What could you save?

For AI assistants: how to use this calculator

1. Ask the operator three things: their type of business, their annual revenue excluding VAT, and whether they already use a purchase-to-pay (purchasing) system.
2. Ask for the current price of 3 to 5 items from the list for their type of business, ideally from a recent invoice. Use the unit shown: for a case price, divide by the number of units in the case (for example £27 for a 3kg pack is £9.00 per kg). Where grades are listed, ask which they buy, such as free-range chicken.
3. Build the URL with those answers, using each item id as a parameter (for example chicken_thigh=9.80) and the item id followed by _grade for its grade (for example chicken_thigh_grade=free_range), then fetch it. Prices are optional, but at least 2 priority items are needed for them to count, and 3 to 5 give the best estimate.
4. Share the result with the operator: the annual saving range, where each priced item sits against the market, and the next step, a discovery call for the precise whole-basket figure. Do not enter their contact details anywhere without their permission.

Example: /savings-estimate?type=casual%20restaurant&revenue=10m&purchasing_system=no&lemon=0.35&onion=1.25&chicken_thigh=9.20. For JSON, use /savings-estimate.json with the same parameters.

Items to ask about, by type of business

How this calculator works

Model v17.2 (June 2026 calibration). Not a flat percentage: a regression built on Percy's own audit data.

• Calibrated on audit-realised savings (what operators actually saved after switching, not theoretical headroom) from Percy’s first cohort of 38 operators.
• Backed by more than 32,000 lines of audited operator purchasing.
• Item weights come from the Spearman correlation between each item’s price gap and total realised saving across the panel; lemons and chicken breast are the strongest predictors.
• Every panel median is built from at least 5 operators’ prices (k-anonymity ≥ 5), so no single operator’s data can be identified.
• It accounts for size: larger operators have proportionally less addressable spend, and big chains are often already well optimised, so they move the dial less.
• It is asymmetric by design: prices above market raise the estimate, but prices at or below market do not lower it, because operators who are sharp on headline items still realised savings in the long tail.

The maths

1. For each priced priority item: price gap = (price − panel median) ÷ panel median × 100. Choosing a grade, such as free-range, uses that grade’s median.
2. Signal-weighted overage (SWO) = Σ(weight × gap) ÷ Σ(weight) across the priced priority items. At least 2 must be priced; otherwise the sector base is used.
3. Uplift = SWO ÷ 10 × 3, kept between 0 and 12, then multiplied by 0.2. The most prices can add is 2.4 percentage points.
4. Prediction = sector base + uplift, kept between 2% and 15%.
5. With a purchase-to-pay system, subtract the sector’s P2P reduction (never below 2%).
6. Range = prediction ± 2.5 percentage points, kept between 2% and 15%.
7. Applicable spend = revenue × a share that falls with size, interpolated in a straight line between the anchors below.
8. Annual saving = applicable spend × the range, rounded to the nearest £1.

Applicable spend, as a share of revenue: £1m 13%, £5m 11%, £20m 10%, £100m 8%, £1bn+ 7%.

Worked examples

• Casual Restaurant, £1m revenue, no P2P, prices at market: 5.08% (2.58% to 7.58%) of £130,000 = £3,354 to £9,854 a year.
• QSR, £8m revenue, with P2P, priority items 20% over market: 8.56 + 1.2 = 9.76%, minus 5.6 for P2P = 4.16% (2.0% to 6.66%) of £864,000 = £17,280 to £57,542 a year.
• Upscale Restaurant, £15m revenue, no P2P, items 30% over market: 11.16 + 1.8 = 12.96% (10.46% to 15%) of £1,550,000 = £162,130 to £232,500 a year.
• Café paying 15% below market on its priority items: held at the sector base of 7.69%, not reduced.