—Why this step matters
Step 9 · Sales forecasting on the roadmap
Every hire, every oven, every line of credit and every standing flour order is a bet on a number somebody forecast. In most businesses that number is a feeling, checked once a year against whether it felt about right.
A forecast that is simply noisy is survivable; you carry a little more buffer. A forecast that is wrong in a consistent direction is not, because you have built a plan on it — and consistent error hides inside an accurate total, where one channel's optimism cancels another's caution and the aggregate looks fine. This step is about finding out which one you have.
What follows is one worked example. The company figures are invented; the method and the published series are not.
Sales Forecasting & Revenue Analysis
The total is nearly right. Both halves of it are badly wrong.
Prepared for Bell & Rye Bakehouse
00Summary of findings
Across all three channels the six-month forecast was out by +2.3% — 43,800 cases forecast against 44,790 actual. On that number the forecast looks close to excellent, and it is the number the business has been judging itself on.
Split by channel, foodservice was +25.0% and grocery retail was −7.0%. Those are 11 times the size of the aggregate error and they point in opposite directions, which is exactly why they cancel and exactly why nobody has noticed.
This is bias, not noise. Every single month, grocery came in under forecast and foodservice came in over. Six months out of six in the same direction is not bad luck — it is a wrong assumption being applied consistently.
Flour, labor and oven time are bought against the forecast. Over-forecasting grocery means buying for cases that never sell; under-forecasting foodservice means turning down orders or paying overtime to fill them. Both cost money, and the accurate total means neither ever shows up in a variance report.
01The question
Bell & Rye builds one forecast, in cases, for the whole bakehouse. It has been within a few percent for two years and is treated as reliable.
Is the forecast actually accurate, or is it accurate in total and wrong everywhere? The distinction decides whether the answer is to leave it alone or to stop forecasting the business as one thing.
02Sources, and what is real
Invented. Bell & Rye Bakehouse, its three channels and all six months of forecast and actual cases. This document is a demonstration of a method, not a report on a real bakery.
Real, and checkable. The two published indexes in section 04, used only to make the point about what a case-based forecast cannot see:
| Series | Title | Jun 2026 |
|---|---|---|
| PCU311812311812 | PPI Industry: Commercial bakeries | 469.6 |
| PCU484121484121 | PPI Industry: General freight trucking, long-distance truckload | 204.6 |
Everything in section 03 is arithmetic on the invented table, and every figure in it can be recomputed from the numbers shown.
03Where the error actually is
| Channel | Forecast | Actual | Bias | Average monthly error |
|---|---|---|---|---|
| Grocery retail | 26,700 | 24,830 | −7.0% | 7.5% |
| Foodservice | 11,550 | 14,440 | +25.0% | 20.0% |
| Direct & market stalls | 5,550 | 5,520 | −0.5% | 0.5% |
| All three | 43,800 | 44,790 | +2.3% | — |
The bottom row is the one the business looks at. The rows above it are the business. Direct & market stalls is the only channel forecast well, at −0.5%, and it is the smallest of the three.
04What a case forecast cannot see
Even a perfect case forecast would miss something, and it is worth naming because it changes what the forecast is for. Over the last year the published price index for commercial bakeries rose +1.8% (PCU311812311812) while long-distance truckload freight rose +21.5% (PCU484121484121).
Cases do not carry that. Two channels with identical case volumes have very different economics when one is delivered forty miles in your own van and the other ships. A forecast in cases tells you what to bake. It does not tell you what the baking is worth, and the gap between those two numbers has been widening.
05How to challenge this
- Check whether the foodservice over-run is one large account rather than the channel. If it is one restaurant group growing, the forecast is not biased — it is missing a customer, which is a different and easier fix.
- Look at whether grocery shortfall tracks a specific store closing or a listing being dropped. Same logic.
- Re-run the six months in dollars rather than cases. If the bias reverses, the problem is mix, not volume.
06Recommendation
- Forecast the three channels separately. It is the same work split three ways, and it is the only change here that would have caught this on its own.
- Track bias, not just error. Error tells you how far off you were. Bias tells you whether you are off in the same direction every time, which is the only version that compounds. A column of six plus signs is the alarm.
- Add a dollars-per-case column by channel before the next planning cycle. It is the cheapest way to stop planning a business in units when the units are worth different amounts.
The honest summary: nothing here says the forecast is bad at forecasting. It says it is being asked to describe three businesses as one, and doing what any single number does in that situation — landing in the middle and being wrong about both ends.