A week of reporting work, now one click and a review.

The report mattered, but building it took most of the week.

Every week an analyst pulled figures from three separate servers, moved them through spreadsheets, rebuilt the charts and updated a forty-plus slide deck for each market, then checked it all before it went to the client.

Building it by hand also meant a slide could carry last week's number without anyone noticing until the client did.

A fast start, with nothing extra asked of the client.

The operator recorded himself doing the work he already did. There were no workshops, no specification and no time taken out of his week.

He walked through his week on screen: logging in to each server, running the queries, exporting, cleaning the spreadsheets, rebuilding the charts and pasting them into the deck. AI turned the recordings into a step-by-step map of the process, and we redesigned it from there. When one page was still missing, a six-minute video of it became its specification.

Recording → step map → redesign → safety checks → build

Only the right steps are automated, and a person answers for each one.

We split the work into one-job steps and decided, step by step, whether a rule, AI or a person should do it. Judgement stayed with the operator. One step was taken out altogether.

Our method for this is the 4Ds: Divide, Decide, Demarcate, Declare.

One click builds all three decks and flags anything out of date.

The tool reads the warehouse directly, works out the measures once and writes them into the client's real deck, in the operator's own slide order. The operator picks the week, presses run and reviews the output.

• Thirty warehouse queries replaced the manual pulls.
• The client's real template. Slides are matched by market and title, so the deck can be reordered and the tool still finds the right page.
• Stale slides are flagged. A page the tool could not refresh is marked in red on the slide and in the speaker notes.
• Nothing leaves the laptop. The tool reads, computes and writes locally, and has no send path of its own.

Where AI does the work. AI built it in days. An AI engineer drafted the step map from the walkthroughs and built the tool, a separate AI reviewer checked each build and sent back work that fell short, and one person led. Rules run it. Every figure in the deck comes from a rule, and none passes through a model.

A week of work became a click and a review.

Ten days from first email to all three markets' decks and output sheets built in one run. No new staff and no platform migration. The client's own operator runs it: he picks the week, presses run and reviews the output. Data stays on his laptop. The operator now spends his time checking the story instead of assembling pages.

What's next. The next stage adds a second data source, separate runs per market, AI-drafted slide commentary for the operator to edit and sign off, and a stop the operator controls. It will be built the same way, redesign first, then build in days.

£13,000 to £17,000 a year of senior analyst time, handed back

• About 7.5 hours of hands-on work a week, 345 hours a year, from a senior analyst.
• At a senior analyst's cost (salary of £55,000 to £75,000 plus employer National Insurance and pension, £37 to £50 an hour), that is £13,000 to £17,000 a year.
• The work was spread across the week, so each sitting pulled him out of something else. Allowing a conservative quarter on top for switching in and out, the real cost was nearer £16,000 to £22,000 a year.
• Now it is one run and a review, and he spends the time saved on other work.