Ashwell Components Ltd, Cambridgeshire
Prepared following a three hour Compass session with the managing director, operations manager and senior estimator.
Ashwell runs a sound, profitable operation with a pattern we see often: skilled people spending a large slice of their week on work that follows rules. We identified roughly £114,000 a year of salaried time going on repetitive, pattern-following tasks, and six places where a focused AI system would hand a useful share of it back. The six are ranked below by annual return against the effort to get them live.
Our recommendation is to build the quote assistant first. It has the strongest return, the best data behind it (six years of quotes with known outcomes), and it removes the pain the team raised most often without being asked. The order-entry extractor follows naturally, because it reuses the same intake plumbing. Between them, payback lands inside the first quarter.
Nothing in this report proposes putting AI between Ashwell and its customers. The gains here are in research, preparation and administration; the relationships stay where they belong, with people.
Ashwell manufactures precision-machined components (around 60% of revenue) and distributes factored lines (around 40%) to OEM customers in agritech, rail and industrial equipment. Twenty-six people work in the machine shop; fifteen in the office across sales, estimating, purchasing and accounts. Systems are Sage 50, a shared RFQ inbox, and a good deal of Excel. There is no CRM, and nothing between the inbox and Sage except people.
The flow of a job: an RFQ arrives by email, usually a drawing plus a covering note. An estimator reads it, checks material prices and past jobs, and builds a quote. If it wins, the purchase order is rekeyed into Sage, acknowledged, scheduled, chased and shipped. Every step works. Most steps are manual.
Hours below were taken from the session and checked against a week of the RFQ inbox and the team's own diaries, not estimated from the doorway. Costs use loaded rates of £38/hour for estimating and £26/hour for office roles, over 46 working weeks.
| Activity | Who | Hours / week | Cost / year |
|---|---|---|---|
| Reading RFQs, looking up history, building quotes | Estimating + sales | 29 | £50,700 |
| Rekeying emailed purchase orders into Sage | Office admin | 18 | £21,500 |
| Answering "where's my order" enquiries | Sales + ops | 9 | £10,800 |
| Chasing suppliers on factored lines and materials | Purchasing | 8 | £9,600 |
| Month-end reporting and forecasting in Excel | Finance + MD | ~8 | £8,100 |
| Credit control chasing | Accounts | 6 | £7,200 |
| Tender and framework documentation (10 majors / year) | Sales + MD | ~4 | £6,000 |
| Identified repetitive, pattern-following work | ~82 | £113,900 |
That is not a criticism of the team; it is the reason the opportunity exists. Work that follows a pattern is exactly the work AI systems do well.
Six opportunities passed our bar (a real return, buildable with today's tools, no dependence on the customer noticing anything except faster service). Ranked by annual return against effort to get live, weighted by how confident we are in the numbers.
| # | Opportunity | Annual return | Effort to live | Payback | Confidence |
|---|---|---|---|---|---|
| 1 | Quote assistant | £30,000 | 5 weeks | ~6 weeks | High |
| 2 | Order-entry extraction | £17,000 | 3 weeks | ~5 weeks | High |
| 3 | Order-status replies | £8,500 | 2 weeks | ~6 weeks | High |
| 4 | Demand forecasting, repeat lines | £12,000 | 5 weeks | ~4 months | Medium |
| 5 | Credit-control drafting | £5,000 | 2 weeks | ~2 months | High |
| 6 | Tender first drafts | £2,500 | 2 weeks | ~3 months | Medium |
| Counted return, all six | £75,000 / yr |
An RFQ lands; the system reads the drawing note and covering email, pulls the nearest past jobs and current material prices, and hands the estimator a structured summary with a draft quote to check rather than a blank screen. The estimator stays the decision-maker; the reading and looking-up goes.
Saves ~60% of quote-preparation time (~17 hrs/week, £30,000/yr). Six years of quotes with outcomes make this the best-evidenced build. Faster quotes also tend to win more; we have counted nothing for that.
Emailed purchase orders are read automatically and arrive in Sage as a draft entry for a one-glance confirmation, instead of eighteen hours a week of rekeying. Shares its intake plumbing with the quote assistant, which is why it comes second: half of it exists once the first build ships.
~80% of rekeying removed (£17,000/yr), and rekeying errors go with it.
"Where's my order" emails are answered with a drafted reply built from Sage and the job cards, sent after a human glance. Customers get answers in minutes; the shop floor stops being interrupted for look-ups.
~£8,500/yr in office time, plus interruptions nobody costed.
For the ~300 SKUs that reorder on a rhythm, a forecast built from three years of Sage history sets suggested stock levels and flags lines drifting away from pattern, replacing the month-end Excel exercise.
~£12,000/yr in carrying and expediting costs, plus a one-off working-capital release we estimate near £40,000. Medium confidence until the history is extracted and checked, which is why it is ranked behind smaller, surer wins.
Aged-debt chasing sequences drafted in the company's own tone, escalating on a schedule, each one approved before it goes. The accounts team runs the judgement calls, not the typing.
~£5,000/yr, and typically a couple of days off debtor days, which we have not counted.
A library of past bids becomes the raw material for first drafts of new ones: the boilerplate assembled, the evidence sections pre-filled, the writing time spent on what actually differentiates.
~£2,500/yr as costed; the larger prize is capacity to bid on frameworks currently left alone, which only Ashwell can value.
A Compass is as much about what to leave alone. For Ashwell, this year:
First, the quote assistant. Strongest return, best data, and the pain the team named unprompted. A Blueprint Day designs it in detail; the build is about five weeks to a working system on live RFQs.
Second, order-entry extraction. It reuses the intake pipeline the first build creates, so the second build is faster than the first.
Third, order-status replies, a two-week build on the same foundations.
Forecasting waits until six months of disciplined Sage exports prove the history is clean; credit-control and tender drafting slot in whenever the office wants them, as they are independent of the rest. Taken together, the first three builds return roughly £55,000 a year for around ten weeks of build effort, with payback inside the first quarter.
If the ranking above holds after discussion, the next step is a Blueprint Day: a full day on site where we design the chosen system in detail with the people who will use it, and leave behind a build-ready specification with a fixed price for the build itself.
The approach is described at moonbootsconsultancy.com/ai-blueprint, or write to [email protected].