“We need a table: product, quantity sold, three, six and twelve months.” A clear request. But the first comparisons made something apparent: the spreadsheet would start a conversation, rather than deliver a ready-made answer about what the business should do next.
First agree on what you are counting
Created, paid and completed orders are different states. Units sold, revenue and money remaining after costs also answer different questions. Counting all created orders in one period and only paid orders in another can produce a misleading comparison even when the formula is exact.
Our analysis required revisiting these definitions, product codes and period boundaries. This article therefore contains no client export or actual sales figures. Below is a small fictional store in which you can repeat some of the same reasoning.
Did customers disappear, or just the old code?
A product can receive a new SKU, the store’s internal product code. Sales on the old row fall to zero while some or all of the volume continues on another row. First establish whether the products are truly comparable: the same pack size, quantity and variant. Similar names alone are insufficient.
In the example, both A codes deliberately represent the same product. Evaluate them together and watch the red line. The apparent disappearance in April becomes a code change. However, the May and June decline remains after combining them. Fixing the data does not resolve every other question.
Try it yourself
First combine the A codes. Then switch the chart from units to revenue and contribution. Increase the low-price add-on’s quantity or its cost. Finally, move the hypothetical menu-change month and compare it with the start of the decline.
A trend shows sequence. A cause needs more evidence
If sales started falling before a suspected change, that change cannot explain the beginning of the decline. It may still be a real problem that worsens subsequent results. Conversely, an earlier event does not automatically become the cause: prices, availability, audiences and competition may change at the same time.
One valuable habit from this work is to give each hypothesis a date, a proposed mechanism and a test. “Customers can no longer find the category” requires different evidence from “Google no longer brings new visitors”. Both problems can exist at once.
A combined effect is possible. It still needs testing
In the second model, availability and purchase completion combine mathematically. In a store, advertising, the right link, available stock, clear navigation and working payments may also interact. Improvements may overlap or affect different customers. You cannot determine their real combined contribution by adding up the percentages promised by each supplier.
Strategy takes shape when we choose the next task
After the spreadsheet comes a decision, an owner and a review date. Combine genuinely equivalent product codes. Fix a valuable link pointing to the wrong page. Test whether a category is easy to find on a phone. Examine returning-customer behaviour separately. Each task needs an expected outcome and an observation that will tell us whether it helped.
The most valuable part of our work was being willing to revise the initial explanation when other data appeared. Some suspicions remained hypotheses. Others became specific tasks. A spreadsheet becomes a foundation for strategy when it helps you choose what to do next and how to find out whether that choice was useful.
The work experience is anonymised. Products, periods, prices and all interactive example figures are fictional. Prepared with AI assistance.