Most of what a founder knows about selling is already captured in call recordings, transcripts and email threads with buyers. The work is turning that material into four documents a new rep can use: a written ideal customer profile, objection notes, a discovery guide and onboarding material. AI can do the first pass across the recordings, and your job as the founder is to correct what it gets wrong.
Why does this matter before the first hire?
Before the first salesperson starts, the founder is the sales process. You know which buyers close, which questions reveal a poor fit, and what to say when a prospect raises price or timing. Little of it is written down, because you’ve never had to explain it to anyone.
A new rep without that material learns by shadowing you and by losing deals. Shadowing helps, and it depends on your time, which is what you’re hiring to get back. Written material lets the rep prepare before their first call and gives you something concrete to point to when their approach drifts. What to do before your first sales hire covers the rest of the groundwork, and the sales process template gives you a place to put it.
When I joined startups as an early sales hire, the most useful thing a founder gave me was their rough notes on why the last few deals closed. A polished pitch deck with no explanation of who it worked on helped far less.
What material do you already have?
Look for:
- Recorded sales and demo calls, with transcripts, if you’ve used a recording tool
- Email threads with prospects, especially where a buyer asked questions or pushed back
- Proposals and follow-up emails you sent after calls
- Notes in your CRM or a shared table, however thin
- Closed-lost deals, and what you remember about why they were lost
Calls carry the most. The buyer describes their problem in their own words, asks the questions they care about, and raises the concerns that stall deals. That’s what a new rep needs to hear, and in many startups it sits in a recording library nobody opens once the call ends. What reaches the CRM is a few fields typed later from memory, which is the gap this post on missing customer conversations covers.
If you haven’t been recording, start now with each buyer’s consent, and fill the gaps from email and memory.
How do you use AI for the first pass?
Pick a set of calls that represents your deals: some that closed, some that were lost and some still open. Include each kind of buyer if you sell to more than one. Then work on one document at a time, giving the AI the transcripts and a specific question. You can do this in Claude or ChatGPT by uploading the transcripts; Claude for sales covers a setup. Keep customer data inside tools your company has approved for it.
Ideal customer profile (ICP). An ICP is a written description of the companies and people most likely to buy from you and stay. Ask the AI to list, for each closed deal, the type of company, the role of the person who brought you in, the problem they described and what made them start looking. Then ask what the closed deals share that the lost ones lack.
Objection notes. Ask for every point where a buyer pushed back, hesitated or raised a concern, grouped by theme, with your response in each case. You end up with the objections you hear and how you’ve handled them, in your own words.
Discovery guide. Discovery is the part of a sales call where you learn whether the buyer has a problem worth solving and the means to buy. Ask the AI to pull out the questions you asked that led to useful answers, and the answers that told you a deal was real or a poor fit.
Onboarding material. Once the first three exist, ask for a short guide a new rep could read in their first week: who buys, why they buy, what they ask and what to listen for. Have it link each section to an example call.
What does the founder need to correct?
The AI reads what was said. It can’t tell which deal was an outlier, which buyer came through a friend, or which objection came up once and never again. It will treat one vivid deal as a pattern, and it will miss what was never said out loud, such as the moment you knew a deal was gone from how the buyer answered.
Go through each draft and delete the patterns that came from a single deal. Add what you know and never said on a call: why you walked away from certain prospects, which signals you trust and which ones you ignore. Then put the points in order of importance, because the AI gives everything equal weight.
When I run this on a founder’s calls, the first draft of the ICP tends to describe everyone they’ve spoken to. The useful version comes after the founder crosses out the deals they wouldn’t want to repeat.
How do you keep it current?
These documents go stale as the product, the market and the buyer change. Give each one an owner, which is you until there’s a head of sales. Every so often, run the same first pass on recent calls and compare it with what’s written. Where they differ, decide whether the market has moved or the new calls are unusual, and update the document if it’s the market.
Once a rep starts, their calls feed the same process. Reading a new rep’s transcripts against the discovery guide shows where they’re following it and where they’ve found a better question, which is worth adding to the guide.
Of everything a founder handed me, the objection notes were the page I went back to most, and I added to them after every lost deal.