AI agents for retail: fewer stockouts, fewer markdowns, buyers back on judgment calls
KOI AI deploys governed agents on top of a retailer's existing POS, ERP, and inventory systems: a forecasting agent that flags stockout and overstock risk by SKU and store, a replenishment agent that drafts purchase orders for a buyer's approval, and a vendor-communications agent that reads inbound vendor email, tracks confirmations, and drafts replies. No purchase order is sent without a human sign-off. In a published engagement, an 80-store specialty retailer cut stockouts by 31%, reduced markdown loss by 22%, and gave each buyer back 15 hours a week.
What agents do in retail
Demand forecasting by SKU and store
Flags stockout risk on the fast movers and overstock risk on the slow ones, store by store.
Replenishment and purchase-order drafting
Drafts purchase orders from the forecast and the vendor terms, for the buyer to approve or adjust.
Vendor-communications triage
Reads inbound vendor email, tracks confirmations and changes, and drafts replies for approval.
Markdown and promotion support
Identifies the items and stores where markdowns are likely and drafts options for the merchant to decide.
Store questions
Answers associates' policy and stock questions from the systems of record.
Returns and exceptions
Classifies exceptions and drafts the resolution, with anything above a threshold going to a person.
The systems agents work in
- ◆Point of sale
- ◆ERP and inventory
- ◆Vendor email
- ◆Planning spreadsheets
- ◆E-commerce platform
Approval and risk controls
- ◆No purchase order is sent without a buyer's sign-off.
- ◆Spend and quantity thresholds route larger orders to a senior approver.
- ◆Every forecast, draft, and action is recorded on the accountability ledger.
- ◆Agents operate within the access you grant to your systems.
What an engagement is measured on
- ◆Stockouts on top sellers
- ◆Markdown loss
- ◆Buyer hours per week
- ◆Inventory turns
- ◆Vendor confirmation time
What agents do not do here
- ◆Forecasts depend on sales history and data quality. New and seasonal items need the buyer's judgment.
- ◆Vendor adoption of structured confirmations varies. The agent works with the email you receive today.
Agentic demand forecasting and vendor automation across an 80-store fleet
A specialty retailer replaced reactive reordering and spreadsheet-driven vendor comms with governed agents that forecast demand and draft purchase orders for human approval.
Read the case study and how the numbers were measured →- AI Readiness Assessment →
2 weeks
- AI Agent Implementation →
4–8 weeks to the first production agent
- Fractional AI Officer →
Ongoing, monthly
Questions from retail buyers
Does the agent place orders on its own?
No. It drafts the purchase order with its reasoning, and a buyer approves, adjusts, or rejects it. Larger orders route to a senior approver by threshold.
What data does forecasting need?
Sales history by SKU and store, current inventory, and vendor lead times, from the systems you already run. The scoping week tells you whether a gap would limit the forecast.
How do buyers work with it day to day?
They approve agent-drafted orders and exceptions instead of building them by hand, and spend the time on vendor negotiation, assortment, and the judgment calls the forecast cannot make.
One process in your business, two business days.
Describe the task. Get a working agent before you spend anything.
Take the free assessment →