AI economy
UK firms are adopting AI. Depth matters more than the label.
Judge whether an AI tool changes a complete workflow or merely adds another drafting step.
Zocri · 2 min read · Sources checked
Adoption can be a shallow change
The ONS July 2026 analysis describes relatively shallow AI adoption among UK businesses with at least ten employees. This is a finding about surveyed businesses, not a verdict on every sole trader or microbusiness. It also does not measure the quality of an individual software product.
The useful question for an owner is how far a tool reaches into an actual job. Drafting an appointment reply is one step. Checking availability, handling a change and leaving the correct record for tomorrow's staff are other steps that may still require separate work.
Find the boundary of the improvement
Map one familiar journey on paper. For a home service, it might run from inquiry to estimate, acceptance, scheduling, visit and payment. Mark where an AI feature starts and stops. Then identify who is responsible for carrying the result into the next stage.
A small improvement can still be worthwhile. The problem is valuing it as though it improves the whole chain. If estimates become faster but jobs remain blocked by uncertain staff availability, the customer may experience little change in the time it takes to get a visit.
Test the handover, not just the draft
Give the tool a request with a missing detail and another with a changed date. Ask a colleague who did not create the draft to finish the job. Watch whether the colleague knows what has been checked, what is uncertain and what the customer was promised.
This is particularly useful in a small team where the owner cannot review every conversation. A good handover makes decisions legible. An unclear one can transfer time from the person writing the message to the person trying to reconstruct it the following morning.
Expand only after a complete job works
Choose one end-to-end journey and write down its acceptance conditions. Include the customer's receipt, the staff record and the recovery path when something fails. Measure completion and correction effort alongside the speed of the first response.
If the workflow is dependable, consider an adjacent step. If it is fragile, improve the handover before adding another AI feature. The ONS evidence motivates this distinction, while the sequence and checks here are our practical analysis. A deeper deployment should earn its place through ordinary work that becomes easier to finish.