← All food signals
Retail

The Overridden Forecast

AI ordering for fresh food is spreading across grocery chains, but store managers often override it, scored on full shelves rather than waste.

What is it?

AI ordering systems propose how much fresh food each store should buy, sized to expected demand. The store manager still places the order. ReFED finds managers often override the smaller order, because their scorecard rewards a full shelf over a lighter bin.

Why now?

Chains are moving AI ordering from pilots into whole estates, and Grocery Outlet’s rollout reaches about 550 stores. The waste savings on offer are still the vendors’ own numbers. A 2026 ReFED report puts the brake on the store floor, where in-stock audits outrank waste in every manager’s review.

Who’s building it?

Ordering-software vendors supply the model; chains and store operators decide how much of it to obey.

Ordering-software vendorsGrocery chainsStore operatorsFood-waste researchersOperations scholars
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — AI ordering vendors sizing each store’s fresh order.
Buyers — grocery chains and independent store operators rolling the system out across departments.
Gatekeepers — store managers who accept, trim or override each order.
What’s at stake
Whether the waste AI ordering saves is real, when the only published figures come from the companies selling it.
Surplus quietly written off each week as routine shrinkage.
In-stock audits that mark down any manager who runs short.
What’s unsettled
Whether any independent study will measure AI ordering’s effect on fresh waste.
Whether chains will rewrite store scorecards to count waste beside availability.
Whether a manager’s override adds local judgement or simply restores the old full-shelf habit.
02 · How the demand matureda progression, not a timeline
Pilot
A chain trials AI ordering in a few fresh departments and reports shrink.
Rollout
The system spreads across whole estates, independent operators and every department.
Adherence
Vendors start reporting how often stores follow the advice, making each override visible to head office.
Scorecard
Whether store metrics start to count waste decides how far the model’s order gets.
03 · Where the market goesthe forks, not a prediction
→Scorecard or shelf: chains may add waste to store reviews so the model’s order survives, or keep in-stock audits and live with overrides.
→Independent studies may confirm the vendor savings in fresh food, or find them far smaller.
→Adherence rates may become the headline metric, turning each manager’s veto into a performance issue.
→Human adjustments may prove worth keeping for some products, even where the model is accurate.
The model can size a better order already; whether it gets placed depends on a store scorecard that was written long before the software arrived.
04 · The Resilience Tech deskadjacent markets, moving now
Proving You Actually Wore It
Returns get screened before the refund.
A returns firm gives each returned item an AI fraud risk score.
CBS News
Robot Gridlock
AI planning moves deeper into the stockroom’s daily flow.
AI planning gave about 25% higher robot throughput in simulation.
MIT News
The Machine Markdown
Stock nearing expiry gets marked down by software.
New Zealand grocers use WhyWaste software to discount stock nearing expiry.
Newsroom (NZ)
05 · The Wire
01
ReFED finds managers frequently override AI waste-cutting orders.
ReFED ↗
02
Grocery Outlet to use Afresh ordering across about 550 stores
Grocery Dive ↗
03
Letting merchants adjust an inventory algorithm’s forecast input raised profit by 4.92% on average versus automation alone.
Management Science (via RePEc) ↗
The panel on this signal
They agreed on
–Published waste savings from AI ordering in fresh food are still the vendors’ own, with no independent measure yet.
–Store scorecards reward full shelves over less waste.
–The override happens at the store, whatever the model’s accuracy, so the brake is organisational.
The demand-planning practitioner disagreed on
–An override is not always bias: planners see local events the history misses, and adjustments can raise profit.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.