Every online store makes merchandising decisions constantly: which products to bundle, what to recommend at checkout, which customers are worth a win-back email this week, when a repeat buyer is due for a reorder nudge. Almost none of those decisions are made from the order data. They're made from instinct, last quarter's bestseller list, or whatever the merchandiser remembers seeing in a dashboard.
That's not a discipline problem. It's a resourcing problem. Mining a store's order history for real, statistically sound patterns (what actually sells together, which customers are actually slipping away, when a segment actually tends to reorder) is specialized, repetitive work. Hiring an analyst to do it for one account is expensive. Doing it for the ten or twenty accounts an agency runs is out of reach for almost everyone.
That gap is where a new role is taking shape: the Merchandising Engineer.
Not quite an analyst, not quite a merchandiser
The title borrows its shape from the same pattern that gave us "GTM Engineer": take a function that used to be a person's full-time job, pair it with the word "Engineer" to signal it now runs on a system instead of a headcount, and give the resulting hybrid its own name.
A Merchandiser decides what to feature, bundle, and discount, usually from experience and a handful of sales reports. A Data Analyst can run the numbers, but hands back a dashboard or a spreadsheet, not a decision, and rarely stays close enough to the merchandising calendar to know what to look for next.
A Merchandising Engineer sits between the two. The job is to keep mining a store's order history for patterns that matter (association rules, customer segments, timing) and to hand back something a marketer or merchandiser can act on immediately: a ranked list, a customer list, an audience, not just a chart.
What the job actually involves
Strip away the title and the day-to-day looks like this:
- Finding what sells together. Market basket analysis over the order history: which product pairs and combinations show up together more often than chance, ranked by how strong and how reliable the pattern is (support, confidence, lift), not just by raw co-occurrence counts.
- Catching cannibalization. Flagging product pairs that substitute for each other instead of complementing, so a bundle recommendation doesn't quietly cannibalize a higher-margin item.
- Segmenting the customer base. Recency, frequency, and monetary value scoring to separate Champions from At-Risk from Hibernating customers, so a campaign targets the segment it's actually meant for.
- Reading order sequences. Looking past single baskets to the order the products get bought in, and how often the same customer rebuys the same item, to time replenishment reminders instead of guessing at a cadence.
- Turning a finding into a list. A pattern is only useful once it's a set of customer emails sized for a campaign and ready to drop into an email tool, not a row in a report nobody opens again.
None of this is exotic. It's the same statistical toolkit retailers have used for years. What's new is doing it continuously, for every account, without waiting on a headcount that doesn't scale past two or three clients.
Why agencies feel this first
Merchants managing one store can eventually justify an analyst, or absorb the guesswork. Agencies can't. An agency running reports for ten client accounts would need ten analysts working in parallel to give every account the same depth of attention, which is precisely the math that makes the role uneconomical to hire for and unavoidable to automate.
That's the framing behind "a Merchandising Engineer on every account": not one more dashboard tool bolted onto the stack, but the hire an agency would make for every client if hiring were free. It reads the client's orders, products, and past analyses; answers questions grounded in that data instead of guessing; starts a new analysis when a question needs fresh work; and builds the resulting audience without anyone opening a spreadsheet.
What this looks like in practice today
This isn't a hypothetical job description. It's the shape of the assistant that now ships in every Affinsy workspace: it reads a client's order history and completed analyses, answers questions from that data (and says so plainly when something isn't in the data instead of guessing), runs a Market Basket or RFM analysis on its own when a question calls for new work, and turns the result into a sized, exportable customer list.
The title is new. The work underneath it isn't: it's the merchandising math that's always mattered, running continuously instead of once a quarter.