# AI for Business Central Inventory: Ask About Stock, Bins, Costing and Variances > Source: https://netray.co/erp-ai/business-central-inventory-ai > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: AI for ERP ## Short answer AI for Business Central inventory lets planners and warehouse leads ask questions about stock, bin content, reorder policy and cost in plain English, answered from the Item Ledger Entry, Warehouse Entry and Value Entry data with the query shown. It finds mismatches between inventory and general ledger, ranks slow stock, and proposes count and reorder-point changes for human approval. ## Kind use-case ## Erp Family microsoft ## Erp Products - Dynamics 365 Business Central ## Industries - Manufacturing - Distribution - Wholesale ## Region Global ## Persona Inventory Manager ## Eyebrow Business Central inventory AI ## Intro Inventory in Business Central is stored in three overlapping views that do not always agree on first look: quantity in the Item Ledger Entry, value in the Value Entry, and for locations that use warehouse functionality, bin quantity in the Warehouse Entry and Bin Content. Add item tracking by lot or serial, multiple locations with transfer routes, and costing methods that differ by item, and a simple question like how much do we have and what is it worth turns into a careful reconciliation. Most inventory managers solve this with a patchwork of Item Availability pages, the Inventory Valuation report, and Excel workbooks. Experienced people know which page tells the truth for which question. New staff do not. When the person who knows leaves, the knowledge leaves with them, and the questions end up in a queue for a consultant. A language model is not a replacement for the costing engine or for the posting rules, and it should never post an item journal on its own authority. What it can do well is read the data, explain what it sees, compare records that humans do not have time to compare, and point to the lines that deserve attention. The key design choice is that it reads through supported APIs and shows its working. This page describes what that looks like for Business Central, what it needs from your tenant, and where a simpler tool such as a saved page or Power BI report is the better answer. ## Pain Points ### Quantity, value and bin disagree **Detail:** Item Ledger Entry, Value Entry and Warehouse Entry are posted by different processes. When cost adjustment has not run, or a posting failed in a batch, the balances diverge and tracing the cause is manual. ### Reorder settings go stale **Detail:** Reorder Point, Safety Stock Quantity, Maximum Inventory and the chosen Reordering Policy are set at go-live and rarely reviewed. Demand shifts, and the planning engine keeps working from outdated numbers. ### Slow and excess stock hides **Detail:** Remaining quantity with no movement for a year sits on shelves and on the balance sheet. Finding it across items, locations and lots needs a report few people have time to build and read. ### Cycle counts are not targeted **Detail:** Physical Inventory Journals and Physical Inventory Counting Periods are often run by calendar rather than by risk. High-value, high-adjustment items get the same attention as stable low-value ones. ### Lot and expiry visibility is poor **Detail:** Item Tracking Lines and expiration dates are available but hard to view across locations. Near-expiry stock is found late, and recall traceability needs several pages to reconstruct. ## Use Cases ### Plain-English stock and availability questions **Description:** Ask what is available to promise for an item at a location, what is reserved against which orders, and what is inbound, with the source documents listed. **Erp Objects:** Item, Item Ledger Entry, Reservation Entry, Sales Line, Purchase Line, Prod. Order Component **Outcome:** Customer service gets an answer with its basis in seconds rather than navigating Item Availability by Event and Period by hand. ### Inventory to G/L reconciliation assistant **Description:** Compares the Inventory Valuation figures with inventory posting group accounts, explains differences such as unposted cost to G/L or expected cost, and lists the entries responsible. **Erp Objects:** Value Entry, G/L Entry, Inventory Posting Setup, Post Inventory Cost to G/L, Adjust Cost - Item Entries **Outcome:** Month-end differences are explained with named entries rather than chased for days. It does not post corrections. ### Reorder policy review **Description:** Analyzes consumption history and lead times per stockkeeping unit and suggests reorder point, safety stock and policy changes, grouped for a planner to accept or reject. **Erp Objects:** Stockkeeping Unit, Item (Reordering Policy, Safety Stock Quantity, Reorder Point), Item Ledger Entry, Item Vendor lead time **Outcome:** Fewer stockouts and less excess when applied carefully. Gains depend on demand variability and data quality. ### Slow-moving and obsolete stock list **Description:** Ranks stock by age since last movement and value, flags items with open demand or substitutes, and drafts a disposition proposal for review. **Erp Objects:** Item Ledger Entry (Remaining Quantity, Posting Date), Item Substitution, Bin Content, Inventory Valuation **Outcome:** A reviewed list for finance and operations before year-end, produced in minutes. Write-down decisions stay with people. ### Risk-based cycle count plan **Description:** Proposes which items and bins to count based on value, past adjustments and movement, then prepares the counting lines for the physical inventory journal as a draft. **Erp Objects:** Phys. Invt. Counting Period, Item Journal Line (Phys. Inventory), Warehouse Entry, Item Ledger Entry adjustments **Outcome:** Counts focus on items that drift, which usually improves accuracy per hour counted. A small share of count time may be freed. ### Lot, expiry and recall tracing **Description:** Answers which customers received a lot, which vendor supplied it, and what stock of that lot remains, by walking item tracking and application entries. **Erp Objects:** Item Ledger Entry (Lot No., Expiration Date), Item Application Entry, Item Tracking Lines, Posted Sales Shipment **Outcome:** Trace time falls from hours of page-hopping to a documented list, which supports recall response. ### Transfer and location balancing **Description:** Spots locations with excess and others with shortages of the same item and proposes transfer orders for a planner to approve. **Erp Objects:** Location, Transfer Header, Transfer Line, Stockkeeping Unit, Item Ledger Entry **Outcome:** Reduces emergency purchases when stock exists elsewhere, subject to transfer lead time and cost. ## Architecture **Overview:** The assistant reads through Business Central APIs and a small set of custom API pages that expose ledger-level tables not in the standard API, aggregating server-side where possible so large ledgers are not pulled in full. A scheduled snapshot of inventory metrics supports trend questions without hammering production. **Layers:** #### Access layer **Detail:** Microsoft Entra application with OAuth 2.0 client credentials mapped to a Business Central user whose permission set is limited to inventory and item read access. #### Data connectors **Detail:** API v2.0 for items and locations, plus custom API pages over Item Ledger Entry, Value Entry, Warehouse Entry and Bin Content built in an AL extension, filtered and paged. #### Snapshot and index **Detail:** A nightly snapshot of on-hand, age and consumption per item and location stored outside the ERP for trend and slow-stock questions, with documents such as count procedures indexed for retrieval. #### Model and reasoning **Detail:** A self-hosted open-weight model with tool calls for each query type, instructed to show the filters used and refuse to estimate numbers it cannot read. #### Approvals and write-back **Detail:** Count lines, transfer orders and reorder changes are created as drafts or suggestions. Posting remains a user action under existing permissions and journal batch templates. ## Deployment Options ### Customer cloud tenant **Fit:** Business Central online users with Azure skills **Detail:** Model and snapshot store run in the customer's own Azure subscription, connected to the BC environment through the API with private networking where available. ### On-premises inference **Fit:** On-prem BC and companies with strict data residency **Detail:** A GPU server in the customer's data center, with the API calls staying inside the network. Updates to model weights are brought in by a controlled process. ### Analytics-first pilot **Fit:** Teams unsure whether AI beats a good Power BI report **Detail:** Start with the snapshot and a handful of questions on a sandbox. If reports meet the need, stop there; if people want free-form questions, extend. ## Compliance ### Inventory valuation and G/L figures are read, not changed. Every number in an answer links to ledger entries, giving auditors a trail. **Framework:** Financial reporting and audit ### For medical device customers, lot and serial tracing uses the item tracking and application entries, and the assistant does not alter genealogy. Validation of the tool for its use is the customer's responsibility. **Framework:** FDA 21 CFR Part 820 and ISO 13485 traceability ### Lot and expiry queries support the traceability records a food business must keep. Completeness depends on how consistently item tracking is used at receipt and shipment. **Framework:** FSMA 204 and food traceability ### Cost and value fields are hidden from roles that cannot see them in Business Central. The assistant applies the same restriction through the user's identity. **Framework:** Access control ## Integration Notes Item Ledger Entry and Value Entry tables can be very large; always filter by item, location and posting date and use server-side paging, and consider a summarizing API page for aggregates. Run Adjust Cost - Item Entries on its normal schedule before relying on cost figures; the assistant should state the date of the last cost adjustment in valuation answers. For locations using bins, join Bin Content and Warehouse Entry rather than inferring bin stock from the item ledger, since Item Ledger Entry does not carry bin. Costing method matters: Standard, FIFO, Average and Specific costing behave differently on variances, and the assistant needs the item's costing method to explain differences correctly. Item tracking: lot and serial numbers sit on Item Ledger Entry and Reservation Entry, and expiration dates require the item tracking code to be configured for them. Writes go through standard journals with journal batches reserved for the assistant, so reviewers can see its drafts separately. ## Product Fit ### ERPray **Why:** Question answering, dashboards and agents over live inventory data with queries shown and role-based access. ### DataRay **Why:** Useful when warehouse data also lives in a WMS database or scanner logs outside Business Central. ## Engagement ### Data and policy review **Duration:** 1-2 weeks **Deliverables:** - Inventory and costing configuration review - Question inventory from planners and warehouse - Permission design - Sandbox access ### Read-only assistant **Duration:** 3-4 weeks **Deliverables:** - Custom API pages for ledger tables - Stock, age and availability questions - Snapshot job and trend views - Accuracy test set ### Analysis workflows **Duration:** 3-4 weeks **Deliverables:** - Reconciliation assistant - Reorder review suggestions - Risk-based count proposals - Reviewer feedback loop ### Production and handover **Duration:** 2 weeks **Deliverables:** - Production connection - Monitoring and cost report - User training - Roadmap for transfers and tracing ## Buyer Questions - How do you avoid pulling entire ledger tables, and what is the effect on production performance? - How does the assistant treat our costing methods and unadjusted cost? - Will it explain differences between inventory and the general ledger with named entries? - Can it read bin and lot data, and what do we need to enable in our configuration? - Who approves changes to reorder points, and where are suggestions stored? - How are custom fields and extensions on items handled? - Could a Power BI report meet our need at lower cost? - How is the model kept current with our data without retraining? ## Faq ### Can AI fix inventory mismatches in Business Central? **Answer:** It can find and explain them, but posting corrections should stay with a person. A typical case is a gap between inventory valuation and the general ledger caused by cost not yet posted to G/L. The assistant names the entries and the likely cause, and your finance team decides whether to run the batch job or post an adjustment. ### Does it replace the planning engine? **Answer:** No. Business Central's planning engine still calculates requirements from reorder policy, demand and supply. The assistant reviews the inputs, such as reorder point and safety stock, against actual consumption and suggests changes. Those changes are applied by a planner, and the planning engine keeps doing the calculation. ### Will it work with warehouse management features? **Answer:** Yes, if the data is exposed. Locations with bins use Warehouse Entry and Bin Content, which are not in the standard API, so we add custom API pages. Warehouse documents such as receipts and picks can be read the same way. Directed put-away setups need a review so answers reflect your bin rules. ### Is Power BI enough instead? **Answer:** For fixed dashboards of known measures, often yes, and it is cheaper. A language layer adds value when people ask new questions each day, need explanations rather than numbers, or want to combine ledger data with documents. Some teams use both, with the assistant querying the same snapshot the reports use. ### How does it handle lots and expiry dates? **Answer:** It reads item tracking data from the item ledger and reservation entries, so lot, serial and expiration answers depend on tracking being used consistently at receipt and shipment. If a location does not track lots, the assistant cannot invent that history, and it should say so instead of guessing. ### What about performance on a large database? **Answer:** Large ledgers are why queries are filtered, paged and where possible aggregated by a purpose-built API page. A nightly snapshot answers trend and age questions without live scans. We test with production-sized data in a sandbox and set limits, so a single question cannot slow the system for other users. ### Can it run on-premises? **Answer:** Yes. Business Central on-premises exposes the same API pages and web services, and the language model can run on a local GPU server. Nothing needs to leave your network. Plan for a controlled way to update model weights and for monitoring of GPU capacity as usage grows. ## Search terms business central inventory ai, ai for business central inventory management, business central item ledger entry query, business central inventory valuation reconciliation, business central reorder point optimization, business central cycle count ai, business central slow moving stock, business central lot traceability, business central copilot inventory, on-prem ai business central inventory --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json