Work / Case study
A sales pipeline the founder runs herself, without writing a line of code.
An early-stage software startup was heading into launch with a waitlist and one non-technical founder doing all the selling. I built an AI-run pipeline that she operates herself. The company signed its first paying customer during the engagement.
- Client
- Software startup, pre-launch
- What I did
- AI-run sales pipeline, founder setup, product work
- Who uses it
- The founder, who doesn't code
Where they started
- Every waitlist email and every follow-up was written by hand.
- Call notes were scattered, and there was no single view of who was where.
- On the product side, changes had no review process and the setup wasn't written down.
What I built
1. A pipeline that AI runs three times a day
A board with one column per stage: waitlist, invited, call booked, deciding, onboarding, won or lost. Three times a day a run does the housekeeping a founder never has time for:
- Picks up new signups and adds them to the board.
- Matches email replies to the right person.
- Spots new bookings and no-shows.
- Reads call transcripts and writes a summary, the to-dos, and tagged product insights: feature request, bug, onboarding issue, pricing signal.
- Runs the follow-up timers, drafts what's due, and reports back on what it did.
2. Rules it can't break
An AI that touches customer email needs hard limits. These are written into the process itself:
- Drafts only. It never sends an email. The founder reads and sends every one.
- It never emails the same person twice about the same thing.
- When it isn't sure, it asks instead of guessing.
- It never marks someone as lost without writing down why.
3. A setup she ran herself
I didn't install it for her. One prompt, pasted in, walks through the whole thing from zero: installing, connecting accounts, building the board, importing the waitlist, live tests, and putting it on a schedule. If her laptop died tomorrow she could set it all up again without me.
4. Outreach prompts
Prompts that draft one personalised email per person on the waitlist, ready for her to review.
"The prompt worked beautifully. Thank you!"
5. A bug that was quietly hurting replies
Links in the AI-drafted emails were expiring before people clicked them. I found the cause and wrote a guided, no-jargon fix that she applied herself. Use of the pipeline was uneven before that fix. It's the kind of problem that makes a team give up on a tool without ever knowing why it let them down.
6. Product work alongside the pipeline
I also worked on the product itself alongside the startup's own engineering: 18 pull requests, 8 of them merged, several within a day. And I built a full working prototype of a redesigned product in about 10 weeks.
Results
- 3× a day
- the pipeline runs on its own, and reports back each time
- 0
- emails the AI can send on its own. It only drafts.
- 18
- pull requests on the product, 8 merged
- ~10 weeks
- to a full working prototype of a redesigned product
Some honest notes. The prototype wasn't adopted as the product; it became the design reference for the main app. Some reporting work is still in review. And we didn't measure hours saved, so I won't quote any.
What I can say is that a founder who doesn't code now runs her own sales pipeline. Its rules are written in plain English, so she can read every one and change them.