Electrical control panels built to order for machine builders and water companies. 34 staff. Scope: panel build: from kitting through wiring, testing and despatch.
Verdict: 6 kinds of mistake, 173 a year, cost £28,140, 84% of it from mistakes the customer finds. The costliest: built to an old drawing revision (£10,752 a year). 7 devices should save £19,512 a year after £5,064 of running costs, including the time they add, for £12,000 one-off (payback 7 months), cutting mistakes that reach the customer from 47 to about 8 a year. 1 device doesn’t pay on its own: keep it only for a safety or customer reason; dropping it would make the net saving about £20,928 for £5,500 one-off. 2 of the mistakes are safety risks: not all are yet made impossible or stopped at the step.
| Mistake | Found today | A year | Cost a year | Devices remove | Saving |
|---|---|---|---|---|---|
| Built to an old drawing revision Kitting · counted | By the customer | 4 | £10,752 | 94% | £9,715 |
| Terminal screw not torqued Safety Wiring · counted | By the customer | 9 | £6,498 | 95% | £6,173 |
| Wire markers missing or wrong Wiring · counted | By the customer | 22 | £4,752 | 95% | £4,514 |
| Wire landed on the wrong terminal Wiring · counted | At final test or check | 96 | £2,592 | 94% | £2,319 |
| Wrong breaker rating fitted Safety Kitting · counted | At final test or check | 30 | £1,890 | 60% | £1,134 |
| Loose parts or offcuts left in the panel Despatch · estimated | By the customer | 12 | £1,656 | 50% | £720 |
A torque fault found on site costs £650 in travel and goodwill and four hours; the same fault prevented at the bench costs nothing. Wrong terminals are caught at final test today, so they cost hours of fault-finding but no cash.
| Device or method | Level | One-off; running a year with time added | Own net saving a year | Owner and date |
|---|---|---|---|---|
| Clutch-set torque drivers for every bench, with a torque-mark pen line on each tightened screw Stops: Terminal screw not torqued | Prevent: can't be done wrong removes about 95% | £2,400 £636 a year | £5,537 payback 5 months | Workshop manager 13 Nov 2026 · Trial |
| Wire markers printed from the CAD wiring list, each showing its destination terminal Stops: Wire landed on the wrong terminal | Prevent: warns before it happens removes about 60% | £1,800 £400 a year | £4,524 payback 5 months | Design lead 11 Dec 2026 · Idea |
| Point-to-point continuity tester run against the CAD netlist before power-up Stops: Wire landed on the wrong terminal | Detect: stops it at the step removes about 85% | £6,500 £2,180 a year | −£1,416 doesn’t pay on its own | Test engineer 29 Jan 2027 · Idea |
| Wire markers printed from the CAD wiring list, each showing its destination terminal Stops: Wire markers missing or wrong | Prevent: can't be done wrong removes about 95% | £0 £0 a year | Same device as above: counted there | Design lead 11 Dec 2026 · Idea |
| Kit trays laid out from the bill of materials, with each breaker's rating label facing up in a marked slot Stops: Wrong breaker rating fitted | Prevent: warns before it happens removes about 60% | £600 £840 a year | £294 payback 24 months | Stores lead 27 Nov 2026 · Idea |
| Drawings printed only from the job record, with the revision in a large box and old prints recalled Stops: Built to an old drawing revision | Prevent: warns before it happens removes about 60% | £0 £0 a year | £1,555 no one-off cost | Design lead 30 Oct 2026 · Trial |
| Shadow board for each panel's loose parts, and a magnetic sweep and photo before the doors close Stops: Loose parts or offcuts left in the panel | Detect: warns at the step removes about 50% | £300 £672 a year | £48 payback 75 months | Workshop manager 6 Nov 2026 · Idea |
| Kitting sheet printed from the job record carries the revision; the kitter scans the drawing barcode, which must match before kitting starts Stops: Built to an old drawing revision | Detect: stops it at the step removes about 85% | £400 £336 a year | £2,928 payback 2 months | Design lead 27 Nov 2026 · Idea |
Each device is judged on what it adds on top of the others, after its running cost and the minutes it adds to every panel. Clutch-set torque drivers make an untightened screw impossible and pay back in about five months. The revision barcode check costs £400 and stops an old drawing at kitting: payback under two months. Printed wire markers serve two mistakes from one printer. The continuity tester is the one to drop or rethink: with the markers in place it adds about £764 a year but its ten minutes on every panel cost more than that, so it pays only if test time is cut elsewhere. The shadow board barely pays (£48 a year): keep it for the customer, not the money. The kit trays pay slowly, and a wrong breaker is a safety risk covered only by a warning: scanning each breaker against the bill of materials at kitting would stop it at the step, and is worth doing whatever the payback. Trial each device on a few panels before buying in bulk, and allow a few weeks before it reaches its full effect.
Mistakes are counted over 12 months, from records where marked Counted, and priced from real cases: cash costs, plus overtime only for the share of hours that causes it. How much each device removes is an assumption by level (can’t be done wrong 95%, check later 30%), to be confirmed by a trial; devices of the same kind on one mistake are not added, and prevention and detection are combined, not added. Detection saves the cost of the mistake escaping, and saves hours only where it is quicker to put right at the step. Running costs, including the time each device adds to every job at the full hourly cost, are taken off. Better reputation, fewer lost customers and safety are not valued: a safety risk is treated by the Checks, not by the payback.
Fictional, continuing the example. After the design, Merrowfield chose the guided route and set up the agent in an AI assistant the business already uses, connected read-only to its test log, its customer complaints log and its service call-out records (its IT confirmed this before it committed). On the first Monday of each month it counts each mistake's escapes, updates the 12-month counts in a new dated copy of the workbook, runs the 15 checks, compares each mistake with what its devices should leave and each device with its due date, logs what it finds and sends the workshop manager a short note. Nothing left the business, and no decision changed, without one of its own people approving it: the agent never changed a device, a status or a drawing.
| Date | What the agent found | What people decided |
|---|---|---|
| 2 Nov 2026 No flag | Drawing revision control in use; no panel built to an old revision since | None needed |
| 2 Nov 2026 Flag | A loose-connection call-out (terminal screw not torqued), a safety risk, on a panel wired on 20 October | Wired on the one bench still without a clutch driver during the trial; drivers fitted to every bench by 13 November, as planned |
| 7 Dec 2026 No flag | Wrong breaker ratings at final test down after the kit trays; drawing barcode check in use | None needed |
| 7 Dec 2026 Flag | Shadow board for loose parts more than your 14 days past its due date and not in use | Workshop manager to make the board this week; in use from 14 December |
| 4 Jan 2027 Flag | Wire markers missing or wrong: 2 customer complaints on panels wired after the printed markers came in, a device meant to make it impossible | Two late wiring changes were marked by hand. Rule: every change is reprinted from the CAD wiring list before it is wired |
| 4 Jan 2027 No flag | Wire landed on the wrong terminal: fewer at final test since the printed markers, within your line | Watch January, the first full month with the markers |
At the check-in on 26 November, after the first monthly run, Geoff and the managing director read the first log together and left both lines where they were. After that the agent, its accounts and its log were Merrowfield's.
Each flag pointed at a device that had stopped working, or never started: a bench without its clutch driver, a shadow board not made, and hand-written markers on late changes. Each was put right within a week. The counts are too few months in to confirm the saving the design expects; the continuity tester, due in January, is next. Reading the monthly note and recording decisions took the workshop manager about 20 minutes a month (estimate). The months after add nothing to the figures above.
Fictional and labelled: these events continue the worked example. The same six events are the sample rows in the Agent log of the worked-example spreadsheet. The agent is part of the guided and done-for-you routes; the self-implement kit does not include it.