Ask a planner what demand looked like last season and they will pull up the sales report. It is the most natural thing in the world, and it is the single most expensive mistake in the plan. Sales are not demand. Sales are the intersection of demand and whatever you happened to have on the shelf.
When a style sells out, the sales curve does not go up and then stop because customers lost interest. It stops because you ran out. The demand kept going. You just stopped being able to see it, because a customer who wanted a medium and found none does not leave a receipt. They leave nothing. And nothing looks, in the data, exactly like an absence of demand.
This is called censoring, and it has a direction. It never makes a style look more popular than it was. It only ever makes your best sellers look smaller, because the best sellers are the ones that stock out. So you systematically under-read your winners, buy them too shallow next time, stock them out again, and confirm the very number that was wrong in the first place.
The demand you cannot see is exactly the demand you most need
Stockouts erase the signal from the styles that were working.
Picture a hero style that sold 800 units and hit zero on hand in week four of a twelve-week season. The sales report says 800. But the sell-through was accelerating when it flatlined, not decelerating. The truthful read is not 800 units of demand. It is 800 units of supply meeting something closer to 1,400 units of want, with 600 customers who tried to buy and could not.
Now do what most plans do: take 800 as the demand number, apply your growth factor, and buy roughly 850 next time. You have just under-bought the same style by 40 percent again, and it will stock out again, this time maybe in week three. The censored number does not just mislead once. It compounds, season after season, and the style that should have been your franchise stays capped by a supply decision you made and then forgot you made.
The over-bought styles do the opposite and are just as misleading. They never stock out, so their full demand is visible, all the way down to the units you had to discount to move. Their sales history looks complete and healthy. So the styles you should buy less of look trustworthy, and the styles you should buy more of look weak. The data is pointing you in exactly the wrong direction.
This asymmetry is what makes censoring so dangerous compared with ordinary forecast error. Random error washes out: some styles you over-read, some you under-read, and across a big assortment it roughly cancels. Censoring does not cancel, because it only ever cuts one way. Every stockout removes demand you would have seen, and never adds demand you would not have. So the bias accumulates in a single direction, concentrated on your best products, and no amount of averaging across the assortment saves you from it. The more successful a style, the more its truth gets hidden.
the full-price sell-through gap between category leaders (71%) and everyone else (57%). Reading true demand instead of censored sales is a large part of how leaders find that margin.
That gap is not mostly about better taste. It is about buying to real demand on the styles that work, so they stay in stock and keep selling at full price, instead of capping them at last year's censored number and watching them stock out into a lost sale.
Reconstructing demand is a modeling job, not a spreadsheet job
You have to estimate what the stockout hid, and you cannot eyeball it.
The fix is to stop treating the sales report as the demand report and start reconstructing what demand actually was, including the part the stockout erased. That is not something you do by hand. It requires modeling the sell-through curve, knowing when each SKU went to zero on hand at each door, and estimating the demand that would have continued if the units had been there.
Tightly's platform does exactly this: it uncensors the signal by looking at the shape of the sell-through before the stockout, the pace at comparable doors that did not run out, and the behavior of similar styles, then reconstructs the demand curve that the availability gap hid. The forecast is then built on demand, not on the truncated sales that pass for demand in a spreadsheet.
It works the other side too. Units that only moved because you cut the price are not clean demand at full price, and the model separates the two. So a style is not credited with strength it only bought with markdown, and it is not penalized for a stockout it never chose. Both distortions come out, and what is left is a read on what customers actually wanted.
The stockout gap sits entirely on your winners
Sales cap at what was on the shelf. True demand kept climbing. The difference is the lost full-price sale you never saw.
This matters most for DTC brands specifically, because DTC amplifies the censoring in ways wholesale does not. A wholesale buyer sees a purchase order and moves on; the sell-through happens at someone else's till. A DTC brand watches demand in real time, hour by hour, on its own site, which sounds like an advantage and becomes a trap. When a size sells out online, the product page often just hides that size or greys it out, so the demand that arrives after the stockout is not merely uncounted, it is actively suppressed at the source. The customer never even gets to add it to a cart. Your own store is deleting the evidence, and doing it faster than any wholesale channel ever could.
The other DTC-specific wrinkle is returns, which are a second layer of distortion sitting on top of the first. Gross sales overstate demand where returns are high, and censored sales understate it where stockouts hit. Read the raw report and both errors are baked in at once. A serious demand read has to strip both: reconstruct the units the stockout hid, and net out the units that came back. Only then are you looking at what customers actually wanted and actually kept, which is the only number worth planning against.
Buy against what customers wanted, not what you had
Once the signal is uncensored, the winners stop being starved.
When the forecast is built on reconstructed demand, the whole shape of the buy changes. The hero styles get depth that reflects what customers actually wanted, so they stay in stock through the season and keep earning at full price. The styles that only ever moved on discount stop getting rewarded for it. And new launches, which have no sales history at all, cold-start from the demand profiles of their nearest historical cluster rather than inheriting a censored average.
The re-forecasting has to be live for any of this to hold. Demand is not a fact you establish once at the start of the season; it moves every day as new sell-through, returns and channel mix arrive. A forecast that reconstructs true demand in January and then sits still is only right for about a week. Tightly's platform re-runs the demand read hourly, so the reconstruction stays current and the winners get chased while there is still season left to sell into, not flagged at the end when the only lever left is markdown.
A stockout is not the end of demand. It is the end of your ability to see it, and the two are very easy to confuse.
The brands that get this right stop being surprised by their own best sellers. They know which styles were capped by supply, not appetite, and they buy them to the demand instead of to the receipt. That is not optimism. It is finally reading the signal the shelf was hiding all along.