Sales are dropping in the South — are our warehouses understocked, or is something else going on?

Sales in the South are down 18% on last quarter. Every other number on your dashboard looks fine. Stock levels say "healthy." Suppliers are delivering on time. The dashboard says there is no problem. The South says there is. We talked to 22 people — store managers, warehouse staff, regional directors, the forecasting team, and the people who built the ordering screen.

Each of these three findings was checked the rule-of-three way before it landed in this report — three independent people said the same thing, the settings out of the computer back them up, and a second pair of eyes (an independent reviewer) agreed the evidence supports the claim. Two more claims came up with weaker support — those are in the Possible section. You can see they exist. They don't have the weight of a finding. Two more claims were about one specific person each — those are dropped, with the reason visible in Dropped.

What this report does not do: tell you what to fix first, recommend a tool, or point at anyone. Those decisions are yours. The pieces of evidence we relied on are linked from each finding and listed in full in the Appendix.

Here is how a supplier delivery is supposed to end up on a shelf in a South store. We've laid out every step. Three of them are broken — the red ones below. Each break links to the finding that explains why.

At the end of 2024, the company switched to a new ordering system. During the switch, someone set a rule that tells the computer to send every supplier delivery to the Midlands warehouse before any other one. It was meant to be temporary — to help the Midlands site cope while everything else was being set up. Nobody put an end date on it. Nobody went back to check. It's still on today^¹.

What people said

When a store manager opens the ordering screen, there's a green message at the top that says "National stock: healthy." The way that message is built, it only checks one warehouse — the Midlands. The check was set up during the 2024 switchover as a placeholder. Someone labelled it "national" before the other warehouses had been added in. The other warehouses were never added in^².

What people said

After months of empty shelves and a green message that kept saying everything was fine, store managers in the South stopped putting in re-orders. The system that records re-orders shows the number coming from South stores dropping by 41% between summer 2025 and spring 2026^³. Two managers told us, in nearly the same words: it took time to put in a re-order, nothing ever came, so they stopped.

What people said

Two of the things people told us were about one specific person each, and only one person said them. Our rule is that we won't put a finding into a report if it points at one person and only one other person said it. We show you that the claims existed, so you can check we played fair. The actual content stays out of this report. If the same pattern comes up in a later investigation, those single voices will join up with the new ones.

Here are the rules we follow on every Deep investigation. Each one leaves a trace in the Appendix, so you can check we kept to them.

How we asked

Every person we spoke to got a Microsoft Teams message first. It said what question we were working on. It said their manager wouldn't see what they answered. It said they could skip any question. They could end the chat any time. They could change anything we'd written down before the report went out. Eighty-seven out of every hundred people we invited said yes. The list of who agreed to what is in ART-01 — names are stripped.

When we refuse a question

If the question you ask us is really about one specific person — "is X bad at their job" — we won't run it. We check before we start. Twice during this investigation, a person we were chatting with asked us not to follow up on something. Both times we stopped. We wrote down what kind of question it was (not their name) in ART-02.

The rule of three

Nothing on one source. Every cause we keep has been checked three independent ways: by our own rules (built-in steps the AI can't skip), by a second pair of eyes (an independent reviewer reading the evidence and the claim, who has to agree), and by trying to prove it wrong (a test designed to break the cause; if it survives, it stands). In process work, the second check doubles as "three independent people said the same thing." Below the three-check bar, a claim moves to Possible — you can see it exists, but it isn't a finding. If it would name a specific person we haven't talked to enough, we drop it, with the reason visible in Dropped.

How we keep voices private

In this report we call people by their job and where they work — "Warehouse Ops · Cardiff," "Store Manager · Bristol." Not by name. If only one person in the company has a particular job, we don't use that job as a label either — we either group them with others, or we move the claim to Possible. We'll delete the anonymous summary too if you ask.

Who checks the report before you see it

Before this report reached you, a person at Confr read it end to end, and so did the person who asked us to run it — M. Okafor at your end. Both of those reviewers see the full Possible and Dropped sections, including what we dropped. You see only what's in this document.

Where the data lives

Everything we collected stays in Europe. It's encrypted on the way and at rest.

Every claim in this report links to one or more pieces of evidence below. Click any blue evidence button next to a finding and it'll bring you here and open the matching entry. The list below is everything we used.