What was there when I arrived
“I feel like I’m an IT manager most of the time.”
The founder, December 2025
Clients pay for white-glove guidance through the move. The roadmap is the backbone of it: every step, in order, for their situation.
Building one meant a coach found a past client who looked similar, copied that roadmap, and adjusted 75 or more tasks by hand to fit the new client. Ten to fifteen minutes each, and the adjustments were judgment calls that lived in two places: the founder’s head and a folder of old roadmaps. Miss one of the rules that only applies to some clients, and the client finds out late.
It looked like personal attention. It was busy work. The coach’s time went into assembling the plan, and the client’s first experience of the roadmap depended on who happened to copy it.
And the business was growing, on a team the founder wanted to keep small.
Why back-office hand work like this is where automation pays off first is in Your Automation Budget Is Backwards.
Why not just hire someone
“I would love to be able to keep it where it’s five people.”
The founder, December 2025
That was the brief, and it rules out the obvious fix. Hire a coordinator to copy roadmaps faster and you’ve added a salary, left the rules in people’s heads, and still not given the client any more of their coach. Why pay someone to repeat a judgment you can write down once?
What I built
The client finishes the intake form. From there:
- A rule decides which tasks apply. Every task in the library is tagged by the parts of a client’s personal situation it depends on. Code matches each client’s answers against those tags. Nothing is left to a guess.
- The AI only writes. It takes the tasks that apply, tailors the notes to this client, sets the timing, and puts them in order. It’s told, in so many words, not to add or remove anything.
- A counter checks the AI. If a task goes missing between the rule and the finished roadmap, I get an email naming the task.
- The roadmap lands with the right coach. It’s built from her existing template, with the move date set and sections that don’t apply to the client left out. The client’s record is created in her CRM, and the coach for that country gets a message with the link and anything unusual to watch for.
Then the part that matters. The coach starts from a roadmap that already fits the client and spends the time on what only a coach can do: their own notes, their own guidance, the first real conversation. A destination the library doesn’t cover yet skips the build and goes straight to a person.
This isn’t built once and frozen. Requirements change, and the founder updates the library herself, so the next client’s roadmap already reflects it.
The numbers
| Before | After | |
|---|---|---|
| Assembling a roadmap | 10 to 15 minutes of copy-and-adjust by a coach | About 2 minutes from intake to a draft roadmap that fits the client, ready for the coach |
| Where the coach’s time goes | Building the plan | The client: personal notes, guidance, one-on-one attention |
| Where the rules lived | The founder’s head and past clients’ roadmaps | One task library of 142 tasks |
| Personal starting points | One per hand-adjusted copy | About 1,300 from one library, each then tailored to the client by the AI and by their coach |
What the build caught
The first version trusted the AI with the wrong job. In January, the AI decided which tasks went into a roadmap. That’s a call about someone’s visa, and a model can drop a line without telling anyone. By launch, code decides what’s in, the AI only decides how it reads, and a counter checks the AI’s work against the code. “The AI will include everything” is a hope. A counter is a control.
Some of the rules had never been written down. Reviewing the test roadmaps, the founder found tasks in her own library that didn’t apply to situations she hadn’t thought about before. Turning a coach’s judgment into tags forces every “it depends” into the open.
A failure that makes no noise. A tag spelled slightly differently from the form’s answer doesn’t throw an error. The task just quietly shows up for the wrong clients, or doesn’t show up at all. That’s how two tasks went missing for one test profile. The founder caught it in the last review round, before launch, and the library’s editing guide lists the exact values each tag allows.
The founder reviewed eighteen test clients before a real one. The test roadmaps ran across country and visa combinations, and the first real client went through after that.
Who runs what
“You’re just, like, working yourself out of a project.”
The founder, August 2026
She owns the content. Her coaches personalize and deliver the roadmaps, and the AI bill is hers. I keep the workflow running. That was the point: I was brought in to take the IT job off her plate, not to hand it back with a manual.
Some businesses I work with run the workflow themselves. She didn’t. Either way, it’s hers.
What it cost, said out loud
It took twice as long as planned. The proposal said five to six weeks; it went live in about twelve. Part of that was waiting on decisions, on both sides.
The library only knows what it’s told. When a rule changes, someone has to update it. The founder does, and that’s the deal: the roadmaps are only as current as the library behind them. Why every automated loop needs a named person who answers for it is in Should Your AI Agent Be on the Org Chart?
Delivery stays personal. Coaches review every roadmap, add their own guidance, and share it with the client themselves.