AI implementation: from one real process to live
Implementation is where AI plans go to die. I start small, with one process, and stay until it's running and your team owns it.
Where implementation goes wrong
Nearly nine in ten organizations now use AI regularly in at least one part of the business, yet only 37% attribute any profit impact to it (McKinsey, 2026). The gap between using AI and getting results from it is implementation.
It usually goes wrong in one of three ways:
- Too big. A company-wide rollout with no single process anyone can point to.
- Too vague. The AI is given a chat window and a loose request, and the output is almost right, which makes it useless.
- Too fragile. One person builds something clever, it runs on their laptop, and it stops when they go on holiday.
How I implement
Start with one process. Something that happens every week, that one person does well, and that has a clear finished result.
Write it down first. I sit with the person who does it best, watch them do it on real work, and ask why at every choice. The judgement calls are the part that's usually missing.
Build it where the work lives. The AI connects to the sheets, inboxes and tools your team already uses. I don't add software you'll have to learn.
Test it on one real project. Your team checks every output. Rules start strict and loosen as the team learns to trust it.
Hand it over. Your team learns to run it and change it, and I check in until it's normal.
What live looked like for one client
At a 25-person marketing agency I built a client dashboard that syncs every hour from the sheets the team already works in. If a sync reads bad data, it keeps the last good version instead of overwriting it. Each client sees only their own work. The team adds a new client themselves in under 10 seconds.
It was live for the first client about seven weeks after we started.
What you own at the end
Everything. The written process, the build, and the ability to change it. Nothing should depend on me, or on any one person's laptop.
Common questions
What does an AI implementation consultant do?
Writes down how a piece of work really gets done, builds AI to carry it out on your existing tools, tests it on real work, and trains your team to run it.
How long does implementation take?
A typical start is one process, one team, a few weeks. A larger build, like the agency's client dashboard, was live for its first client about seven weeks in.
How much does it cost?
It depends on the process, so I scope it on the first call rather than publish a number.
Why start with just one process?
Because one process that's properly defined and actually used teaches you more than a company-wide plan nobody can point to.