Spare parts for commercial catering equipment (ovens, dishwashers, fryers, refrigeration), sold to service engineers and kitchens across the UK from a warehouse in Telford. 84 staff.
Verdict: About 3,800 contacts a month take 533 handling hours, £127,997 a year (£2.81 a contact). 40% (give or take 6%) is failure demand, caused by something not done or not done right, costing £75,881 a year; biggest: Wrong or damaged part received (£37,523); Invoice or credit query (£10,782); Where is my order (£10,542). A further 10% is avoidable value demand self-service could absorb (£20,419 a year). 19% of contacts are repeats. Staffing: Overall 60% of productive hours go on handling contacts, yet the rota misses the target in 2 of 6 bands (worst Mon 10:30-17:00, 57% in time). Reshaped, the work needs about 260 paid hours a week against 270: the rota’s shape, not headcount: move about 12 rota hours a week into the short bands. The 6 actions to do should remove about 307 contacts a month, freeing 476 hours a year (0.3 of a full-time post), worth £16,230 a year with 80% of the hours put to use (overtime or agency cut), for £9,760 in the first year (£5,400 cash): a first-year net of £4,441 (£8,801 counting cash only; -£2,659 if they do half as well), £14,070 a year after. Best for the money: “Log every promised call-back with a same-day deadline”.
Failure demand caused by something not done, or not done right Avoidable value wanted, but self-service could answer it Value demand what the team is there for
Each sampled contact is scaled up by its channel’s count from the phone, email and chat systems. The ± is the give-or-take on each reason’s share (95%, treating contacts as independent; contacts on the same day are alike, so the true margin is wider). Repeat contacts are 19% of the total; 84% are resolved first time. Satisfaction: 81% satisfied; effort not measured. A further 10% of contacts ask which part fits a model: wanted, but a parts finder could answer most of them (£20,419 a year).
| Reason | Contacts a month | Repeats | Resolved first time | Cost a year | Removed a month |
|---|---|---|---|---|---|
| Wrong or damaged part received 13 in the sample | 167 | 31% | 69% | £37,523 | 50 |
| Invoice or credit query 18 in the sample | 229 | 17% | 73% | £10,782 | 28 |
| Where is my order 41 in the sample | 501 | 30% | 78% | £10,542 | 142 |
| Did not understand returns process 26 in the sample | 333 | 19% | 77% | £10,255 | 40 |
| Chasing a call-back or reply 21 in the sample | 271 | 86% | 61% | £6,780 | 48 |
Wrong or damaged parts cost most because each one also brings a re-delivery and a return (£14 a contact on average, from the courier invoices), on top of the handling time. But only 13 of the 300 sampled contacts were about them, so the figure is rough: count the re-deliveries in the courier invoices before relying on it.
| Channel | Contacts a month | In the sample | Average minutes |
|---|---|---|---|
| Phone | 2,210 | 173 | 7.9 |
| 1,150 | 86 | 9.9 | |
| Chat | 440 | 41 | 7.0 |
| Band | Calls and chats an hour | On the rota | Answering after shrinkage | Answered in time | Rota needed | Rota hours a week short (+) or spare (−) | In time after the fixes |
|---|---|---|---|---|---|---|---|
| Mon 08:30-10:30 | 42 | 7 | 6 | 58% | 9.1 | +4 | 70% |
| Mon 10:30-17:00 | 18 | 4 | 3 | 57% | 5.3 | +8 | 65% |
| Tue-Fri 08:30-10:30 | 19 | 6 | 5 | 94% | 5.3 | −6 | 96% |
| Tue-Fri 10:30-13:00 | 12 | 5 | 4 | 95% | 3.9 | −11 | 96% |
| Tue-Fri 13:00-15:00 | 9 | 4 | 3 | 92% | 3.9 | 0 | 94% |
| Tue-Fri 15:00-17:00 | 7 | 3 | 2 | 81% | 2.6 | −3 | 84% |
Erlang C on 6.6 minutes on the line (the follow-up done afterwards is left out), treating each chat as a call and assuming nobody hangs up while waiting: a guide to where the rota is thin, not a forecast. Monday morning has its own, lower shrinkage (no training or meetings then). The fixes cut calls and chats by 9%; on their own they leave both Monday bands short, so move rota hours from the Tuesday to Friday mornings, which have hours to spare.
| Action | Plan | Contacts removed a month | Value a year | First-year cost | First-year net | Net, cash only |
|---|---|---|---|---|---|---|
| Scan every part against the order at packing Wrong or damaged part received · your own trial · 50% cut if it works, 60% chance · 13 sampled | Do | 50 | £10,745 | £5,360 | £4,042 | £5,802 |
| Dispatch email with a tracking link on every order Where is my order · test result elsewhere · 45% cut if it works, 50% chance · 41 sampled | Do | 109 | £1,872 | £1,760 | −£122 | £838 |
| Engineer's job number and delivery note on every invoice Invoice or credit query · best practice · 40% cut if it works, 30% chance · 18 sampled | Do | 28 | £1,059 | £1,000 | −£73 | £327 |
| One-page returns guide and online returns form Did not understand returns process · best practice · 40% cut if it works, 30% chance · 26 sampled | Do | 40 | £1,008 | £720 | £162 | £482 |
| Log every promised call-back with a same-day deadline Chasing a call-back or reply · best practice · 50% cut if it works, 30% chance · 21 sampled | Do | 40 | £821 | £320 | £398 | £718 |
| Tell customers the same day when a part goes on back-order Where is my order and Chasing a call-back or reply · best practice · 25% cut if it works, 30% chance · 41 sampled | Do | 41 | £726 | £600 | £35 | £635 |
| Website chatbot answering order status Where is my order · hunch · 20% cut if it works, 20% chance · 41 sampled | Park | 0 | – | – | – | – |
| Parts finder by appliance model on the website Which part fits my model · best practice · 30% cut if it works, 30% chance · 30 sampled | Park | 0 | – | – | – | – |
Freed hours count only as overtime cut, at time and a half on the hourly cost, and are capped at the £7,800 a year spent on overtime. Team time is costed at the team’s rate, yours at £40 an hour. Cash is ex VAT.
Contact counts come from the phone, email and chat systems over 3 months. The reasons, handling minutes and repeats come from a sample of 300 contacts, weighted by channel and classified by 1 person (30 second-checked); the failure-demand share is give or take 6%, treating contacts as independent: contacts on the same day are alike, so the true margin is wider, and reasons with few sampled contacts are rough. Handling cost uses £15.20 an hour and 24% shrinkage. Staffing uses Erlang C on calls and chats at 6.6 minutes on the line, treating each chat as a call, at whole people answering, and assumes nobody hangs up while waiting. Assumed: the chance each action works (60% tried here, 50% a test elsewhere, 30% best practice, 20% a hunch), 3 months to take effect, and 80% of freed hours put to use (overtime or agency cut); freed hours are worth nothing unless they cut a cost or go to paying work. Customers kept by better service are not counted.
The freed hours come to 0.3 of a full-time post: no post goes, and no ninth person is needed if the rota is reshaped. Re-run the sample in six months, with a second person checking a tenth, to see whether failure demand has fallen.
Fictional, continuing the example. Hessingley started the six actions on 2 November 2026 and chose the guided route. The agent was set up in an AI assistant the business already uses, connected read-only to the phone, email and chat systems and the helpdesk (its IT confirmed this before it committed). Early each month it re-runs the review in a new dated copy: the month's counts, and a sample of the month's contacts, each given a reason from the Reasons sheet, with 10% classified again by the team leader without seeing the agent's answer. It compares the month with the review's figures, using the owner's lines on Inputs: failure demand up more than 8 points, one failure or avoidable value reason up more than 25%, or a reason with an action on it not below the review after 3 months. Nothing left the business, and no decision changed, without one of its own people approving it.
| Date | What the agent flagged | What was decided |
|---|---|---|
| 7 Dec No flag | November re-run on 214 sampled contacts: failure demand 37% (give or take 6%) against 40% at the review; 3,710 contacts. Nothing over a line. | Nothing needed. The operations director read the monthly note. |
| 8 Jan Flag | “Wrong or damaged part received” was up 46% since the review, to 244 contacts a month. Eleven of the 15 sampled were picked by the temporary Christmas staff. | The operations director put a second person on checking temporary pickers’ orders at packing until the scanners go live in February. |
| 8 Feb Flag | The team leader’s second check classified 4 of 21 contacts differently, all between invoice queries and chasing a reply. The agent flagged no reason figures that month. | The team leader and the operations director agreed that a chase about an invoice counts as an invoice query, updated the notes on the Reasons sheet, and January was re-classified. |
| 8 Mar Flag | Three months after the actions started, “Invoice or credit query” was not below the review: 236 contacts a month against 229. Invoices still lacked the engineer’s job number. | The change was stuck in the finance system supplier’s queue. The operations director got it a date: 20 March. “Where is my order” was down to 290 a month from 501. |
| 8 Apr No flag | March re-run on 219 sampled contacts: failure demand 31% against 40% at the review; 3,420 contacts. Nothing over a line. | Nothing needed. The figures were shared with the team. |
At the check-in on 22 December, about day 45 and after the first monthly run, Geoff and the operations director read the first note and the log together. No rule needed changing, and Geoff's upkeep of the set-up ended there. What the agent added. The method is unchanged: same reasons, same sampling, same checks, same decisions. The agent made sure the review was re-run every month, which a busy team would do once a year at best. One flag caught a cause the actions had not covered (temporary pickers), one kept the classification honest before any figure was trusted, and one found an action that had not happened. Failure demand fell from 40% to 31% by March; that is the actions working, and the review's own figures already count their value, so the months after add no £ figure.