The products your customers buy together are already telling you something useful. Most businesses just never ask the question. Basket analysis does not require a data scientist or a specialist tool; it needs the transaction data you almost certainly already have.

What basket analysis really tells a mid-market retailer, on the data you already capture

Basket analysis, the study of what gets bought together, has a reputation as a thing the giants do. Amazon’s people who bought this also bought that. It sounds like the preserve of businesses with recommendation engines and armies of data scientists. It is not. The data it needs is the till and order data you already keep, for accounting, every single day. The technique is older than the internet, and a mid-market retailer or wholesaler is sitting on everything required to use it.

Affinity analysis is not just for Amazon.

Every order you take is a basket: a set of things bought together, by someone, at a moment. You keep those records because you have to, to invoice, to account, to manage stock. Affinity analysis, to give it its plain name, is simply reading back across all those baskets to find which things tend to travel together. You do not need a new system to do it. You need the order lines you already have, in a state clean enough to read. The capability has been in your accounts system the whole time, doing nothing.

What it tells you.

Done properly, it informs real merchandising and ranging decisions. Which products truly pull each other through, so they are worth placing together, promoting together, or bundling. Which lines are bought alone, so a discount on them does nothing for the rest of the basket. Where a gap sits, a product your customers clearly ought to be buying alongside something else and are not, which is range you are missing. For a wholesaler, which lines anchor an order and which ride along, which tells you what to protect on price and what to use to build the basket. These are decisions you make anyway, on instinct. This is making them on evidence.

A coincidence is not a pattern.

The trap, and the reason crude basket analysis embarrasses people, is mistaking the obvious for the useful. Bread and milk turn up in the same baskets constantly, but not because one drives the other. They are simply both bought by nearly everyone. A true affinity is when two things appear together far more often than their individual popularity would predict, and rarely apart. That is the signal worth acting on. The discipline is to measure how often a pair occurs together against how often you would expect it to by chance alone, and to keep only the pairs that beat chance by a margin. Skip that and you will proudly discover that your two best sellers sell well. Do it properly and you find the quiet, genuine links that change a layout or a promotion.

The questions to put to your own basket data.

You can frame the whole exercise as a short list of questions. Which pairs of products sell together far more than their popularity alone would explain. Which of my high-margin lines ride along with a popular low-margin one, so the loss leader is earning its place. Which products are islands, bought alone, where cross-promotion is wasted effort. And which expected pairings are missing, the affinity that should be there and is not, which is a ranging or placement gap. Those four questions, asked of data you already hold, are most of the value, and none of them needs a new platform to answer.

So this week, pull a month of order lines and look at one product you care about: what else is in the basket when it sells, and is any of it there more often than chance would put it there. That single look usually surprises people. Running this properly, separating the real affinities from the coincidences and tying them to ranging and promotion decisions, is everyday machine learning work for us on top of a clean data platform.

It is the kind of analysis we built for Mzuri, working their basket data in Azure SQL to surface the patterns that move a decision. The data was already theirs. The work was reading it well.