Where the next 10 points of margin live in multi-branch retail
In a multi-branch retail business, the next 10 points of margin are rarely on the marketing line. They are in the small operational decisions that happen 50,000 times a week and get logged in spreadsheets. Here is where the leaks usually are.
The standard back-office failure
Mid-sized retail chains — pharmacies, grocers, kirana networks, multi-branch QSRs — tend to share a common pattern. The headline problem looks like inventory imbalance: bestsellers going out of stock at one branch while sitting as dead stock at another. The underlying issue is that operational decisions are happening downstream of an ERP nobody uses as a real-time decision tool.
Fill-in and sell-out are spreadsheet-driven. Returns are handled at each branch with no central visibility. SKU-mix decisions are made on intuition, not data. Each of those is a small leak; together they are a meaningful margin drag.
The three workstreams that compound
ERP integration plus an inventory tool
The first job is to make branch-level inventory visible in real time — not as a monthly close-out report but as a working tool for the operations team. The fill-in and sell-out automation that follows is downstream of that visibility. A thin digitisation layer that integrates with the existing ERP and surfaces the data the ops team needs in a tool they can actually use is usually a 6–8-week build, not a replatforming project.
Returns dashboard
Returns are the silent margin leak. A dashboard that surfaces returns by product, by branch and by month, with reverse-logistics cost attached, makes the leak visible. Once visible, the ops team can prioritise interventions — usually returns drop 25–40% within two quarters of getting the dashboard live.
Store-level product-fit ML
The most interesting workstream. A model that predicts which SKUs will sell better at which branch, based on historical sales patterns, returns data, seasonality and category indices. The model doesn't replace category-manager judgement — it makes the judgement cheaper to scale across dozens of branches.
Concretely, models of this kind surface patterns like: certain antibiotic SKUs have structurally higher sell-through in branches with a hospital within 800 metres; certain wellness SKUs have near-zero sell-through outside dense urban catchments. None of these are surprising in hindsight. The point is none of them were being acted on systematically.
The order of operations matters
Three principles, in order of importance:
- Plug into the existing ERP rather than replacing it. Replatforming projects in mid-sized retail businesses fail more often than they succeed. Don't try.
- Build around the operations team, not around a dashboard for leadership. The tool gets used because the people doing the work can use it. Dashboards only the CXO sees don't change behaviour.
- ML is a layer on top of structured data, not a substitute for the structured data. The model works because the integration work that came first made the inputs clean.
What this is worth on the P&L
A reasonable expectation, fully shipped: inventory costs down 25–35%, returns down 30–40%, revenue up 10–20%. The revenue lift comes downstream of the inventory work — SKUs that previously stocked-out at the wrong branches start being available where the demand is.
If a chain has more than 20 branches, it already has a returns problem it can't see, a SKU-fit problem it can't measure, and a fill-in problem that's eating 5–10 points of CM. The fix isn't a new ERP. The fix is a tool on top of the ERP that already exists, built around the operations team. The ML model is the last 20% — the first 80% is the visibility and the workflow.
Brick-and-mortar margin lives in the small decisions. The small decisions only get better when they're visible at the moment they're being made.