How it works
One job at a time, measured against today.
Three steps. You always know what is being built, what it should save, and who approves the work.
The three steps
Each step ends with a decision that is yours: go on, or stop there.
-
Plan
A two-week paid look. We pick the one job costing you the most, measure how long it takes today, and agree what success looks like before anything is built.
-
Build
One assistant for that job, built for a fixed price and tested on real past work before it touches anything live.
-
Keep it running
We review the results with you each month, fix what needs fixing, and choose the next job together.
1. Plan: the paid look
Two weeks to find the one job worth fixing first. I sit with the people who do the work and record how long the repeat jobs take today. You get a short list of jobs ranked by what they cost you, one job picked with its starting numbers, and a written plan with a fixed price for the build. If AI is not the right fix, you hear that too.
The best first jobs have a paper trail: an inbox, past files or past examples that show what a good result looks like. If a job lives only in someone's head, the first step is getting it written down. More on that in how to pick the first job.
2. Build: one assistant for one job
I build one assistant for the job we picked and test it on real past work before it touches anything live. It starts with training wheels, the way a new employee does: it drafts, a person reviews, and the instructions get better every time it gets something wrong. Once it gets it right consistently, it does more on its own. Anything that goes to a client still waits for a person.
The first week decides whether a team adopts something, so I test and smooth it before handing it over, and train the people who will use it.
3. Keep it running
Anything you build will quietly break sooner or later: a password changes, a program updates, a new kind of request shows up. So the build includes checks that tell us when something is off, and each month we review the results with you, fix what needs fixing, and choose the next job together.
What I need from you
Less than most owners expect.
- One person who knows the jobSomeone who does the work today and can say what a good result looks like.
- Access to the tools the job lives inEmail, practice software, shared drives. Business accounts, set up properly.
- Past examplesThe more real history the job has, the faster the assistant learns it.
What stays in your hands
AI should make your team more capable, not make your clients wonder who did the work.
- A person approves the workNothing goes to a client or into a file without someone on your team signing off.
- One job at a timeSmall, clear steps your team can learn and trust, instead of a big change all at once.
- Measured against todayWe record the starting numbers, so the results are plain to see.
Start with a conversation
Tell Devan about the repeat work that takes the most time in your practice. If AI is a good fit, he will say so. If it is not, he will say that too.
Prefer email? [email protected]