Every study ships with the data as used and the code that produced the figures, so your own analyst can reproduce any number without asking me.
Every study ships with the data as used, the code that produced the figures, and a workbook holding every number behind every slide. Your analyst can reproduce any figure without asking me.
Where something is assumed, it says assumed, on the slide, in the workbook, in the file name if necessary. Most published demonstrations use assumed unit costs, because the public files carry none. One runs on a generated file. Both are stated on the page, next to the money.
Some questions your data cannot answer. When that happens the deliverable says so, explains what is missing, and designs the test or the export that would answer it.
In the published pricing study, two competitor pairs were refused by name because their prices always moved together. That is one pricing decision across three products, not shoppers switching between them. A confident number there would have been a guess.
In the promotion study, roughly half a million promotion weeks on slow-moving products were skipped rather than estimated. Their gaps could be stockouts or delistings, and a baseline built on gaps is fiction.
The basket work could have been presented as rescuing the promotion calendar. Most of that apparent rescue turned out to be a side effect of how the count was built, so the study reports the honest figure and hands over the threshold the client can judge for themselves.
Before running anything, I write down which category should come back weakest, then check whether it did. Estimates are fitted on early years and tested on later ones. Where they drift, the drift is reported.
One export: product by store by week or day, with units and revenue, two years or more. Unit costs where you have them. Without them the study gives directions, not exact prices. Promotion flags if your system records them. Any column names, CSV or Excel.
No system access, no integration, no IT project. Customer names and personal data are not required for any of these studies and should not be sent.
Files go through a private workspace under a signed data processing agreement, are used only for your engagement, and are deleted at the end of it on request.
Everything: the coefficients, the intervals, the formulas, the workbook and the code. None of it depends on me staying involved.
Consumer brands and retailers with at least a couple of years of product-level sales history, where people set the prices rather than an enterprise pricing system. Usually a CFO, a finance director or a commercial director who wants an answer, not a software project.
I'm Kevin Larretche, a data science and FP&A consultant. I spent my FP&A years at Devialet, in premium consumer electronics: owning budgets, closing months, and defending forecasts in the meetings where a number has to hold up. Rennes School of Business.
I build the statistical models myself, which is why every study ships with its code. Analysts who have never sat in a budget review produce coefficients; finance teams without the statistics produce opinions. A price coefficient on its own is not a decision, and turning it into a yes or no on a specific product is the part that needs both.
I am not a grocery specialist, and that is deliberate. The published studies run on retail data because that is what is public. The seat I have sat in is the one you are sitting in. I work from France and Hong Kong, in English or French, and studies are contracted through SQU Solutions, Hong Kong. LinkedIn.
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.