Grocery retail· 2026· Pricing, reconstruction

What if the file has no prices in it?

No shelf prices, no costs, no promotion flags. The normal price was rebuilt from the history, proved against a second retailer, and only then used to price 39 moves.

The price reconstruction study cover: rebuilding the price from receipts alone.
The full deck The reconstruction, its validation, the exhibits and the limits Open PDF
The story

Most client exports look like this one rather than like a clean file. Receipts show what shoppers paid. They do not show what the product normally costs, so there is no way to tell a promotion from an ordinary cheap week. Volume reacts very differently to each. Everything downstream depends on getting that one column back.

What I did
  • Rebuilt the normal shelf price from each product-store's own price history, rather than assuming the most common price, or the highest one, is the shelf price.
  • Validated it before using it: ran the same reconstruction against a second public retailer whose file does record the true shelf price and the real promotion flags, and measured how often it agreed.
  • Estimated price response product by product and store by store, with displays and leaflet features controlled separately so merchandising is not billed to price.
  • Quoted the prize as a range produced by ordinary year-to-year volume swings, not as a single figure with false precision.
  • Capped every move near the price territory the product has actually traded in, and staged the rest.
Chapter 01

You cannot measure a discount without a normal price

A promotion and an ordinary cheap week look identical on a receipt, and they do very different things to demand. Without the list price, every promotion number downstream is guesswork.

Why a discount cannot be measured without a normal reference price.
The problem, stated plainly. The file records what was paid, never what was asked.
Rebuild the price, then prove it against a second retailer before using it.
Rebuild the price, then check it. An unchecked reconstruction is still a guess, so it was run against a retailer whose true prices are recorded before it was used anywhere.
The check is the whole difference. Run against a second public retailer whose file does carry the true shelf price, the reconstruction came back 0.11% below the truth on average and found 85% of the promotions that retailer actually ran. That is also why the promotion study on this file could be run at all.
Chapter 02

What price does to volume here

Volume here barely reacts to price. Most products sit well above −1, which is why a raise can pay and why deep discounting cannot.

Price response measured product by product and store by store, each with its interval.
Measured product by product, each with its interval. Published analyses of this same public file report responses three to five times steeper.
Why the difference, and why it matters. The usual reason is that volume is compared to price without separating out the displays and leaflet features running in the same weeks, so the merchandising effect gets charged to price. The stakes are commercial, not academic: at the steeper number a 5% rise looks like it would cost 14% of volume and nobody dares move. At the controlled number it costs about 4%, and the raise pays.
Chapter 03

How the prize is stated

Three rules, all of which make the number smaller and none of which is optional.

01

A range, not a single figure

Volume moves year to year for reasons that have nothing to do with price. The prize is quoted as the range those ordinary swings produce.

02

Capped, not optimal

Where the arithmetic points far above any price the product has ever carried, the move is capped near observed territory and walked in steps.

03

Each move pre-cleared

Every recommendation carries the volume it should lose beside the volume its own margin can afford to lose, both computed before anything changes.

04

Two things are not the client's

The unit costs are an assumed model and the shelf price is reconstructed. Both are labelled everywhere they appear.

The prize stated as a range, capped, and staged rather than as a single figure.
$56k to $63k in a realistic year, at assumed costs, and $179k over three years if the same moves hold. The band is drawn rather than hidden.
Reflection

What this study taught me

1. Rebuilding the list price took most of the study.

Everything a client wants to know about promotions and pricing runs through the list price, and it is the column most exports do not have.

2. A reconstruction is only worth what its validation is worth.

Anyone can infer a normal price. Running the same method against a retailer whose truth is recorded, and reporting the bias and the hit rate, is what separates it from a plausible guess. That test took longer than the reconstruction did.

3. Controls change the decision, not just the coefficient.

Three to five times steeper is not an academic disagreement. It is the difference between a business that dares to raise a price and one that does not.

4. Say which numbers are not theirs.

Assumed costs and a reconstructed price are both legitimate. Quietly mixing them in with measured figures is not. Labelling them everywhere costs nothing and is the reason the rest gets believed.

Interested in what your data can support?

Start with the question, not a proposal. If your export cannot support the study you had in mind, that is worth knowing in twenty minutes.

Ask what your data can support

  • No system access, no IT project, one export
  • Signed engagement letter and data processing agreement
  • France & Hong Kong, in English or French
kevin.larretche@squ.solutions How the first conversation works

What to put in the first email

The question. What you would do differently if you knew the answer.
The data. Roughly how many products, stores or channels, how far back, and whether unit costs exist anywhere.
The timing. When a decision has to be made.
All studies
Next study

Where did the growth actually come from?

↗