On Built This Week, AI in finance is mostly about reading: data rooms, 10-Ks, profit and loss statements and transaction logs. The finance companies we hosted use models to pull answers out of documents in seconds and leave the judgment to people. Our own experiments with AI trading bots were fun and roughly broke even at best, which taught us that AI speeds up analysis but doesn't hand you an edge.
But if you can say no in, you know, say just a few hours of work instead of a hundred hours of work, then you can look at that many more companies.
— August Kiles, Episode 31
How companies we talked to use AI in finance
Emblem: due diligence for private equity and venture funds
Before AI, a private equity firm that got a data room sent associates to comb through it for dozens or hundreds of hours. Emblem indexes the data room the moment it arrives, answers questions in seconds, extracts financials, builds models and exports reports to Word, PowerPoint and Excel. August Kiles said it is most useful at the private equity stage, where data rooms have thousands of files, and that associates still do the last 5%. He doesn't see it taking entry-level jobs; it lets funds get to "no" faster.
Public: an API brokerage with AI research
Public launched Alpha, an investment research bot, a couple of months after ChatGPT came out, then added AI features for earnings call summaries, explaining why your portfolio is moving and portfolio construction. AI lead Jake Trefethen described "generated assets," custom indexes built from a plain-language strategy. Head of Product Emily Kurtz explained the new trading API as another front end for people who want programmatic access to their account.
Collective: agents in the solopreneur back office
Hooman Radfar runs an online back office for solopreneurs and says it saves members about $10,000 a year on average. Internally, Collective benchmarks each service flow, such as bookkeeping categorization and reconciliation, and moves it to agents for consistency and speed. For members, a chat tool now answers account questions like what you owe in quarterly taxes. His goal is a "CFO in your pocket": a super agent that queries an accounting agent and others on the member's behalf.
AI One: investigations in a regulated firm
Emanuele Melis's favorite deployment handles customer emails about money that used to take a person about a week to investigate. One agent pulls the transactions, another checks the documents, another reads the monitoring logs, and after a couple of hours a person has a report with every data point needed to reply. Every workflow starts with the "autonomy slider" at zero, so the agent only drafts and researches until people trust it.
What we built
- Public API trading bot: Jordan vibe coded it in Replit in three to six hours. Every 30 seconds ChatGPT decided whether to buy or sell IBIT using prices from Polygon. After about a dozen trades, $300 became $299. Jordan's words: "This is not investment advice."
- Kalshi prediction market trading bot: it flagged whale bets, coordinated entries and volume spikes. Jordan won 8 of 12 trades and still lost about $7 because fees ate the margin and, as he put it, it's bots trading on bots.
- Solopreneur finance dashboard for Collective: upload a P&L and balance sheet to get a cash-flow forecast, insurance compliance check and lending readiness report. Jordan had to switch to Gemini because the first model couldn't take that much financial data.
- Portfolio macro scenario simulator: Sam's tool for Emblem runs a scenario like rising regulatory pressure across a 30-company seed portfolio and sorts companies into advantage, stress and watch zones. August said a large share of Emblem's potential customers could use it.
We also use AI on our own numbers. In Episode 11, Claude turned Google's 57-page 10-K into a CFO-style dashboard; ChatGPT gave too much text and fewer visuals. Sam said we do the same with our own QuickBooks data and throw the dashboard away after.
What's still hard
- There's no free alpha. After the Kalshi bot, Jordan concluded the only real way to make money on those markets is inside information. Consensus signals get traded away, and fees take the rest.
- Vibe coding near money is risky. Emily drew the line clearly:
Obviously, like, making me vibe code something that we're gonna put into production that directly impacts, their withdrawals, funds, whatever it may be, is it's pretty risky.
— Emily Kurtz, Episode 8
- AI-written code can backfire. Emblem leaned on AI coding too early in 2025, shipped too many bugs and pulled back for most of the year before newer models made AI-generated code the majority again.
- Prototype to production is the last mile. Hooman said people still don't grasp how fast you can prototype, but moving from prototype to production is the hard part.
- Integrations lag. Collective had to build its own gateway to wrap an agent around Gusto because, as Hooman noted, it has no MCP.
Episodes on AI in finance
- Episode 8: Can Public Make You a Better Investor? with Emily Kurtz and Jake Trefethen (August 2025).
- Episode 11: a 10-K turned into a CFO dashboard with Claude (September 2025).
- Episode 14: Hooman Radfar on Collective and the $1.4 trillion solopreneur market (September 2025).
- Episode 26: Jordan's Kalshi prediction market bot (January 2026).
- Episode 31: August Kiles on Emblem and 100-hour due diligence (February 2026).
- Episode 46: Emanuele Melis of AI One on agents in regulated firms (June 2026).
FAQ
How is AI used in finance?
The companies we hosted use it to read and summarize financial documents: Emblem indexes private equity data rooms and builds financial models, Public summarizes earnings calls and market moves, Collective is moving bookkeeping to agents, and AI One uses agents to investigate money-related customer requests in a regulated firm.
Will AI replace financial analysts?
Emblem's August Kiles doesn't think so. Associates still do the last 5% and decide what to ask; the gain is ruling out deals in a few hours instead of a hundred, so a fund can look at more companies.
Can an AI trading bot make money?
Not in our tests. Jordan's ChatGPT bot on Public's API turned $300 into $299, and his Kalshi prediction-market bot won 8 of 12 trades but lost about $7 to fees. His takeaway was that the only real edge seemed to be inside information. None of this is investment advice.
Which AI is best for financial analysis?
In our tests Claude turned a 57-page Google 10-K into a CFO-style dashboard far better than ChatGPT. Emblem uses Claude and Gemini most and said the newest Opus made a big jump in building financial models in spreadsheets.