The one question worth asking your own sales data

The usual approach to business data is to look at it and hope something stands out. You open the sales sheet, sort a few columns, notice that one month was strong, and close it again none the wiser. This is not a failure of skill or software. It is what happens when analysis has no question, because data does not volunteer conclusions — it answers what it is asked, and an unasked question gets no answer however long you stare.

The fix is to start with one question specific enough to have an answer and consequential enough that the answer changes something. For most small businesses, one question outperforms every other candidate: what is different about the customers who come back compared with the customers who do not? It qualifies on both counts. It is answerable from records you already keep, and its answer points at a lever, because whatever distinguishes the two groups is something you have at least partial control over.

Repeat customers are the most valuable population in most businesses for an arithmetic reason: they cost nothing to acquire. Every rupee of acquisition spending buys a first purchase, and everything after that arrives free. So a business that understands what produces a second purchase has found the cheapest available growth, and it does not require more traffic, a bigger budget or a new channel.

The question is also unusually well-suited to small data. It does not need many customers to be informative, because it is a comparison of two groups rather than an estimate of a population value, and a clear difference between groups is visible at modest numbers. It works with the crudest possible records — a list of sales with a customer identifier, a date, a product and an amount. And its answer is nearly always actionable, because the distinguishing factor tends to be something you decide: which channel you spend on, which product you lead with, how you follow up. Compare this with the average vanity metric, which is precise, effortlessly available, and attached to no lever at all.

Start by splitting customers into two groups: those who bought more than once, and those who bought exactly once. Take care with the window — a customer who bought their first item last week has not had the chance to return, so exclude anyone whose first purchase falls inside your typical repurchase interval, or you will classify recent buyers as one-time customers and manufacture a pattern.

Then compare the two groups on the attributes you happen to have. Acquisition source is usually the most revealing. First product bought is next, and often the most immediately useful. First order value matters, and the direction is not obvious in advance — sometimes small first orders lead to loyalty because they are a low-risk trial, sometimes large ones do because they indicate a serious buyer. Then the time of year they arrived, whether they used a discount, and their location if that varies meaningfully. For each attribute, work out the proportion of each group falling into each category, not the raw counts, since the two groups are different sizes and raw counts will mislead you every time. A spreadsheet pivot table does all of this in a few minutes.

You are looking for a difference large enough to be worth acting on. If a third of customers acquired through one channel return, against a twentieth from another, that is a gap worth investigating regardless of statistical formality. If the figures are close, the honest conclusion is that this attribute does not distinguish the groups, and saying so is a real result — it removes a hypothesis and stops you from acting on noise.

The important discipline is not to convert a difference into a cause too quickly. A channel whose customers return more often may be attracting better-suited buyers, or it may be that those buyers were already familiar with you from somewhere else and the channel merely collected them. A product associated with loyalty may create the habit, or it may simply be what committed buyers choose first. Both readings imply different actions, and the data will not separate them. What it does reliably give you is a much better question to investigate, and the cheapest way to investigate is to ask several people in the group. A ten-minute conversation with five repeat customers about why they came back will usually explain the pattern the table found.

A finding that produces no change was entertainment. The action depends on what the difference turns out to be, and the mapping is mostly obvious once stated. If a channel's customers return more, shift spending toward it and check that the pattern holds as volume grows, since channels frequently degrade when scaled. If one product is associated with repeat buying, consider leading with it, featuring it, or making it the natural first purchase — while remembering the causal ambiguity above.

If discount-acquired customers return much less, that is among the most valuable things you can learn, because it means discounting is buying revenue rather than customers, and the acquisition cost of a discounted first order is higher than it appears. If a season or a location explains it, that shapes when and where you spend. In every case, treat the change as a test rather than a conclusion: write down what you expect to happen and check in a few months whether the repeat rate moved. That converts a one-off analysis into something that compounds, because each round narrows what you believe about your own business.

It cannot establish cause, and that limit is structural rather than a matter of getting more data. You are observing groups that differ in many ways at once, including ways you do not record, so any single attribute you find is entangled with others — the channel that brings loyal customers may also bring older customers, in a particular city, buying a particular product, and the table cannot separate those. It also says nothing about people who never bought at all, who are absent by construction and may be the larger and more interesting group.

Small numbers deserve genuine caution. With a few dozen customers per group, one or two individuals move a percentage noticeably, and a difference that looks decisive can evaporate next quarter. Look for gaps large enough that a couple of people could not explain them, and prefer patterns that persist across several periods to a single striking result. And a difference you find is a description of the past: it held for the customers you have already acquired, under the conditions that applied then. It is a hypothesis about the future, not a fact about it, which is why the next step is always a test rather than a commitment.