Twice a year, the range review at Larsen & Holt - an 85-store convenience grocery chain - arrived at the same argument, and twice a year the argument ended the same way. The deli counter lost money on paper: heavy labour, punishing waste, a margin line that made the finance director wince. And twice a year the category director defended it with the sentence that has protected weak categories in every grocery business since the invention of the trading meeting: it drives footfall.

Maybe it did. Nobody actually knew. The claim was, strictly speaking, a testable one - do deli buyers cross-shop, what do their baskets carry beyond the counter, and does the answer hold in a suburban 400-square-metre store the way it might in the two city-centre flagships? But testing it meant joining POS transactions to loyalty identities to labour scheduling to waste logs, four systems with four owners, and the data team’s estimate for that project was six weeks. Six weeks was longer than anyone’s patience and shorter than nobody’s backlog: the dashboard queue already ran four months, and category managers had learned to stop asking. So the deli survived on folklore, the review moved to the next line, and the decision was deferred - in effect, made - for another two quarters of labour cost and shrink.

That deferral has a market price, and the industry has been paying it at scale for as long as anyone has measured. IHL Group, which has tracked what it calls inventory distortion for eighteen years, puts the global cost of stocking the wrong things - out-of-stocks plus overstocks - at .7 trillion a year, roughly 6.5 percent of retail sales. For a banner Larsen & Holt’s size, industry-average distortion implies a leak north of €25 million a year, distributed across thousands of SKU-store combinations too small to see individually and too numerous to survive aggregation into a weekly report. The whole problem, top to bottom, is a visibility problem. The folklore fills the space where the data should be.

Out-of-stocksshoppers came to buy, left empty-handed .2T

Overstocksmarkdowns, spoilage, tied-up capital $554B

Scaled down: a €400M grocery banner at industry-average distortion is leaking roughly €26M a year - most of it invisible in weekly reports.

Source: IHL Group, “Fixing Inventory Distortion” (2024) and 2025 update; retailers using AI/ML show 2.3× sales and 2.5× profit growth vs. peers.

The trading meeting, replayed

The spring review went differently, because by then the Analytics Engine sat over all four systems - read-only, nothing migrated - and the CEO had developed a habit that mildly terrorised the room: asking the testable question and waiting.

> Of loyalty customers who bought from the deli counter in the last six months, what share of their spend is outside the deli, and how does their basket compare to matched non-deli shoppers - by store?

The answer took under a minute, and it embarrassed both sides of the argument. In nineteen urban and commuter stores, the folklore was true and then some: deli buyers shopped 2.3 times more often, carried baskets a third larger, and the counter’s direct loss was cheap marketing by any honest accounting. In the other stores, the deli was exactly what the P&L said - a labour-heavy counter selling to a small, loyal, unprofitable-and-staying-that-way clientele who cross-shopped no more than anyone else. The follow-up question - full counter economics per store, labour and waste included, ranked - turned the range review’s oldest stalemate into a schedule: keep and invest in nineteen, convert the rest to pre-packed within two resets, redeploy the labour hours to the categories that had been begging for them.

The decision also stopped being the end of the story, which is the part retail range reviews never used to have. The freed counter space and labour hours went to chilled ready-meals, and instead of waiting two quarters to learn whether the bet paid, the trading director pinned a conversion dashboard the day the first reset finished: sales per square metre in the converted zone against its deli baseline, waste, and - the question that actually settles it - whether the old deli loyalists’ total spend held. Six weeks in, one store cluster was lagging badly enough to trigger the threshold she’d set; the planogram was wrong for small formats, and it was fixed in week seven rather than discovered in the autumn review. A range decision became a monitored experiment, which is what every range decision claims to be and almost none are.

No new analysis technique was involved. Every part of that answer had been sitting in Larsen & Holt’s own databases for years. What changed is that the person who owned the decision could ask the question during the decision, instead of routing it through an analyst, a data engineer and a six-week estimate - the loop that quietly converts “let’s find out” into “let’s revisit next half.” Walmart understood this earlier than almost anyone: its Data Café was built precisely to collapse trading questions from two or three weeks to about thirty minutes, on the stated logic that an insight arriving after the sales week is an insight about money already lost. Walmart needed the world’s largest private cloud and a staffed war-room in Bentonville. The reflex it bought is now a text box.

The layer that never sleeps

The decision-grade questions get the headlines; the standing rules pay the rent. Larsen & Holt’s mid-management runs a few dozen of them now, each one a sentence, each one watching for a specific, named way the business loses money.

The store operations director wrote the one that pays for the product by itself: flag any top-500 SKU that sells zero units in any store for a full trading day while the ERP shows more than five units on hand. That single rule is a phantom-inventory and dead-shelf detector across 85 stores - the ERP says the shelf is stocked, the shelf is empty or the stock is in a back room, and the weekly report will never notice because one SKU in one store rounds to nothing. It fires a handful of times a week. Each fire used to be a silent multi-day stockout.

The pricing team’s rule catches the failure mode that everyone has a scar from: a promotion that goes live in the leaflet but never activates at the till in one region. Zero redemptions four hours after launch now triggers a text; the fix happens the same morning instead of after a week of customers paying the wrong price and saying so on social media. And every regional manager starts Monday with a generated digest - stores ranked by like-for-like against last week, anything more than eight percent adrift flagged with the three categories driving the gap - which replaced a forty-tab spreadsheet that two analysts spent Thursday and Friday assembling and most recipients opened as far as tab three.

Monthly category review 30 days

Weekly trading report 7 days

Analytics Engine alert <1 hour

Walmart’s Data Café cut trading questions from 2–3 weeks to ~30 minutes.

Walmart Data Café: Forbes/Bernard Marr (2017), threshold alerts included. Bar for alert row enlarged for visibility.

Those two analysts are the quiet ending of this story. Their job had degenerated into data logistics: exporting, pasting, reconciling, formatting, shipping the pack, starting again. With the assembly gone, one now runs the promotion-effectiveness analysis the chain had talked about for three years - using the Analytics Engine’s ad-hoc layer to test cannibalisation store-by-store in afternoons rather than quarters - and the other owns the alert portfolio itself, tuning thresholds and retiring rules that stopped earning their interruptions. Headcount unchanged; output unrecognisable. The dashboard backlog, meanwhile, stopped growing, because the routine asks that fed it never reach a queue anymore.

What a retailer actually buys here

Strip the software away and what Larsen & Holt acquired is three habits. Transparency: the CEO’s testable question gets tested, in the room, against the company’s own data - so folklore loses its natural habitat. Awareness: the thousand small leaks that aggregate into the industry’s 6.5 percent - the empty shelf the system thinks is full, the promo that never fired, the store drifting quietly off plan - announce themselves within the hour, to the one person who can act, instead of compounding until they are large enough to appear in a report. And informed decisions at the top: range calls, space allocation, labour deployment made on evidence that arrives while the decision is still open.

The deli counter, for what it’s worth, is thriving in nineteen stores. The folklore turned out to be regionally true. It took four systems, one sentence and about fifty seconds to find out which regions - after six years of finding out nothing.