Where AI Helps in a Warehouse, and Where It Does Not
Perspective · AI in Business · AI Practice · 2026-10-08T09:00:00.000Z
AI is useful for forecasting, slotting and exceptions. It is a poor fit for fixing broken process.
Where AI earns its place
- Demand forecasting. Predicting what will sell, and when, so you stock the right quantity. Seasonal and promotion-driven items benefit most.
- Slotting. Deciding where each item should live so fast movers sit close to dispatch and pickers walk less.
- Exception spotting. Flagging counts that do not match, picks that take unusually long, or orders likely to miss their cut-off, so supervisors look at the right things first.
- Labour planning. Estimating how many people each shift needs from expected order volume.
- Reading inbound paperwork. Pulling item and quantity details from delivery notes and invoices to speed up receiving.
Where it does not help
- Broken process. If goods are not scanned at receiving, put-away and picking, the system never knows where stock really is. AI trained on that data gives confident wrong answers.
- Poor item master data. Missing sizes, weights or units make slotting and forecasting unreliable.
- Rare, one-off events. A model needs repeated patterns. A once-a-year disruption is a job for a planner.
- Safety and rule-based decisions. Load limits, hazardous goods rules and legal requirements should stay as fixed rules, not predictions.
- Replacing floor leadership. Good supervisors know things the data does not show.
A quick readiness checklist
[ ] Stock counts in the system match what is on the shelf. [ ] Every item has size, weight and unit of measure recorded. [ ] Goods are scanned at receiving, put-away and picking. [ ] Every storage location exists in the system. [ ] You have at least a year of order history (more for seasonal items).
If you cannot tick most of these, spend the next quarter on process and data. It will pay back whether or not you ever use AI.
How to start
- Pick one question, for example "which 200 items should sit nearest the dispatch bay?".
- Record today's result so you can prove the change.
- Run the model alongside your current method for a few weeks. Let people review its suggestions.
- Measure and decide. Keep it only if it beats the baseline.
- Expand slowly to the next question.
Is AI worth it for a small warehouse?
Often not yet. A well-set-up warehouse management system with clean data and simple rules delivers most of the gain.
Do we need our own data scientist?
Not to begin. You need someone who owns the question, someone who owns the data, and a partner who can build and test the model.
What is the first sign it is working?
Supervisors start trusting its exception list and spend less time hunting for problems.