How to Run a Rollout Pilot That Tells You the Truth

An IT rollout pilot has one job: to be wrong early, while wrong is still cheap. Run well, it breaks your playbook in two stores instead of two hundred. Run badly, it becomes a ceremony, three easy sites, a green tick, and a program that meets reality for the first time in wave two with no budget left for the lesson.

Choose sites that fight back

The classic pilot mistake is selecting for success: the newest store, closest to head office, with the friendliest manager. That pilot can only confirm what you hoped. Select instead for coverage of your variance:

• One standard site, as a control
• Your smallest, most awkward format
• The oldest building in the estate
• A regional site with real freight distance

If the estate has a format you dread, pilot it. The dread is information.

Measure like it's science

Decide the metrics before the first pilot night:

• Task timings, step by step, against the playbook's assumptions
• Deviation count: every point where the crew improvised
• Evidence quality: did the checklist and photo requirements work in practice?
• Outage duration from the business's perspective
• The morning after: tickets, complaints, anything the store noticed

Deviations are the gold. Every improvisation marks a playbook gap that two hundred sites would otherwise each discover independently.

Debrief the crews, properly

The install crew knows things the dashboard doesn't: the step that reads clearly but works badly, the tool nobody packed, the part of the kit that arrives tangled. Debrief within a day, while it's fresh, and treat "the playbook was wrong here" as the most valuable sentence anyone can say. Crews who get punished for reporting problems stop reporting problems, and the program loses its sensors.

Update, then re-pilot if the changes are big

Pilot findings flow into a revised playbook, revised timings and sometimes revised staging. Minor updates: proceed to wave one, watching first-visit success closely. Major surgery, changed method, changed kit, changed window assumptions: run a second pilot. A week spent re-piloting costs less than a wave delivered on an unproven revision.

The gate must have teeth

Pre-agree the criteria that release wave one: timings within tolerance, deviations resolved, evidence pack workable, sign-off achieved without heroics. If the pilot misses, the gate holds. The most expensive words in deployment are "we're behind, let's push on anyway", spoken at exactly the moment the pilot has just explained what pushing on will cost.

The pilot's findings feed the deployment playbook, which then carries the program. Get the pilot honest and the rest of the rollout inherits its honesty.

Want a pilot designed to find trouble? Speak to an expert.