AI Inventory and Margin Anomaly Detection

The problem: the losses are already in your data

A company running SAP Business One at any real volume posts thousands of A/R Invoice lines, Goods Receipts, Goods Issues and Inventory Transfers a month. Every one of those documents is correct in isolation and the system will never complain about any of them. That is precisely why margin leaks survive.

A salesperson applies a discount that puts a line below moving-average cost — the A/R Invoice posts cleanly. A price list is updated for one item group and a decimal is misplaced — the Sales Order takes it without objection. A warehouse count adjusts an item down by a few units every month at the same bin, and each Journal Entry is individually unremarkable. An item stops selling in March and is still carrying stock and warehouse space in November, because nothing in the system is designed to raise its hand.

By the time these appear in a report, they appear as an aggregate: gross margin down two points, inventory value up, and no obvious cause. Finding the cause means someone manually trawling documents, which nobody has time to do, so the pattern continues.

What ECOSIRE builds

ECOSIRE builds a per-tenant SAP Business One add-on that reads your transactional history, learns what normal looks like for your business, and surfaces the specific documents and items that deviate from it. This is scoped and built for your company — it is not a pre-built product you download.

Margin analysis at line level

The add-on reads A/R Invoice and A/R Credit Memo lines through the Service Layer, joins them to item cost (moving average or FIFO as configured in your company), and computes realised margin per line, per document, per Business Partner, per item and per sales employee. It flags lines sold below cost, lines below a configured margin floor, and — more usefully — lines whose margin is a statistical outlier against that item's own recent history or against comparable customers. Rebates, freight recovery and other charge lines can be included in the margin basis if your finance team defines them that way during scoping.

Price-list drift detection

Price lists, special prices for Business Partners, period and volume discounts and discount groups are all snapshotted on a schedule. The add-on compares each snapshot against the last and against the item's cost trend, and raises an alert when a price moves by more than a threshold, when a price falls below cost, when a special price diverges from the base list beyond an expected band, or when an item's price has not moved while its cost has climbed. That last case is the one that never gets caught manually and quietly erodes margin for months.

Dead, slow and excess stock

Using Goods Issue, Delivery and Inventory Transfer history against on-hand quantity per Warehouse, the add-on classifies items into movement bands, computes days of cover against recent demand, identifies stock with no outbound movement inside a configurable window, and highlights excess relative to reorder parameters on Item Master Data. Results are expressed in value, not just quantity, so the review conversation starts with the items actually tying up cash.

Shrink and adjustment patterns

Inventory adjustments, Goods Receipts and Goods Issues without a linked purchase or sales document, negative-quantity corrections and cycle-count variances are aggregated by Warehouse, item group, bin, user and time of day. The add-on scores repeated patterns rather than single events — a warehouse that adjusts the same item group down at the same interval, or a user whose correction rate sits far outside their peers' — and presents the underlying documents so a supervisor can look at the actual transactions.

How the scoring works

Detection is a layered approach rather than one black box. Deterministic rules cover the cases you already know matter — below cost, below floor, no movement in N days. Statistical methods handle the rest: seasonally-aware baselines per item and item group, robust z-scores and interquartile range tests for outliers, and change-point detection on cost and price series. Where you want narrative explanation of a flagged cluster, an LLM summarisation layer can be included, operating strictly on the aggregates the add-on has already computed — it explains findings, it does not decide them, and it never writes to SAP Business One. Every alert carries the rule or statistic that produced it and links to the source document, so a reviewer can always see why something surfaced.

Where the findings live

Anomalies are written to User-Defined Tables and surfaced through a User-Defined Object review form inside SAP Business One, with status, assignee and resolution notes, so investigation happens in the ERP rather than in an exported spreadsheet. Approval procedures can gate a Sales Order or A/R Invoice that breaches a margin floor, sending it for sign-off instead of posting silently. Crystal Reports provide the periodic management view, and scheduled email digests push the highest-value open items to the people who own them.

What it does not do

The add-on detects, ranks and routes. It does not automatically adjust prices, write off stock, or post corrections. Every remediation stays a human decision taken in SAP Business One through the normal document flow.

Who this is for

Distributors and wholesalers with wide catalogues and negotiated pricing, where per-line margin is invisible in aggregate reporting. Manufacturers carrying components whose cost moves faster than their sell price. Multi-warehouse retail and trading operations where shrink is suspected but never evidenced. Any SAP Business One site whose finance team knows margin is leaking but cannot point at the documents.

Technical fit

Built against SAP Business One 10.0 on Microsoft SQL Server or SAP HANA, on-premise or cloud-hosted. Reads are performed through the Service Layer and, where volume demands it, through read-only queries against the company database; writes of anomaly records use UDTs and UDOs via the DI API and Service Layer. No SAP tables or stored procedures are modified. Analysis runs on a schedule you set, sized so it does not contend with your transactional load — typically overnight, with an on-demand run available from the review form.

How delivery works

1. Scoping call. We review your margin definition, your costing method, your price-list structure, your warehouse and bin layout, and the anomaly types that actually matter to you. We look at data volumes to size the analysis run.

2. Fixed quote. You get a written scope naming each detector, its inputs, its thresholds, the review workflow and the reporting, at a fixed price. Nothing is built before you approve it.

3. Build. ECOSIRE develops the add-on and calibrates baselines against a copy of your historical data, so thresholds are tuned to your business rather than to generic defaults.

4. Install in test. Installed into your test company database and run against your real history. You review the first output together with us and we tune sensitivity — this step matters, because an alert list nobody trusts is an alert list nobody reads.

5. Install in production. After your written sign-off, we install into production, set the analysis schedule, and configure recipients and any approval gates.

6. Support. A defined post-go-live support window covers defects and threshold recalibration once you have seen live output.

Typical build and delivery is two to four weeks from approved scope, depending on the number of detectors and the depth of history to be baselined.