Merchandising Director FAQ — retail AI decision platform

Treat the AI as a junior buyer producing a first-pass proposal. Senior buyers review, override and curate. Their overrides feed back into the model as supervision signal, so the system learns merchant intent over time. The role shifts from "produce the buy" to "validate and curate the buy", which frees roughly 30-40% of buyer time for higher-value work like vendor negotiation and tail-category curation. The AI handles volume and pattern recognition. The buyer handles judgment, trade-offs and the calls the model cannot see.

The cycle compresses by two to four weeks. AI surfaces demand signals on volume and assortment four to six weeks earlier than spreadsheet-driven planning, because it ingests sell-through, search and competitor signals continuously. Buyers walk into the buy meeting with a draft assortment instead of building it during the meeting. Late-cycle changes drop sharply, and the team has more room to refine the tail. The cadence also stabilises: weekly review loops replace the quarterly fire drill that most pre-season planning currently runs on.

Two design choices are non-negotiable. First, the AI must explain its recommendation in business terms, not model internals. A buyer needs to read "pulled forward by warm-weather signal in cluster B", not feature-importance bars. Second, every recommendation must carry a one-click override, and the model has to learn from that override. Without both, buyers either disengage or game the system. Get them right and judgment compounds: the model encodes merchant intuition while still doing the heavy pattern lifting. See our piece on retail business rules for the rule-design angle.

Months zero to three sit at 30-50% adoption as buyers experiment in parallel with their existing spreadsheets. Months three to nine usually dip before recovering, because the honeymoon ends and the model has to prove itself on the team's hard cases. Steady state from month nine sits at 70-85% for well-implemented rollouts. Anything under 50% at month nine signals the model is ignoring buyer constraints, and the fix is rule and feedback design, not more training data. Measure adoption per buyer, not per team average.

Split overrides into two tiers. Line-item overrides on a single SKU or store stay free: the buyer wins, the model records the delta and learns. Strategic overrides that shift category trend, brand mix or pricing posture require sign-off from the category lead and feed back as model parameters, not as one-shot corrections. This keeps individual judgment fluid while protecting the model from drifting on a few loud opinions. Audit overrides monthly: a buyer overriding 70% of recommendations is a model bug, not a discipline problem.

Resistance comes in three flavours. First: the AI does not understand my constraints. Fix it by adding those constraints as explicit rules, not by arguing. Second: the AI is worse than my judgment. Fix it by measuring both on the 80% volume base where the model is meant to win, not on the 20% edge cases where the buyer wins. Third: this is here to replace me. Fix it by promoting the first cohort of AI-fluent buyers visibly within twelve months. Each flavour has a structural fix, not a training fix.

Three metrics matter at the category level. Sell-through rate typically moves 200-500 basis points on AI-driven assortments versus control. Markdown depth drops 100-300 basis points because pre-season buys land closer to actual demand. Assortment productivity, measured as revenue per SKU at the same margin, rises 5-15% as the long tail gets tighter curation. Track all three against a held-out control category for the first two seasons. Vanity metrics like forecast accuracy do not belong on a merch dashboard.

The biggest impact lands on tail-SKU decisions. AI surfaces signals on slow-moving and new-to-assortment SKUs six weeks before the buy meeting, where spreadsheet planning typically reads those signals three weeks after commitment. That window is what unlocks tighter curation without hurting breadth. Core SKUs benefit less, because their patterns are stable and any decent planner already gets them right. Plan the AI to earn its keep on the tail, where humans run out of attention. See our pre-season AI assortment planning piece for the full workflow.

AI-fluent category managers handle 30-50% more SKU breadth at the same quality bar. That puts them on a faster trajectory to senior buyer and category director roles, because span of control is the gating factor most companies use for promotion. Refusers stagnate: they end up running smaller and smaller scopes as their peers absorb broader categories. Frame the rollout as career investment, not as workflow change. The buyers who own the AI early are the ones running the function in five years.