AI stock-redistribution agent connected to the ERP

Continuous reading of SAP stock levels, matching near-expiry surplus ↔ out-of-stock territories, economic validation before any transfer instruction.

−31 %

expiry write-offs (tracked SKUs, 6 months)

+7,3 %

sales in underserved territories

< 1 h

to a validated redistribution instruction

Context

The problem

Expiry on the surplus side

Stock nearing its use-by date in overstocked or slow-moving warehouses. The time to analyze SAP and decide manually closes the window: the product is destroyed.

Stockout on the demand side

The same SKU out of stock in a neighboring territory while surplus sits a few hundred kilometers away. The match is never made in time.

SAP data with no decision loop

Every signal exists in the ERP (levels, use-by dates, demand signals), but turning it into an economically viable transfer instruction takes several days.

Analytical load on planners

Two to three hours each morning pulling reports and simulating moves by hand, at the expense of higher-value trade-off decisions.

Solution

Deployed architecture

Ingestion

Continuous SAP reading

Stock levels, expiry dates and demand signals ingested as a stream: no weekly batch export.

Matching

Network opportunity detection

For each near-expiry SKU, a scan of the entire network to locate the zones running a deficit on the same reference.

Economics

Transport viability gate

Transfer cost weighed against the loss avoided and the incremental margin. Only profitable moves become instructions.

Delivery

Prioritized instruction queue

Ready-to-execute instructions for the supply team: no intermediate analysis needed on standard cases.

Results

Measured results

Measured in production conditions on the deployed scope.

−31 %

Expiry write-offs

Across the monitored SKU portfolio, measured over six months post-deployment.

+7,3 %

Sales in underserved territories

Uplift in zones previously subject to recurring stockouts, now fed by redistributed surplus.

< 1 h

Decision speed

Versus 2 to 5 days of planner analysis on the same scenarios.

Load

Planner time freed up

A significant cut in daily analysis time: capacity reallocated to strategic trade-offs.

Lessons learned

What we took away

The agent does not replace SAP: it reads the ERP and produces instructions, while logistics execution stays in the existing processes.

The economic gate (transport vs loss) prevents absurd transfers and secures business adoption.

Real-time matters more than model sophistication: a decision on day 3 for a perishable SKU is a non-decision.

Involving planners in the human validation queue for edge cases accelerates trust.

Start on a subset of SKUs with high write-off value and uneven rotation across regions.

Get started

A similar case on your site?

Tell us about the process, the data available and your deployment constraints. We will assess feasibility within 48 hours.