Seats aren't adoption.
Your team has AI. Your company doesn't.
We put AI inside the work — and it isn't finished until your team is running it.
You bought the licenses and a few people use them to rewrite emails. Nothing about how the company works has changed, because individual productivity doesn't compound and nobody redesigned the work. We start with a two-week paid workflow audit, agree in writing on exactly what changes, then build it and train your team to run it without us.
The problem
You bought the licenses a year and a half ago. Somebody on the leadership team is enthusiastic about it. A few people use it to rewrite emails. Nothing about how the company works is any different.
That isn't the tools failing, and it isn't your team being slow. Individual productivity gains don't compound. One person works out something clever on Tuesday and nobody else hears about it. The AI can't see the systems where your work actually lives, so it helps with the writing and not much else. Nobody decided which work was worth changing, so the visible work got attention and the expensive work didn't. And nobody owns any of it, which means it's nobody's job.
You've been here before with something else. You bought a tool, or hired a consultant, or ran a process improvement, and six months later everyone was back to the old way. It reverted because nothing underneath it got redesigned, so there was nothing for the new thing to hold onto. That's the same reason this hasn't landed, and it's the part we work on first.
What we install
We find the work that's actually expensive, rebuild it with AI inside it, and train your people to run it without us. What we're aiming at is capacity — not people working longer, but the work that used to eat the week no longer eating it.
We're not going to tell you AI will run your accounts payable. It won't. Nothing we build replaces a process end to end, and you should be skeptical of anyone who tells you otherwise. What we do is put AI into specific steps and leave the judgment and the approval with your people. In an AP process that looks like extracting the invoice, coding it, checking it against the contract or the PO, and routing the exception to a person. Four changes, each one testable, and not one of them is a claim that a model runs your finance function.
What you end up with is the workflows themselves, running on your systems under your accounts. Connectors into the tools where the work already lives, so the AI can see your business instead of guessing at it. A shared skills library with a practice for publishing to it, so what one person works out on Tuesday is how the team works by Thursday. Training for the people who'll use it, and a named owner inside your company who we train to run and extend all of it after we're gone. And a written list of everything we found and didn't do, in priority order, so you know what's next.
If you're paying for AI and nothing about how your company works has changed, that's a 30-minute conversation.
How it works
A workflow audit first, and it's paid. We map where your team's hours actually go and rank that work by what it costs you and what can realistically change. Then we write the document that governs everything after it. Every workflow we propose to change gets named, along with where the AI goes, what it produces, the quality bar it has to clear, and what happens when it can't handle a case. You read it, you argue with it, you sign it.
Start here
Workflow audit
$10K Two weeks
Stop here and you own the map — you spent $10K finding out what's worth doing. Keep going and the $10K comes off the build. The audit is also what tells us which depth below actually fits.
Then you pick the depth
One function
$35K
Four to six weeks
Sales, finance, delivery or operations — one of them, rebuilt end to end.
Company-wide
$95K
Three to four months
The expensive work across every function, prioritized by what the audit found.
Full install
$250K
Six months
Every function, your company knowledge base built alongside it, offline deployment where data can't leave the building, and us in your leadership rhythm throughout.
Fixed fees, agreed before we start. Model usage runs tens to hundreds of dollars a month, sized during the audit.
What we need from you: a named internal owner, attendance at the working sessions, and system access in the first two weeks.
The knowledge problem
Your business knows an enormous amount and can get almost none of it back. Three years of recorded meetings nobody will rewatch. Decisions sitting in a chat thread from last spring. And the person who understands how it all actually works is one resignation away from taking it with them.
For one software company's product org, we turned three years of meeting recordings into a structured issues log: failure modes organized by module, flags on the ones that fail silently, confidence levels, and a source on every entry so any line can be traced back to the conversation it came from. Then the test that mattered. Their departing domain expert reviewed it and confirmed it was right, before he left. That's the first question anyone should ask about AI reading three years of anything, and it's the one we wanted answered.
A log like that is built to get used. It takes a validation pass from the people who know the material, then a prioritization pass, and then it loads into the planning tool as a punch list the team burns down.
The same problem shows up smaller every week, in your meetings. Your notetaker captures the commitment beautifully and then it dies in a document nobody opens. It got recorded and it never got actioned. We build the connection most tools skip: meetings flow from your calendar through transcription into structured notes and a searchable knowledge base, and the commitments made in the room become tracked tasks in your task system, with the context attached, without anyone typing them in.
If your company's memory is one person's notice period away from walking out the door, that's a 30-minute conversation.
What finished means
We don't consider a workflow finished until your team is using it in production.
Not a demo, and not a pilot that quietly stops. The workflows we stand behind are the ones named in the audit — which is why the audit comes first.
Full terms in the agreement.
We run on this ourselves
Most firms selling AI run their own business on email and spreadsheets. You can check whether we do.
The meeting pipeline above is ours, and it's how this business keeps track of what it said it would do. Our proposals get written with our full client context and our own design instead of from a template. When we needed to get up to speed on government contracting, we built ourselves a primer and did it in a couple of hours instead of a couple of weeks. And we ran our whole sales-call history through a model, which turned up the pattern that changed how we sell: the deals we lost, we'd known early we were going to lose, and we chased them anyway. That one gates the deals we take now.
We'd rather say this plainly than let you assume otherwise. The deepest system we've built is our own. It's also why we know what actually installs in a business and what quietly dies a month after kickoff.
Common questions
Our data can’t leave the building.
Then it doesn’t. We deploy open models on infrastructure you control where the work requires it, and everything else gets built on your accounts under your access policies. You get a written data policy naming exactly what each vendor can and cannot see. That’s part of the design rather than a disclaimer at the end.
Which AI vendor do you use?
The one the job needs. We’re not a reseller and we take no vendor commissions, so the recommendation is about fit, and everything gets built so the model behind it can be swapped.
What happens when the models change?
They will. That’s why the value sits in the system around the model: your data pipelines, your process design, your integrations. When a better model ships, we swap it in and the rest keeps working.
Will this replace our people?
What we build keeps a person in the loop on anything that requires judgment, and the workflows get designed around approval rather than around removal. What you do with the capacity you get back is your decision and not ours. You should also be suspicious of anyone selling you AI on headcount math they can’t stand behind.
Should we just buy an off-the-shelf AI tool?
Sometimes, yes, and we’ll tell you when. Off-the-shelf wins when your process matches the tool’s assumptions. When it doesn’t, you’re back to contorting the business to fit the software, now with AI inside it.
What does it cost to run?
Model usage for systems like these usually runs tens to hundreds of dollars a month rather than thousands. We size it during the workflow audit, before you commit to anything.
What if the audit tells us not to bother?
Then we’ll say so, and you’ll have the map and the reasoning behind it. We’d rather lose the build than sell you one you don’t need.
More of what we do
Custom software
You’ve outgrown spreadsheets and off-the-shelf tools. We build software your team will actually use, and you own all of it.
Read more →Due diligence
A buyer’s lens on your business before a buyer takes one — findings ranked by valuation risk.
Read more →Financial models
Stop deciding on last month’s P&L. Forward-looking models with scenarios you can test before you commit.
Read more →Let's find the expensive work.
A 30-minute conversation about where your business is, what's not working, and whether we can help. If we can, we'll tell you exactly what we'd do first. If we can't, we'll say so.
Not ready to talk?
Get the Catalyst — Kevin's weekly newsletter on designing businesses that scale without depending on you.