Insight

What the first week of an AI rollout looks like

On day one we install one controlled workflow. By day seven you can see it working and see the exceptions.

Blash AI · 19 June 2026

The first week of a well-run AI rollout, counted from the day the build is installed, ends with one workflow live on real work: narrow in scope, reviewed by a human at every step, and logging every exception it meets. Seven days is enough to produce working evidence, and working evidence is the only thing that moves an AI programme past the pilot stage.

The pattern is deliberately unglamorous. Wide rollouts fail in predictable ways, because every extra workflow multiplies the integrations, the edge cases and the number of people who must change how they work at once.

What happens on day one of an AI rollout?

One workflow is installed against a narrow slice of real volume, scoped to a single intake channel and a single job type. The baseline was agreed before the week began, so everyone already knows what the manual version costs and what improvement would count as success. A person reviews every output before it goes anywhere, which means the worst case for the week is some reviewing time spent, and the best case is a running system.

What should you be able to see by day seven?

Three things. First, the workflow handling live items without being nursed. Second, the exception log: the items the system declined to handle and passed to a person, which tells you where its confidence boundaries sit. Third, the first movement against the baseline, whether in captured enquiries, recovered hours or faster turnaround.

One of our builds, for a Midlands home-services business, went live in 7 working days and captured 54 missed or after-hours enquiries in its first month, with 81% of them logged with complete job details before anyone reviewed them. The month-one log settled the value question faster than any projection could have.

Why start narrow when the ambition is wider?

Because a narrow system running on Friday teaches you more than a broad system scheduled for next quarter. The exception log from week one becomes the tuning input for week two, and each expansion after that is grounded in evidence about how the system behaves on your work rather than on a vendor benchmark.

Scope then grows on the same terms it started with: one addition at a time, each measured against the baseline it inherits. Programmes built this way compound. Programmes launched wide tend to produce a long integration phase, a stale business case and a quiet cancellation.

Put it to work

Bring us the workflow this applies to

Book an AI audit
The newsletter

AI worth your inbox

The tools, launches and shifts that actually matter, in plain English. No paywall, unsubscribe at any time.