Sample discovery calls Composite, not real clients

The playbook in motion.

The bold lines are the actual questions we ask, each labelled with the part of the conversation it is probing. The recommendation falls straight out of the answers. One call with someone new to AI, one with a power user already running hard. Same skill in both: right-sizing to the person instead of handing everyone the same lockdown.

New to AI

The service pro

Dana runs a two-person landscaping business. Books by phone and email. Has never really used AI.

Velocity What AI tools do you use today, if any?
Honestly none. My nephew showed me ChatGPT once. I haven't touched it.
Velocity Walk me through a normal job, first call to getting paid.
Someone calls, I visit the property, take phone notes, then at 9pm I write a quote in Word and chase them if they go quiet.
Velocity Where does the time actually go?
Quotes and chasing. And I lose jobs because I take two days to send a quote and someone beat me.
Stakes What is in those quotes? Anything sensitive?
Name, address, phone, pricing. Nothing like credit cards. Just me and my laptop.
Goals If one thing got easier tomorrow?
Getting quotes out same day instead of at midnight.

What we heard

Stage 0, low sensitivity, own machine, no AI yet. The pain is all in one place: quote turnaround. Nothing to lock down, so this is a clean first win.

What we recommend

Start with the intake-to-quote flow, nothing else. Set Dana up on commercial-terms tooling from day one, then build an assistant that turns the on-site notes into a drafted quote and follow-up. Dana reviews and sends.

You get

Same-day quotes. The 9pm block mostly gone. More jobs won on reply speed.

The contract

2 to 3 weeks. One flow built end to end, plus a sit-down to run it. Fixed price. A second area only once this one pays for itself.

Already using AI

The fast-but-messy power user

Marcus is a solo SaaS founder running hard with AI. Came in because something felt off, not because anything broke.

Velocity How are you using AI day to day?
Heavy. Agent in my editor, I let it run commands, it drives a browser to test my app. Sometimes two at once.
Velocity Full codebase access, runs without you approving each step?
Yeah, turned off confirmations a while ago. Way faster.
Stakes What can it reach if it does the wrong thing? Production?
...It is on my main machine. My production database creds are in a .env in one of those repos. So, yeah.
Stakes The browser it drives, separate profile or your own?
My own. Logged into everything. That is the point, it can just go.
Data terms What plan, and is the training setting off?
Pro plan. No idea where that setting is. And yeah, I have pasted customer support chats in.
Blast radius Secrets in your repo history? Any spend cap?
Probably, in history. No cap. Got a scary bill once.
Goals What brought you in, if nothing is broken?
I move fast and I know I am one bad command from a disaster. I don't want to slow down, I want to not blow up.

What we heard

Stage 3, genuinely fast, every safety dial wide open. Production reachable from an autonomous agent, the agent on his logged-in browser, training on with customer data in the mix, secrets likely in history, no spend cap. It has not bitten him. That is luck, not safety.

What we recommend, cheapest first

  • Free, today: flip the training opt-out.
  • Free / discipline: scan history for secrets, rotate, gitignore, add a spend cap.
  • Discipline: a separate agent browser profile, and production behind scoped tokens and a separate identity so a bad command cannot reach the real database.
  • A few dollars a month: support-data tools onto commercial terms.
  • Upside: route routine work to cheaper models, set up real multi-agent orchestration.

You get

The same velocity, the disaster scenarios closed, a lower bill. He keeps skip-permissions where it is safe, because now safe is actually true.

The contract

2 to 3 weeks of hardening and optimization. We map the blast radius, close it in order, and leave him faster and calmer than he came in.

Ready for your own?

The two-week paid discovery runs this conversation for real and ends with a written audit, a prioritized map, and one working prototype.

Start a discovery Browse sample engagements