Case study: 55 hours of research work, done with Finn in 19.5
An early user tracked six weeks of research workflows in a spreadsheet: hours with Finn, hours manual, and notes on quality. We’re publishing the log as recorded, including the error Finn made along the way.
One of our early users is a portfolio manager who tracks his own time. From early February to mid-March, he ran his research workload through Finn and logged every task: what it was, hours with Finn, hours the same work takes him manually, and notes on output quality. He was already using ChatGPT and similar tools daily, so the baseline here is not "no AI." It's a working PM who uses AI, comparing on his actual coverage.
The totals: 19.5 hours with Finn. 55 manual.
The log
Here is the log, as recorded:
Workflow | Finn time | Manual time | Observed outcome |
|---|---|---|---|
Product comparison and moat analysis (Adobe vs peers) | 3h | 4.5h | Strong insight quality; report well-made, with one stock-price error noted. |
Thesis validation (Transocean) | 3h | 5h | Helpful counterpoints that addressed the blind spots in the thesis. |
Business and scenario analysis (Oracle and SaaS downturn scenarios) | 3h | 7h | Business segments explained clearly; scenario work described as strong. |
Earnings updates across the portfolio | 2h | 6h | Newsletter-style synthesis; described as brilliantly done. |
Durability and moat analysis (Salesforce and Adobe) | 4h | 10h | One of the strongest outputs; quality and moat analysis described as top-notch. |
What the numbers show
The hours that disappeared were assembly hours. Pulling filings, re-reading transcripts, rebuilding context from last quarter before an earnings call. The judgment hours didn't move much, which is what you'd want, that part of the work is his.
Errors and caveats
Finn got a stock price wrong in the Adobe report and the PM caught it. He was testing an early version of Finn. We have added verification steps since then that catch this class of error more often, though no system catches everything.
Not everything compressed equally. The deep-dive and synthesis work saved the most time. Some of the recurring market updates were rated decent but marginal for how this particular PM works. The deep-dive work is where the subscription pays for itself.
And the obvious caveat: this is one user, and the manual hours are his own estimates of equivalent effort. It's a documented single-user comparison, not a study. We're publishing it because the per-task detail tells you more than any aggregate claim could, and because it matches what we hear from other early users.
What Finn is
We're publishing this because almost every prospect asks the same question: what does Finn actually do day to day, on real coverage. This time log answers it better than a demo.
Finn is an AI research analyst that works over email. It holds your portfolio context, watchlists, and open theses between sessions, runs deep dives and earnings prep, and monitors what you tell it to.
See it on your own book: request access, send one or two names you cover, and the first report is back the next morning.