What should you look for when choosing an AI automation agency?

A list of agencies does not choose for you, and on the front page they all look alike. What does help: put the same route from your own company to every agency, and ask for an answer on paper.

Three things you want back in that answer. What runs after two weeks. Where it runs, in your environment or on the agency's platform. And who maintains it afterwards.

Ask for numbers as well. How many systems run in production today, at how many companies, and how long it took from intake to live. An agency that does not name its numbers usually does not have them.

Most surprises are not in the build but in the year after it. Do you pay licences, do you pay per user, and what does a bug cost that only shows up a month later.

Ask as well what happens if you stop. Do you get your data fully exported, are the accounts in your name, and can an administrator switch the system off without calling the agency.

Ask on top of that what it is built on. A connector tool is quick as long as the route is straight and the volume stays small. Once exceptions turn up, once a document has to be judged, or once someone wants to look back at what happened, a tool like that gets stuck.

A first assignment that is a whole programme is hard to check. A first version small enough to look over is not.

So ask for a scoped route and for a price that is fixed up front, rather than for a track with an hourly rate and an open end.

We put the answer on paper before a line of code is written. Within two weeks the first working version runs on real data, where possible in your own environment and your own accounts, and the price for that one use case is fixed up front.

We write the code and keep it running. You pay no licences and no amount per user, and your data and accounts stay yours.