16 June 2026

Copilot is installed. Nobody is using it.

Here is a conversation we keep having in 2026. A construction business bought Microsoft Copilot licences, sometimes dozens of them, because it seemed like the responsible way to "do AI". Twelve months later, the finance director is looking at the renewal and asking what the business got for it.

The honest answer is usually: a handful of people summarise the odd email, one enthusiast in the commercial team uses it properly, and everyone else forgot it exists. The licence did not fail. The rollout did.

Why utilisation stays low

It arrived as a feature, not a change. A button appeared in Word and Outlook. Nobody explained what it was for, in the language of the jobs people actually do. So it got clicked twice, produced something mediocre, and was dismissed.

Nobody translated it into construction. Generic training shows Copilot drafting a marketing email. Your QS does not write marketing emails. They write payment notices, scope clarifications and monthly commentary, and nobody showed them how it helps with those.

The data underneath is a mess. Copilot works over your SharePoint and Outlook. If your SharePoint is a decade of unstructured folders, the answers it gives are as confused as the filing. Weak data foundations quietly cap what the tool can do, then the tool gets the blame.

There were no agreed use cases. "Use AI more" is not an instruction anyone can follow. People need to know the five specific tasks in their role where the tool saves real time, and permission to use it for them.

What activation actually looks like

Turning shelfware into value is not a technology project. It is a short, practical adoption programme.

Pick the workflows first. For each role, identify a small number of weekly tasks where the tool demonstrably helps: minutes into actions, first-draft reports, inbox catch-up after leave, summarising a long document trail before a meeting.

Train by role, on real work. An hour with the bid team on their live tender beats a day of generic features. People adopt what solves their Tuesday.

Fix the ground it stands on. Straighten the SharePoint structure, naming and permissions so the assistant has something coherent to read. This is unglamorous and it is where half the value comes from.

Set the rules. What can and cannot go into which tools, what must be reviewed by whom, and what is out of bounds. Clear guardrails create confidence, and confident people use the tool more, not less.

Measure something. Utilisation, hours saved on the chosen workflows, adoption by team. Renewal conversations are easy when the number is visible.

The bigger point

This is not really about Copilot. The same pattern applies to any AI capability a construction business buys: the licence is the cheap part. The value lives in the translation, from generic capability to your workflows, your documents, your people.

That translation is precisely the work most rollouts skip, because it is not on the invoice. Budget for it, and a modest licence outperforms an expensive platform. Skip it, and the renewal conversation will keep being awkward.

If you already own the licences, you are one focused activation away from actually getting paid back for them. That is a far better starting point than buying anything new.