Illustrative example — this shows how a job like this typically runs, not a specific client engagement. Figures are representative, not measured outcomes.
Turning 'We Should Do Something With AI' Into a Shipped Feature and a Team That Uses It
How an 8-person SaaS startup went from board pressure and ad hoc ChatGPT use to one AI feature customers actually wanted, and a team trained to use AI safely.
Where they started
Investors and prospects had started asking the founders what their AI strategy was. A direct competitor had shipped a chatbot feature that customers found useless. Inside the company, a few staff were already using ChatGPT for support replies and code, quietly and inconsistently, with nobody having thought about what customer data was safe to paste into it.
The founders didn't know whether to build an AI feature into the core product, which of several ideas would actually help customers versus just look like an AI feature, and had no shared approach for how the team should use AI tools day to day. Building the wrong thing, or letting customer data leak into a public tool, both felt like real risks — and nobody at the company had the time or the mandate to work it out.
What three days with the team turned up
From the first read-throughCalls with six existing customers turned up nobody wanting to 'talk' to the software — they wanted the twice-weekly manual slog of re-scheduling around no-shows and cancellations handled automatically.
Automatic re-scheduling suggestions when a job falls through scored well above four other AI-flavoured ideas on customer value versus what it would cost to build.
Staff were pasting support tickets and, occasionally, customer job details into personal ChatGPT accounts, with no one having decided whether that was actually safe.
Everyone agreed AI mattered, but with no one holding the mandate to decide, it kept getting pushed to 'next quarter'.
How the three days ran
Done in 3 business daysTalking to customers instead of guessing
Called six existing customers about where manual work actually hurt, and ranked what came up by how often it bit and how much time it cost.
Scoring five ideas, building one
Scored five possible AI product ideas against customer value and build effort. Automatic re-scheduling suggestions came out well ahead — the other four were shelved before a line of code was written.
Rules for using AI, and someone to own it
Wrote a one-page policy on what can and can't be pasted into AI tools, picked one tool for the whole team to standardise on, ran a 90-minute hands-on session, and named one person as the ongoing owner of AI decisions.
What was handed over
- Customer interview findings on where manual work actually hurt
- Five AI product ideas scored on customer value versus build effort, with a build-or-shelve call on each
- A one-page policy on what can and can't go into AI tools, written for non-technical staff
- One AI tool standardised on, with the team trained and a named internal owner
What we built it with
We pick tools with low ongoing licence costs, solid privacy, and no lock-in — so you can take the work elsewhere if you ever want to.