Self-improving sales research
A sales intelligence system that learns from every call
Paste a potential customer's website and, two minutes later, the salesperson has a short, fact-checked prep sheet — who they are, whether they're worth pursuing, and how to approach them. Paste the call transcript afterward, and the system gets smarter for the next one.
Salespeople prepare for every call from scratch — and the hard-won lessons about what actually wins deals leave with whoever learned them.
One loop: paste a potential customer and get a fact-checked prep sheet in about two minutes; paste the transcript afterward, and the system learns. Every rep walks in deeply prepared, and every call makes the next one sharper.
| Before | After | |
|---|---|---|
| Prep per call | 30–60 minutes of hand-research across a dozen tabs | about 2 minutes — paste and go |
| What the team learns | leaves with whoever ran the call | stays in the system and builds up |
| Each call | starts from zero | sharpens the next one |
| To run it at scale | a dedicated sales-operations team | one person and the tools they already use |
| What the salesperson walks in with | a hunch | a fact-checked brief and a clear recommendation |
Prep dropped from an hour to two minutes — but that's the small part. The real win is the loop: every call makes the next one smarter, on its own. Sales know-how that normally needs a team to maintain (and usually never accumulates) now builds up by itself, and a new hire starts with everything the team has learned instead of from scratch. The result is the kind of deep, current prep that wins deals — on every call, not just the ones someone had time for. And there's nothing to host or maintain: it runs inside the tools they already use.
What I built
A sales-prep system the company's sales lead uses on every active deal. It isn't an app or a website — there's nothing to log into, host, or maintain. It runs as a set of AI routines inside the tools the team already works in (Claude and Notion).
- Research and prep. Paste a website, email, or company name and, in about two minutes, a prep sheet is ready. The system reads the prospect's whole website (skipping dead links), checks their social profiles and directory listings, works out how well they fit as a customer, detects which competing tools they already use by reading the code of their site, and writes it all up — grounded in a curated library of the company's own sales knowledge.
- A clear recommendation. Every prospect gets two simple ratings — how well they fit, and how worth pursuing they are — which combine into one of four calls: pursue now, play the long game, wrong fit, or pass. Each rating comes with a one-line reason tied to something specific the system found.
- Honest about what it knows. Every fact is tagged confirmed, inferred, or estimated. The system never assumes which tools a company uses — it only says so when it finds real proof on their site. Every brief arrives marked needs a quick review, and shaky ones carry a warning at the top. A blank field beats a made-up one.
- Catches bad-fit deals early. It automatically flags the warning signs that make a deal a poor fit, and lowers the score when it sees them.
- It learns from every call. Paste the transcript afterward and the brief updates itself — ratings corrected, objections logged, the approaches that worked recorded — plus the one lesson the rep would pass to the next person. That last answer is the most valuable thing the system collects.
- It compiles what's working. Across all those notes it surfaces patterns — labeling each by how much evidence backs it — and, each quarter, rewrites the team's full sales playbook from everything it has learned, including an honest list of what's still unknown.
Underneath it sits a hand-built library of the company's sales knowledge: its features mapped to real customer problems, how it compares to competitors, proven answers to common objections, why customers buy, and patterns from real won deals. The system reads all of it before it writes anything.
Why it matters
The payoff is straightforward: every salesperson walks into every call deeply prepared — the kind of preparation that wins deals, now applied to every deal instead of only the ones someone had time to research. The speed (an hour down to two minutes) is what makes that possible; the loop is what makes it stick. Every call makes the next one smarter, automatically — the system corrects its own judgment, learns which approaches land, and turns that into shared knowledge that no longer depends on any one person remembering. A new salesperson starts with everything the team has learned, not a slideshow.
Normally, "a sales system that learns from every conversation" describes an expensive enterprise product with a team maintaining it. Here it's one person's build, running entirely inside the tools the team already uses.
The hard part wasn't doing the research automatically — it was making the result trustworthy enough to use live, in front of a customer. An AI that states something false with confidence is worse than no prep at all. So accuracy is treated as a hard rule, not a nice-to-have: every fact is tagged confirmed, inferred, or estimated; the system never guesses which tools a company uses — it only says so when it finds real proof on their website; and every brief arrives marked "needs a quick review." A blank field always beats a made-up one. That discipline is what lets a salesperson open the brief mid-call and actually trust it.
Doing the work and improving the system are the same act here: every call feeds it, and it sharpens every call. That's the real shift — not a tool bolted on next to the team, but something that gets better the more they use it, quietly raising the floor on every conversation the team has.
How it rates prospects, how it works out which tools a company already uses, and the curated sales knowledge it draws on are what make the output trustworthy enough to use in front of a customer — and the part clients pay for.