Most retailers are building next season’s assortment for a customer who has already left the building.
The mechanic is well known and rarely challenged. Legacy assortment planning starts with last year. Planners copy the prior range, adjust for known wins and losses, then apply a growth number. It is fast, familiar, and blind to anything that changed.
That last word is the problem. Blind.
The shopper who bought your category last spring is not the same person you are buying for now. They traded down or traded up. They switched channels. Some of them are on GLP-1 medications and eating differently. Some found a brand on their feed three weeks ago that did not exist in your historical file. A plan built on a description of who they used to be is a confident answer to a question nobody is asking anymore.
The seasonal forecast is the wrong unit of decision
The industry knows this. The vendors know this. Uniform distribution strategies, static option counts, and disconnected pre-season and in-season workflows cannot keep pace with localized demand, shifting trends, and continuous volatility.
The deeper issue is that we still treat assortment as a once-a-season bet placed months before the season opens. Seasonal forecasts assume stable demand, but today’s retail environment is driven by real-time signals like social media trends, promotions, and rapid shifts in consumer behavior.
Stable demand is a fiction we buy against because the planning calendar rewards it. It is easier to lock a number in February than to defend a range that keeps moving. But easy is not the same as right, and the market does not grade on a curve.
Here is the honest read on the better approach. The process runs as a loop, not a once-a-season pass. The model learns what sold where, solves for the best mix inside real limits, and every cycle starts from what the last one taught it.
Loop, not pass. That single reframe is worth more than any feature list on a vendor deck.
Live signal is the fix and also the trap
So the instinct is to chase the freshest signal you can find. Wire in social trends, search spikes, real-time sell-through, sentiment. Inject the now.
Good instinct. Bad execution, most of the time.
Feed raw live signal straight into the buy and you have not modernized merchandising. You have built a slot machine. You over-fit a two-week TikTok spike, you whipsaw replenishment, you torch your open-to-buy chasing a trend that peaked before your PO cleared. Noise dressed as insight is more dangerous than a stale plan, because it feels like progress.
The failure is not too much signal or too little. It is placement. Where in the decision does each type of data belong?
Structure comes from history. Category role, store clusters, size curves, margin guardrails, the mission each location actually serves. That is exactly the reconciliation work happening in the market right now: retailers need to define each location’s mission clearly and align assortment. History is good at structure because structure changes slowly.
Deltas come from live signal. Inside that fixed structure, live signal answers the fast questions. Which trend expresses deeper in which cluster. Which store is moving early and deserves an in-season chase. What breadth to hold in reserve because you cannot yet see where it lands.
And the whole thing needs a governor. AI breaks forecasts into reusable components, evaluates tradeoffs in real time, and quantifies the financial impact of acting or not acting so decisions align with margin and service goals. That quantification is the part most teams skip, and it is the part that keeps live signal from becoming a casino.
Put the Cost of Doing Nothing on the board
This is where CODN earns its keep. A stale assortment does not fail loudly on buy day. Nobody gets fired in February. It bleeds quietly across the season through markdowns, dead SKUs, and demand you never captured because the range did not reflect who was actually walking in.
Those numbers are real and large. In North America, understocking alone runs into double digits as a share of inventory value, and mid-tier operators are getting squeezed while value and premium players expand. The buy you copy-pasted with a growth number is not free. It carries a price you pay in Q3, and almost no planning team writes that price on the whiteboard before they lock the buy.
One more discipline. Do not let the machine run alone. The mature pattern is a human seam: most enterprise deployments run a hybrid model. The system proposes the optimized range, merchants approve or override any algorithmic suggestion. Signal proposes. Merchants dispose. Guardrails hold the line.
The retailers who win the next two seasons will not be the ones with the most data or the flashiest live feed. They will be the ones with the cleanest seam between what history knows and what the market is telling them at 9 a.m. today. Structure from the past. Deltas from the present. Budget held back for the future you cannot yet see.
Stop buying for last year’s shopper. That customer already changed. The only question left is whether your plan noticed.