01 the business
Health food retailer
3 shops plus online, 4,500 products
Reorder points were set years ago. Best sellers run out while slow lines fill the stockroom. Stock is split across the website, the shops and a small warehouse, and supplier delays are noticed when customers complain.
02 before
The problem
Reorder points were set once and never updated, so best sellers run out while slow lines fill the shelves. Stock is spread across a webshop, a shop floor and a warehouse, and nobody sees the whole picture. Supplier delays only get noticed when a customer complains.
- Popular items run out every few weeks.
- Cash sits in overstocked slow lines.
- Nobody sees stock across all locations at once.
03 the build
What the AI does
- Step 1: Pulls stock on hand, open purchase orders and recent sales from Shopify, Cin7, Unleashed or Lightspeed every day.
- Step 2: Works out how fast each item is selling, allowing for seasonality, promotions and supplier lead times.
- Step 3: Flags items likely to run out before the next delivery could arrive, and items that are overstocked.
- Step 4: Groups suggested reorders by supplier and drafts the purchase orders with quantities and reasons.
- Step 5: Sends a short daily digest to Slack or email, with urgent warnings sent straight away.
paused: waiting for a person to approve
How it was put together
- Pulled stock, open purchase orders and recent sales from each system daily.
- Worked out sales speed per item, allowing for seasonality, promotions and lead times.
- Flagged items likely to run out before the next delivery, and overstocked items.
- Grouped suggested reorders by supplier and drafted purchase orders with reasons.
04 human in the loop
Where a person stays in charge
The person who owns purchasing reviews the suggested quantities and approves each purchase order before it is sent to a supplier. They can override any suggestion and note why.
- The buyer approves every purchase order before it is sent.
- Any suggestion can be overridden with a note, which feeds back into the rules.
- Urgent warnings go out immediately; everything else waits for the daily digest.
05 rollout
From kickoff to live in 4 weeks
- Weeks 1–2Connect sales and stock systems and clean supplier lead times.
- Week 3Back-test suggestions against the last six months.
- Week 4Start the daily digest and draft purchase orders.
06 after
What changes
You hear about stock problems while there is still time to act. Reordering becomes a short daily check rather than a weekly scramble, and cash is less tied up in slow-moving stock.
Assume the top 200 items each run out twice a year for 3 days, selling $40 a day each.
200 × 2 × 3 days × $40 = $48,000 a year of lost sales on best sellers alone, before counting customers who switch shops.
What to measure
- Stock-outs on the top 200 items.
- Inventory value in slow-moving stock.
- Buyer time spent on reordering.
- Supplier on-time delivery.
07 evidence
What the research says
Gartner research put the average cost of poor data quality at $12.9 million a year per organisation.
Gartner, Data quality: why it matters and how to achieve it (opens in a new tab) · 2020
About 75% of generative AI's potential value sits in four areas: customer operations, marketing and sales, software engineering, and R&D.
08 lessons
Mistakes to avoid
- Wrong lead times. Suggestions are only as good as the supplier data.
- Ignoring promotions. Planned sales spikes need to be in the model.
- Ordering automatically. Keep the buyer's approval.
09 questions
Common questions
- Does it work for clinics and restaurants?
- Yes. Consumables and ingredients follow the same logic, with shorter shelf lives taken into account.
- Which systems does it read?
- Shopify, Cin7, Unleashed, Lightspeed, NetSuite and the main accounting tools.