Skip to content

How it works

No black box. Here is the arithmetic.

Every figure is produced by ordinary, testable code, using methods taught in undergraduate finance and run by chains with analysts on staff. This is the whole method, so you can check our working.

Cash forecast

Holt smoothing, then the things you already know.

A baseline is projected from your trailing twelve weeks using Holt's linear exponential smoothing. It carries both a level and a trend, so a business that is growing forecasts as growing. Then the certainties are layered on top.

How one forecast week is built Three inputs stack into one weekly figure: a smoothed baseline from the trailing twelve weeks, plus recurring bills dated to the week they fall, plus open invoices dated to when they are due. The result is a projected closing cash balance for each of thirteen weeks. Trailing 12 weeks Your own takings, smoothed (Holt exponential smoothing) Recurring bills Rent, insurance, loan payments placed on the week they land Open invoices Money owed to you, dated to when it is actually due + Closing cash, 13 weeks zero week 12: short The forecast names the week you run short, not just that risk exists. Nine weeks of warning is enough time to collect, delay, or renegotiate.
Nothing here is an industry average. The baseline is your own trailing twelve weeks; the spikes are your bills and your invoices, on the dates they actually fall.

Why smoothing, not an average

A plain average of twelve weeks throws away direction. Holt carries a trend term, so a steady 2% weekly climb keeps climbing in the forecast instead of flattening out.

Why thirteen weeks

One quarter is the standard treasury horizon: long enough to see a problem coming, short enough that the numbers still mean something.

What it will not do

It cannot know about a booking you have not entered or a supplier rise you have not recorded. It forecasts the business as you have described it.

Customer segments

Three scores, eight segments.

RFM scoring is the oldest reliable idea in retail analytics. Each customer gets three scores from one to five — how recently they bought, how often, and how much — and the combination places them in a segment.

Customer segments by recency and frequency A grid with how recently a customer bought on the horizontal axis and how often on the vertical. Champions and Loyal sit top right; New and Potential Loyalist bottom right; At Risk and Needs Attention in the middle left; Hibernating and Lost bottom left. Bought more recently Buys more often At Risk Used to be regulars. Reach out this week. Loyal Steady and dependable. Ask them to refer. Champions Recent, frequent, high spend. Protect these people. Hibernating Long gone, low value. One cheap win-back. Needs Attention Slipping. Still winnable with a timely nudge. Potential Loyalist Coming back already. Give them a reason to stay. Lost Do not spend real money chasing these. New First visit was recent. The second visit decides it.
How much they spend is the third axis, and it decides ties: a lapsed big spender is At Risk rather than Hibernating, because they are worth the phone call.

Break-even

The sales figure that pays for the doors being open.

Break-even is fixed costs divided by gross margin. For the demonstration restaurant, $17,388 a month of rent, insurance and other fixed costs against a 74% gross margin gives a break-even of $23,499 — and the last thirty days came in $26,966 above it.

The formula

fixed costs
divided by gross margin
= break-even sales

Gross margin comes from your own ledger — sales less the costs that scale with them — so the answer reflects how you actually trade, not how a textbook restaurant trades.

Break-even for the demonstration restaurant Monthly sales of $50,465 against a break-even point of $23,499, leaving $26,966 of headroom above break-even. Last 30 days Sales $50,465 everything that came in Break-even $23,499 the point the doors pay for themselves Headroom above break-even $26,966 what is left to cover profit and reinvestment break-even crossed here A business trading below this line is losing money every day it opens, however busy it feels.
The scale is real: the break-even block is 47% of the bar because $23,499 is 47% of $50,465.

Health score

Four components, no mystery index.

The letter grade on the dashboard is the average of four scores, and each one names the figure that produced it. You can always see why it moved.

Profitability
Net margin over the last thirty days.
Revenue trend
Weekly direction over twelve weeks, discounted by how well the line actually fits.
Cash position
Months of runway at your current burn, or cash-flow positive.
Receivables
The share of open invoice value that is overdue.

Why there is no AI

Arithmetic first, and in this case only.

Plenty of tools would put a chatbot on top of this and call it insight. A language model is good at explaining and prioritising and bad at arithmetic, so ours was never allowed to do any — and today there is no AI feature switched on at all.

  1. 1

    You record

    Sales, costs, recipes and customers, typed in or imported.

  2. 2

    Code computes

    Forecast, margins, segments and scores, from unit-tested functions.

  3. 3

    A digest is built

    The finished figures are packaged as ground truth. Nothing raw, nothing invented.

  4. 4

    A model could explain

    If we ever switch one on it reads only that digest, and never your underlying books.

The practical consequence is worth stating plainly: every number on every screen today is computed by code you could audit, and would be identical if a model were involved. The analysis has never depended on one.

Check the working yourself.

Load the demonstration restaurant and every figure on this page is reproduced in front of you, from the same code. No account needed.

Open the demo