A skilled trade business can finish the physical job and still have considerable work left to do. Notes need organising, materials need recording and the customer needs an accurate account of what was completed. That surrounding administration is a practical place to examine AI.

What happened

Fergus announced an AI-first direction for its trades job-management software on 13 September 2026. Its documentation describes an AI Assistant for finding information and supporting administrative tasks within the product. It also distinguishes that assistant from the broader Fergus AI experience, marked as forthcoming in the help material. The announcement should therefore not be read as proof that every described capability was generally available at launch. The announcement and the product explanation set out those distinctions.

The useful business angle is the work around a trade, rather than a claim that software can replace the physical expertise.

Why it matters

Consider a hypothetical maintenance firm whose technicians record notes at the end of a visit. A useful assistant might help organise that information into a draft job record. The technician would still need to confirm what was done, which materials were used and whether further work is required.

The quality of the original information matters. If a note omits a part or confuses two visits, producing a fluent summary will not fix the underlying uncertainty. It may make the incomplete account look more convincing.

I would therefore test the full chain from field information to customer-facing output. Check whether the proposed record is accurate, whether staff can correct it easily and whether the next person has enough context to proceed.

The evaluation should also include the awkward cases: a return visit, an incomplete job, an unexpected additional part or a customer who disputes the description. These situations help reveal whether the assistant reduces work or merely moves it to someone else.

The bigger shift

My reading is that AI adoption in smaller operational businesses may be most useful when it starts with recurring administrative friction. The task does not need to sound futuristic to matter.

But administrative work contains judgement. Turning notes into an invoice can affect what a customer believes they agreed to pay for. Updating a job record can change what the next technician expects to find. The organisation should identify where a draft becomes a commitment.

A pilot should therefore include clear review ownership. The person approving the output needs access to the underlying job information and time to resolve an uncertainty. An approval button without that context adds little value.

Measure the complete process rather than the speed of text generation. Useful questions include how long a checked record takes to complete, how often it needs correction and whether missing information is identified earlier. Include staff training and support effort in the cost.

The priority is a practical fit between the tool and the work, consistent with AI strategy based on business needs.

My take

I would start with one recurring administrative bottleneck that the team already understands. Give it a clear beginning, a clear end and a named person who can judge whether the result is usable.

Run a small supervised trial using ordinary jobs and a few difficult examples. Ask technicians and office staff where the effort changed, rather than relying only on the software’s activity count.

If the process produces more complete records with less rework, there is a credible case to expand it. The valuable outcome is a better-run trade business, with people retaining responsibility for the work and the promises made to customers.