The problem
A retail chain with more than 40 branches had no shortage of data. Sales, stock, branch performance, and campaigns were all being captured, and reporting sat on top of most of it. On paper the problem was solved.
The reality was a phone call.
The chief executive walks into a store, sees something he does not like, and rings the back office. Give me all this branch’s departments and how they compare to the company average. Sport and toys is behind. Why? Show me its stock, show me the shelves, compare it to branches the same size. Give me the manager’s number. And tell me about staff turnover there, because turnover breaks a store.
Every one of those answers already existed in the company’s own systems. None of them could be had without interrupting someone and waiting.
That wait is expensive in this business. The chain sells thousands of products with no brand of its own, constantly switching suppliers, pack sizes, and price points to find what moves. At any hour, something somewhere has drifted, and the money is in noticing early. Sales of an item might be up 5% across every store and flat in one branch. Is it out of stock? Did something else take the shelf? Did a local supplier substitute a variant? By the time a report answers that, the week is gone.
The same delay costs real money further upstream. Seasonal stock that arrives two days after the season starts is stock carried for a year — sales lost, cash tied up, and branches paying more to local suppliers to cover the gap.
There was a quieter problem too. The reporting was maintained by an outside consultant on retainer, one of roughly twenty companies he looked after. The understanding of how the business’s own numbers worked lived with him rather than with the business.
The solution
The same questions, asked in plain language, answered in seconds, on a phone.
That last part is not a detail. The reporting system runs on a desktop, and the people who need these answers are standing in a store. A tool that needs a laptop gets looked at in the morning. A question you can ask from the shop floor gets asked while the decision is still live.
What it does:
- Tells you how any store, region, or department is performing right now
- Compares a branch against similar branches, the chain average, or last season
- Explains why a number moved, not just that it did
- Prepares you for a store visit or a meeting before you walk in
- Warns you about stock and delivery problems before they reach the shelves
- Sends the reports you would otherwise ask someone to build
The warnings deserve a word, because this is where it beats the alerts most companies already have. A conventional alert is a light: red or green. Every false alarm costs credibility until people stop reading them. This gives you something closer to a colleague’s judgement — this shipment looks delayed, here are the products it affects, these branches will feel it first. That stays useful even when it turns out to be nothing, because it explains itself.
The hardest part of building it is not the technology. It is teaching the assistant how the business defines its own numbers.
Ask a simple-sounding question — how much stock do we have on the way? — and for a chain importing internationally the honest answer depends on what you mean. Goods sit at an overseas supplier, on a ship, in customs, in quarantine, in the central warehouse, in local warehouses, and with local suppliers, all at once. Everyone in the business knows which of those they mean. Nothing in the data does.
The same applies to growth that should exclude newly acquired businesses, to comparisons that need to strip out currency movements, to which stores are genuinely comparable, and to seasonal peaks tied to holidays that fall on different dates each year. Get those definitions wrong and you get confident, plausible, wrong answers. Getting them right is most of the work, and it is what makes the difference between an assistant people trust and one they double-check.
Why Node8
We build for the business, not for the data team. The people using this are store and regional managers, buyers, and executives. They do not want a query tool. They want the answer, and they will stop using anything that makes them work for it.
We start from your definitions. Before answering anything, we sit down and write out what your company actually means by growth, by comparable stores, by stock on the way. Most projects skip this and produce something that demos well and gets quietly abandoned once someone checks a number against the finance pack.
We work with the systems you have. No warehouse rebuild, no replacing your reporting. We start from a working copy of your data so you have something real in weeks, then connect deeper to your core systems as it proves itself.
We are careful about where the answers appear. Connecting a general chatbot straight to your systems produces confident nonsense — it has no idea what your business means by any of these terms. And for staff who are not going to scrutinise an AI’s reasoning, a purpose-built application is safer than an open-ended chat window: it bounds what can be asked and what comes back, and when a number is wrong it is clear who owns it.
This is a proven shape, not an experiment. The same approach runs in production at a multi-subsidiary industrial group, with the same architecture and their own business definitions built in. The technology carries across. The definitions are rebuilt for every client, because they are the business.
The gain here is not a cleverer algorithm. It is getting the right information to the right person at the moment they need it, which in most retail businesses is worth more than any model.
Node8’s enterprise AI work is at /tech.