’s note ends with six things to watch. Whether the protein segment keeps compounding next quarter. Whether the company raises prices. Whether import volumes fall. Whether US spot holds above the level that keeps the window open. For a chatbot, that’s a list you’d have to remember to revisit. For Finn, it’s a standing instruction. Each line becomes a watch item, and when one of them moves, the PM hears about it without asking twice.
A chat is sharp, and then it forgets. A research note is where coverage starts. That, more than any single section above, is what we mean when we call Finn a junior analyst.
The honest version
Chatbots are genuinely impressive, and the answer up top is proof of it. Reach for one when you want a smart reply to a question you’ll ask once.
A portfolio isn’t that. It’s a set of views you have to hold, test, and test again while filings stack up and your attention runs thin. That’s a job, not a chat. Finn holds the view, does the digging in the background, and emails you before it occurs to you to ask.
A chat is accountable for a plausible paragraph. An analyst is accountable for the universe — including the boring parts.
Curious how it reads on your own book? Send Finn a story that hit your sector this week and ask it the same three questions. The first note is the demo.