The blind window

The bin is an autopsy

Every fresh operator I have worked with can tell me what their waste number was last week. Very few can tell me which decision produced it.

That is not a reporting failure. It is a structural one. The waste number is generated at the end of a product's life, which means it is generated after every choice that determined it has already been made and closed. By the time a tray of chicken thighs is marked down and then written off, the order that created it was placed four days ago, the forecast it was ordered against was run five days ago, and the replenishment parameter that shaped the forecast was last reviewed some time in the previous financial year.

The bin is an autopsy. It tells you the patient died. It does not tell you which of the last six decisions killed them.

This matters more than it sounds, because almost every improvement programme in fresh is built around the autopsy. Waste reports, shrink reviews, category deep-dives, root-cause meetings: all of them start from the outcome and work backwards. Working backwards from an outcome is a reasonable way to allocate blame. It is a poor way to change behaviour, because the behaviour you need to change happened in a different week, in a different system, by a person who was not in the room.

The blind window

There is a more useful place to look, and it sits at the front of the process rather than the back.

For any fresh line, there is a moment when the order becomes irreversible. Cut-off passes, the order consolidates, the supplier picks, and from that point nothing you learn can change the quantity arriving. There is also a moment when demand becomes knowable: when enough of the selling period has happened that you could say with confidence what the week actually wanted.

The second moment almost always comes after the first. The gap between them is the blind window.

Everything that goes wrong in fresh goes wrong inside that gap. You committed a quantity, then the weather turned, then a competitor promoted, then the school holiday started a week earlier than the calendar in your system, then nothing happened at all and the number you committed was simply too high. The blind window is the period during which you are exposed to being wrong and cannot do anything about it.

It is worth calculating properly, line by line, because the number is usually larger than people expect:

• Order lead time. From cut-off to arrival at store. Include the consolidation step, which people routinely forget.
• Signal latency. From a sale happening to that sale appearing in the data the ordering decision uses. In most mid-market operations this is between one and three days, and in more than a few it is a week, because the ordering system reads a batch file rather than the till.
• Decision cadence. How often the order can be adjusted. A line ordered twice a week has a structurally longer window than the same line ordered daily, whatever the lead time says.

Add them together and compare the total against the product's saleable life. When the window is longer than the shelf life, you are not running a planning process. You are placing a bet and waiting to see.

Why the weekly review cannot see it

The standard fresh waste review is weekly, by store, by category, against a percentage target. It is a good instrument for one job and a poor instrument for another.

What it does well is tell you where the money went. Store 412 is running at nearly double the estate average on bakery. Chilled ready meals deteriorated sharply in the last fortnight. Those are real facts and worth knowing.

What it cannot do is tell you whether any of it was avoidable. Two stores with identical waste percentages can be in completely different situations. One of them ordered against a signal that was clearly visible three days before cut-off and ignored it. The other ordered correctly against everything knowable at the time and got caught by a demand shift that nobody could have seen. The first is a process failure you can fix. The second is the cost of operating in a category with a clock on it.

The review shows both as the same number, and so the improvement effort gets spread evenly across two problems, one of which cannot be solved at all.

There is a second distortion worth naming. Waste and availability are the same dial, and most organisations measure them in separate meetings, owned by separate people, against separate targets. Supply chain is accountable for waste. Trading is accountable for availability. Both can hit their number in the same period while the business loses money, because the only reliable way to move one is to concede ground on the other, and nobody in either meeting is asked about the trade.

Two failures that look identical

Once you have the window in view, fresh waste separates cleanly into two kinds, and they need opposite responses.

Forecast error. The signal was not there. Demand moved for a reason that was not knowable before cut-off, or was knowable only to someone who was not consulted. The order was a reasonable decision that turned out badly.

Decision error. The signal was there and the process could not act on it. The data arrived after cut-off. The person who saw it had no authority to change the order. The system allowed an override but the override took forty minutes of manual work per line and nobody had forty minutes. The parameter that would have prevented it was set three years ago by someone who has left.

These produce the same entry in the waste report and require entirely different investment. Forecast error is fixed with better models, richer signals, more granular data. That work is expensive, slow to show results, and frequently the right thing to do.

Decision error is fixed by changing when decisions happen and who is allowed to make them. It is usually cheaper, faster and less glamorous, and it is where most mid-market fresh operations are actually losing money.

The uncomfortable part is that the industry's spending is weighted heavily toward the first. Forecasting is a product category with vendors, demos and a procurement process. Shortening a decision cycle is an internal argument about process ownership, and nobody has a stand at the trade show selling it.

The forecast got better. It still arrived after the order went out.

Narrowing the window

The window can be shortened at four points. They are listed here in ascending order of cost, which is the opposite of the order most programmes attempt them in.

Move the cut-off. The single cheapest intervention available, and the most commonly overlooked. Many fresh cut-offs are set where they are because of a warehouse constraint that stopped being true years ago, or because a supplier asked for margin at the start of a contract that has since been renegotiated twice. Pushing cut-off later by four hours can take a full day out of the window on a fast line. It costs a conversation.

Shorten signal latency. Getting yesterday's sales into the ordering decision before the decision is made, rather than in the weekly pack afterwards. This is an integration job rather than a modelling one, and the work is usually smaller than it looks because the data already exists in the estate. It just arrives on the wrong schedule for the person who needs it.

Increase decision frequency. Moving a line from twice-weekly to daily ordering halves its exposure. The constraint is rarely the system. It is delivery economics, picking capacity and the supplier agreement, which is why this gets decided commercially rather than operationally and often never gets raised at all.

Improve the forecast. Genuinely valuable, genuinely expensive, and the last thing to do rather than the first. A better forecast pays back properly once the window is short enough for the improvement to reach a decision that is still open. Do it before that and you are refining an input to a choice that has already been made.

The pattern is consistent: the interventions that cost least are structural and unglamorous, and the one that costs most is the one with a budget line already attached to it.

What to measure instead

Total waste as a percentage of sales is a scoreboard. It is not a diagnostic, and treating it as one is why so many fresh improvement programmes stall after the easy gains.

Four measures that actually move something:

• Window length by line. Lead time plus signal latency plus decision interval, compared against saleable life. Sort the range by the ratio. The lines where the window exceeds shelf life are where your avoidable waste is concentrated, and the list is almost always shorter than people fear.
• Avoidable share. Of last period's waste, how much was committed against a signal that was already visible at cut-off? This requires reconstructing what was knowable at the moment of the decision, which is more work than pulling a report and is the only number here that changes what anyone does on Monday.
• Override rate and outcome. How often does a human change the system's suggested order, and are they right when they do? A low override rate in a category this volatile usually means the process makes overriding too hard, not that the system is trusted. A high rate with poor outcomes means the opposite problem.
• Paired waste and availability. Always together, never separately, at the same level of granularity, in the same meeting. If a category improved waste by a point while availability fell by two, nothing was gained and something was quietly lost.

None of these require new software. They require a week of honest reconstruction and a willingness to find out that the answer is unflattering.

Where this leads

The argument is narrow and I want to keep it that way. Fresh waste is not primarily a forecasting problem, and it is not primarily a discipline problem. It is a timing problem. Orders are committed before demand is knowable, and the distance between those two moments determines how much you lose. Shorten the distance and the same people, the same system and the same forecast produce a different result.

That distance is measurable. Most businesses have never measured it.