Retail decisioning glossary: 15 terms that matter

Most retail vocabulary describes the past. Sales, stock, margin, sell-through — these are nouns for things that already happened. Decisioning is the vocabulary of what to do next, and most retail teams don't share it precisely.

That gap has a cost. When merchandising, supply chain, and data each use "recommendation" or "rule" to mean different things, the output of an expensive model stalls. In our field work, the share of system-generated recommendations a chain ignores climbs toward 80% when the terms and rules aren't shared across teams. A common definition is not pedantry. It is the precondition for acting on what your stack already knows.

This glossary defines the 15 terms that matter. Each entry takes a position — Solya's framing — and names what the term is most often confused with. The big terms link out to their full treatment.

Core concepts

Decision layer. The part of a retail stack that turns data and predictions into specific, executed decisions, under the chain's own business rules. It sits above data and modelling, below execution — an architectural position, not a feature you bolt on. It is often confused with a smarter dashboard or a better forecast, and it is neither. See what a decision layer is in retail.

Decision intelligence. The discipline of engineering how decisions get made, executed, and improved — not how data gets displayed. Business intelligence answers "what happened?"; decision intelligence answers "what should we do, and did it work?". The two are routinely conflated because both start with data — but the output differs: a chart versus an executed action. See decision intelligence vs business intelligence.

Operational AI. AI embedded in the day-to-day operating cadence of the business, producing decisions that reach the floor — not AI that lives in a notebook or a quarterly model review. The test is simple. If the output never changes a price, a transfer, or an order, it is not operational. See what operational AI is in retail.

Decisioning. The act of arbitrating between options with different costs and risks, then committing to one. A forecast is an input to decisioning; it is not decisioning itself. The term is often used loosely for any automated step. Decisioning is specifically the arbitration, not the data feed before it or the execution after it.

These four sit at the centre because they are the most abused. A platform that does BI calls itself decision intelligence. A forecasting engine calls its output decisions. Naming the boundaries is what separates a real decision layer from a relabelled dashboard — a distinction we draw in decision platform vs forecasting vs BI.

Inventory decisions

Replenishment. The decision of what to reorder, how much, and when, to keep the right stock in the right place. It is a recurring arbitration against demand, lead time, and shelf capacity — not a fixed reorder point. It is confused with forecasting; the forecast feeds it, but the reorder quantity is the decision. See continuous replenishment vs the weekly meeting.

Allocation. The decision of how to split a finite quantity of a SKU across stores or channels, weighing each location's demand, stock, and selling profile. It is not a forecast and not a simple even split. Good allocation is the difference between a product that sells through everywhere and one that piles up in the wrong stores. See the retail allocation engine.

Assortment. The set of products a retailer chooses to carry, for a given store cluster and season. Assortment is a decision about breadth and depth, made before the season starts, under range and budget constraints. It is often confused with the catalogue — the catalogue is everything available; the assortment is the deliberate selection. See AI assortment planning before the season.

Overstock. Inventory held above what demand will absorb at full price before it loses value. Overstock is not just "too much stock" — it is stock whose carrying cost and markdown risk now exceed its expected margin. It is the silent twin of the stock-out, and ERPs rarely flag it until it is expensive. See the real cost of overstock.

Stock-out (OOS). The state where demand exists but the product is not available to buy, in store or online. A stock-out is a lost sale plus a customer who may not return. It is routinely under-measured because the demand it kills leaves no trace in the sales data. See the real cost of stockouts.

Replenishment, allocation, and assortment are the three core inventory decisions. Each is an arbitration, not a report. Overstock and stock-out are the two failure states they exist to prevent — symmetrical errors, both expensive, and both invisible to a stack that only describes.

Pricing and markdown

Markdown. A permanent price reduction taken to clear stock that will not sell through at full price. Markdown is a margin-versus-sell-through arbitration, made per SKU, per store, per week. It is confused with promotion — a markdown is structural and usually irreversible; a promotion is tactical and time-boxed. Many chains still take 70% of markdown decisions by hand.

Sell-through. The percentage of received stock sold over a period, at full price or in total. It is the single most important signal of whether a buying and pricing plan is working. Sell-through is a measurement, not a decision — but every markdown, transfer, and reorder decision reads it as an input. Low sell-through is what triggers a markdown; high sell-through is what triggers a reorder.

Dynamic pricing. Adjusting price continuously in response to demand, stock, competition, and elasticity, rather than on a fixed calendar. Done well, it is a stream of pricing decisions, each justified by its own conditions. It is often reduced to a rule engine — "if stock > X, cut price 10%" — which is automation, not decisioning. See dynamic pricing: rules vs decisions.

Promo lift. The incremental sales a promotion generates above the baseline, net of what would have sold anyway. Promo lift is a decision input, not a marketing vanity metric, and the hard part is isolating true incrementality from cannibalisation and pull-forward. A promotion with high gross sales and zero net lift destroyed margin. See promo lift is a decision, not a marketing number.

Pricing decisions are where description and decisioning diverge most sharply. Sell-through tells you what happened; markdown, dynamic pricing, and promo decisions are what you do about it. A stack that surfaces sell-through but leaves the pricing arbitration to a Friday spreadsheet has stopped exactly one step short of the decision.

Forecasting and AI

Demand sensing. Short-horizon demand estimation that reacts to recent, granular signals — point-of-sale velocity, weather, local events — rather than long-range seasonality alone. It sharpens the input to replenishment and allocation decisions. Demand sensing is still a forecast, not a decision; its value is realised only when a decision consumes it. See demand sensing and decision readiness.

Business rules. The hard constraints a retailer imposes on every decision — margin floors, supplier minimums, commercial calendars, brand handling, cluster logic. They are not a filter bolted on after a model runs; in a real decision layer they sit at the core, so a recommendation that breaches a floor is noise. Yet roughly 80% of retail business rules are misused — buried, ignored, or applied too late. See why 80% of retail business rules are misused.

Decision-to-execution loop. The closed cycle where a decision is made, executed in the operational systems, measured by its effect, and fed back to improve the next decision. A decision that cannot reach the ERP, WMS, or pricing system without manual re-entry is an opinion, not a decision. The loop is what separates a learning system from a frozen rule engine. See the closed decision-to-execution loop.

Adoption rate. The share of system-generated decisions that operators actually execute rather than override or ignore. Adoption is the truest measure of a decisioning system's value — a model with 95% accuracy and 20% adoption changes nothing. It is the metric most vendors avoid and most buyers should demand. See why the merch director owns retail AI adoption.

These four terms describe the machinery and the proof. Demand sensing and business rules are inputs; the decision-to-execution loop is the mechanism; adoption rate is the verdict. A high adoption rate is the only evidence that the vocabulary above became practice — that decisions were made, trusted, and executed, not just displayed.

Why the shared vocabulary pays off

A glossary looks like housekeeping. It is closer to plumbing. When merchandising, supply chain, and the data team mean the same thing by "decision", "rule", and "recommendation", the system's output stops getting lost in translation between functions.

The terms here are not interchangeable, and the confusions are not harmless. Calling a forecast a decision is why ML investments stall. Calling a dashboard decision intelligence is why KPIs stay flat. Naming the gap precisely is the first decision — the one that makes every later decision executable.

---

Want to map your decisioning vocabulary to your stack?

At Solya, we offer retail data and operations leaders a personalized 30-minute diagnostic. We map where your decisions are actually made today, and which of these terms describe a real capability versus a relabelled one. You'll walk away with:

• A clear read on where your stack stops at description and where it actually decides
• The terms in this glossary mapped to your real components, not vendor slides
• The first high-ROI decisions to bring under a real decision layer