Why the merch director owns retail AI — not IT

There's a quiet assumption that runs through most retail organizations the moment a decision platform or "AI initiative" appears on the roadmap. The merch director reads the slide, recognizes the words — forecasting, allocation, markdown optimization — and files the whole thing under someone else's responsibility. "That's an IT project. Or a data project. I'll be consulted when they need the business view."

It's an understandable reflex. The vocabulary is technical. The sponsor is often the CIO or a freshly minted Chief Data Officer. The kickoff happens in a room full of engineers. Everything signals that this belongs to a function whose job is systems, not to a function whose job is commerce.

The reflex is also wrong — and expensively so. Because the decisions this system is being built to automate are not IT decisions that happen to touch merch. They are merch decisions, end to end. What to buy and in what depth.

How to split that buy across stores and channels. When to protect price and when to let it go. The platform is, in substance, an attempt to industrialize the merch director's own craft. Treating it as IT's project is treating your own job as someone else's.

This article argues the opposite of the reflex, hard. Not "merch should be in the loop." Merch should own it — and ownership means something far more specific than attending steering committees.

The decisions being automated are merch decisions

Strip away the technical packaging and look at what a retail decision platform actually decides. The list is short, and every item is a buying or merchandising call that has had an owner for as long as retail has existed.

Buying. How much of each style, in what size curve, at what initial markup, from which vendor, with what reorder option held in reserve. A model that recommends an open-to-buy allocation is making the same call a buyer makes. It just goes faster, across more options than any human can hold in their head at once.

Allocation. Which stores get depth, which get a token presence, which get nothing. How a limited first delivery is split when demand signals are still thin. Whether a flagship gets the full size run while satellites get core sizes only. These are category-role decisions, and the merch team has always made them — often on instinct refined over years of watching what sells where.

Markdown. When to take the first cut, how deep, on which clusters, in what sequence. Whether to protect a brand's price image past a certain week or let the category clear. This is where merch judgment is at its most loaded, because every markdown is a bet against the residual full-price demand the buyer believes is still in the assortment. As we've argued before, most of these markdown calls are still made by hand on a Monday-morning spreadsheet — which is precisely the work a decision platform proposes to absorb.

None of these are data problems with a merch flavor. They are merch problems that a data system is being pointed at — which is why digital merchandising is really a decision function, not a storefront. The forecasting model is a means; the buying, allocation, and markdown decisions are the ends. When the ends belong to merch and the means belong to IT, the only sane conclusion is that the project belongs to merch, tooled by IT — not the reverse.

The pre-season buy makes this sharpest. Assortment is decided once a season and then lives for six to nine months, and no in-season cleverness recovers a bad pre-season call. If a system is going to weigh in on that decision, the person accountable for the decision had better own the system. That person is the merch director, not the head of infrastructure.

What happens when merch delegates it to IT

Hand the project to IT, show up as a consulted stakeholder, and a predictable sequence unfolds. It rarely fails loudly. It fails by being ignored.

IT and the data team build a competent model. They gather requirements the way engineering gathers requirements — a workshop, a list of rules, a sign-off. The merch team contributes what it can articulate in a two-hour session: the obvious thresholds, the headline margin floors, a handful of vendor terms someone happens to remember. Thirty or forty rules get captured. Everyone signs the document.

The problem is that the merch team's real decision logic is not a list of thirty rules. It is hundreds of contextual judgments that never make it into a workshop because nobody thinks to say them out loud — they're "just how we do it." The brand you never mark down before week ten because its price image is too fragile. The vendor who quietly takes back excess if you don't burn the relationship by clearing too early.

The cluster where a 35% margin floor is real and the one where it's negotiable. The category whose role is traffic, not margin, so you'd rather run thin than discount. As we've written, the vast majority of a retailer's business rules live in people's heads, never formalized and never executable by a system.

The model never absorbs any of that, because the people who hold it treated the project as someone else's and gave it two hours. So the system ships, starts producing recommendations, and the merch team opens them — and discards most of them. Roughly 70% get ignored, not because the math is wrong but because each rejected recommendation quietly violates a rule the model was never told about.

From IT's side this looks like an adoption problem, or worse, like merch resistance to change. From merch's side it looks like proof the system "doesn't understand the business." Both readings miss the cause.

The system doesn't understand the business because the only people who could have taught it stood at arm's length and called it IT's project. The 70%-ignored outcome isn't a model defect. It's the predictable result of the decision's owner declining to own the system that makes the decision.

And the cost compounds. Teams revert to their spreadsheets, the platform becomes shelfware, the CIO's investment gets quietly questioned. And the next time a decision tool is proposed, merch is even more convinced these projects don't work. The cycle hardens the very belief that caused the failure.

Ownership means encoding merch judgment, not rubber-stamping a rollout

So what does owning it actually look like? It is not chairing the steering committee. It is not validating the UI. It is not being the executive who says yes at the gate. Those are the trappings of involvement, and they're exactly what produces the 70% failure while everyone feels consulted.

Real ownership means treating the merch team's judgment as a first-class input to the system — as load-bearing as the demand forecast, and given the same engineering seriousness. Concretely, three things have to be true.

The unwritten rules get written down — deliberately, and over time. Not extracted in one workshop and frozen. The merch director has to drive a sustained effort to surface the contextual logic the team applies daily. That means the brand calendars, the vendor return windows, the cluster-specific floors, the category roles, the "we never do X before Y" reflexes.

This is merch work, not IT work. An engineer cannot elicit a rule nobody knows they're applying. Only the person who applies it can.

Those rules go into the system as living, owned objects — not buried in code. A rule encoded once in a script and frozen is worse than no rule, because it silently drifts from reality the first time a vendor contract changes. Ownership means the merch team can see which rules are active, and which ones a given recommendation respected or violated. They can change a rule when the business changes, without filing an IT ticket and waiting a release cycle. If your buyers can't read and edit the logic, you don't own it; IT does.

Someone in merch is explicitly accountable for that logic staying alive. Not the CIO, not a consultant, not "the project." A named role inside the merch organization whose job is to keep the encoded judgment current as brands, vendors, and strategy shift. Without that, even a perfectly built system drifts back into the spreadsheet within two seasons.

This is the difference between a merch director who owns the system and one who approves it. The owner shapes what the system knows. The approver inherits what someone else decided it should know — and then spends the next year explaining why the team won't use it.

Reframing the fear: AI executes your buyers' judgment at scale

Underneath the "it's IT's project" reflex there is usually a quieter, less spoken fear: that the real point of all this is to replace buyers with an algorithm. If the model decides what to buy and how to mark it down, what's left for the merchant? It's a reasonable thing to be wary of, and it's worth naming directly rather than letting it steer behavior from the shadows.

But the fear inverts what's actually happening. A decision platform doesn't replace the buyer's judgment — it has no judgment of its own to substitute. It can only execute the logic it's given. Left to pure statistical optimization with no merch rules encoded, it produces exactly the recommendations your team already rejects 70% of the time.

The model is not a rival merchant. It is a mechanism that applies merch judgment at a scale and frequency no human can reach by hand.

Consider what a buyer's judgment can actually cover manually. A great merchant can hold maybe a few hundred SKUs in active, considered attention — the strategic styles, the risky bets, the heroes of the season. Below that line, thousands of long-tail SKU/store decisions get made on rough heuristics or not made at all, because there isn't enough human time. That long tail is where margin quietly leaks: the slow seller marked down a week too late in forty stores, the size break unaddressed, the transfer never made.

Encode the buyer's judgment into the system and that judgment now reaches the entire tail. The brand calendar the merch director enforces by memory on the top forty styles gets enforced on all four thousand. The markdown discipline a great buyer applies to hero products gets applied to the whole assortment, consistently, every week, without fatigue. The buyer's expertise stops being a scarce resource rationed across the strategic top of the range and becomes the operating logic of the entire catalogue.

That is leverage, not replacement. The buyer's time stops being spent re-deciding the same thousand routine calls every Monday. It redirects to the work only a human can do: reading a new trend early, negotiating the vendor relationship, making the genuinely ambiguous strategic bet.

The system handles the volume the merchant defined; the merchant handles the judgment the system can't originate. A buyer afraid of being replaced should be far more afraid of the alternative. A system built by IT executes nobody's judgment and burns trust across the whole organization.

The question worth asking before the next kickoff

When the next decision-platform initiative lands on the roadmap, the merch director has one decision to make before any model is trained. And it determines everything that follows: is this our project, or are we waiting to be consulted on it?

This is precisely where Solya starts from a different premise. The platform is built so that the chain's merch logic lives in the decision engine as a first-class, owned input: vendor terms, brand calendars, cluster floors, category roles. It is declared and edited by the merch team rather than buried in code. Buying, allocation, and markdown decisions are then generated to respect that logic by construction, so the merchant's judgment is what scales — not a statistical optimum that ignores it.

If the answer to the question is "we're waiting to be consulted," the 70%-ignored outcome is already written. If the answer is "this is ours," the merch director is no longer protecting their craft from automation. They're arming it.

For the adoption and override questions buying teams face during rollout, see the Merchandising Director FAQ.

---

Is your team about to own its decision AI — or inherit it?

At Solya, we offer merchandising and buying leaders a personalized 30-minute diagnostic. It assesses how much of your team's real decision logic currently lives in heads versus in your systems: vendor terms, brand calendars, cluster floors, category roles. And what it would take to make that judgment the operating logic of your buying, allocation, and markdown decisions.

You'll walk away with:

• A read on how much of your merch judgment your current tools actually encode
• An estimate of the margin leaking in the long tail your buyers can't reach by hand
• The first decisions to bring under owned, merch-defined automation — and the order to do it in