Services

The 90-Day AI Operating Partner Pilot

You own a business doing $5–50M a year. You already use AI and pay for it. What you don't have is someone to say where it goes, build it, and keep it running — so you never end up being the software developer.

That's the job. I work with your team, find where the business is slow, and use AI to cut operating cost or raise revenue. We start with 90 days and one number. If the number moves, we keep going for the year.

Tell me where the business is slow →

The partnership

A fractional AI CTO, not a project

A project asks "what should we build?" and then ends. The question I work on is bigger: what are we doing to get your business using AI across the board? That doesn't end after one build, so neither does the work.

I work with your team. The people doing the work know where it's slow. I sit with them, find the bottlenecks, and tie each one to a number.
Then I fix them, one at a time. Each job either brings a cost line down or a revenue line up. I build it, put it in front of the people who'll use it, and measure it against how the work gets done today.
Then I keep it running. Models change, vendors change, your own process changes. Somebody has to watch for that, and it shouldn't be you.
Already have a CTO? Then I work alongside them, and the focus is on putting automation into the work your team does by hand.

The pilot

How the 90 days run

Week 1

We pick the number

One number your business runs on — a cost line or a revenue line — and a target for it. We set both together, in writing, before anything gets built.
Weeks 2–12

I build and put it to work

I build the jobs that move that number and get your team using them. Thirty minutes a week, we look at the number together, so you always know where it stands.
Day 90

We look at the number

If it moved, I offer you the year: the same partnership, working down the rest of the bottlenecks. If it didn't, the guarantee below kicks in.

The guarantee

If the number hasn't moved, I keep working

In week 1 we pick one number and set a target. If it hasn't moved by day 90, I keep working, without charging you, until it does.

Two conditions, and they're both about your side of the table:

Your team gives me access to the systems. I can't move a number in software I can't get into.
Thirty minutes a week, every week. One check-in with whoever owns the number. That's how we both know it's moving.

Included with the pilot

Three things that come with it

Bonus 1

Your AI roadmap, in writing

Every bottleneck I find, in order, with the number each one moves. It's yours to keep, whatever happens on day 90.

Bonus 2

A training session for your team

The people who use the tools learn how to run them. So the questions go to them and to me, not to you.

Bonus 3

Your Proof Run carried straight in

If we've already done a Free 2-Week Proof Run, what it built and measured comes with you. Week 1 starts from something already working, not a blank page.

What stays true

How I work, every job

Built on your systems. Your AI account, your tools, your data. Nothing runs on anything I control, so there is nothing to migrate off if we stop.
You own the code. It's your business. The only exception is a product I already own and bring with me — that part stays mine, and I'll say so before we start.
Every job gets one question first. Whose day does this change, and does their life get better or worse? Taking away someone's data entry is a relief. Taking away the job that feeds their family, or the personal touch with a prospect, is not. If it isn't lined up with your people, they won't use it.
AI has a meter, and it's yours. Whatever I build runs on your own AI account, so the provider bills you directly. You see every penny and you can cap it wherever you like. It's usually small — working a list of 523 written-off leads cost about $200 in model usage from end to end — but it isn't zero, and anyone who tells you their AI system is free is billing you for it somewhere else.

What stops it doing something stupid

The question owners ask me most, in one form or another. Anything I build that talks to your customers or touches your records gets the same rails I built for the recruiting agents that text thousands of people a month:

It can only do what it's been given permission to do. The list of allowed actions lives in code you can read, not in the model's head.
It sends one message per turn, never a burst, and it won't message anyone outside the hours you set.
If it keeps failing to get somewhere with one person, it stops on its own and hands them back to you.
It drops anyone who no longer qualifies, so it isn't chasing a customer who already bought.
Every decision it makes is logged, so when you ask "why did it do that?" there is an answer.
Anything with a clock in it — the follow-up in five minutes, the next-day 9am message, the 48-hour close — gets tested on a fake clock in seconds, before it ever runs on a real one.

Why the rails exist: one unguarded loop I found later had run 178,745 requests and about $280 in five days on a cheap model. On a top-tier model the same loop would have cost about $8,000. It's the kind of thing you only see if somebody is looking, which is part of the job.

What I won't take on

If we can't name the cost line or the revenue line a job moves, it isn't work I want. I'll say so before either of us spends anything, and that rule doesn't loosen once you're a paying client — it's the whole reason the number on day 90 means something.

Start the pilot

Tell me what the business does and where it's slow. That's enough for me to say whether it's a fit and set up a first conversation.

Not ready for 90 days? Start with the Free 2-Week Proof Run: one job, built and measured, and you keep it either way. If you go on to the pilot, it comes with you.

Tell me where the business is slow

Two or three sentences. What the business does, what your team does by hand, and roughly what it costs you.