SAPUse Case

SAP Business One purchasing AI

AI for SAP Business One Purchasing, From Purchase Request to A/P Invoice

Short answer

AI for SAP Business One purchasing lets buyers ask plain-English questions about purchase requests, purchase orders, goods receipt POs and A/P invoices and get answers built from live Business One data, with the query shown. It reads through the Service Layer or read-only SQL views, respects Business One authorizations, and can run on-prem so supplier prices and terms stay inside your network.

ERP: SAP Business One, SAP Business One on HANA
Industries: Manufacturing, Distribution, Electronics
Written for: Purchasing Manager

Discuss your SAP Business One environment See use cases

A typical Business One purchasing team is two to six people who live in the Purchasing A/P menu: Purchase Request, Purchase Quotation, Purchase Order, Goods Receipt PO, Goods Return, A/P Invoice and A/P Credit Memo. The document chain is clean because each document is copied from the previous one, but nothing in the standard screens tells a buyer which open lines are actually late or which supplier keeps shipping short.

The questions that eat the day are cross-document ones. Which POs have been received but not yet invoiced? Which A/P invoices were posted at a price different from the PO? Which items are below minimum stock with nothing on order? Each answer lives in the rows of OPOR and POR1, OPDN and PDN1, OPCH and PCH1, and the people who can query those tables are usually the same two people who maintain the Crystal Reports.

SAP's own AI direction is aimed at the cloud portfolio and S/4HANA, and Business One customers frequently run SQL Server editions on their own servers with supplier pricing that is commercially sensitive. A buyer does not want that data pasted into a public chatbot, and the owner does not want to buy another cloud subscription to ask a question about last quarter's receipts.

This page explains how a private AI layer connects to Business One purchasing, which documents and tables it touches, where it genuinely helps a small buying team, and where a saved query or a user-defined query is still the simpler answer.

What usually gets in the way

The problems we hear most from purchasing manager teams running SAP Business One.

Open PO status needs a report nobody built

Business One shows open purchase orders per document, but a view of open lines by supplier, ship date and days late usually means a custom Crystal Report or a query pasted into the Query Manager that only one person maintains.

Received not invoiced is a month-end scramble

Goods Receipt POs that never get matched to an A/P Invoice sit in a clearing account. Finding them means joining PDN1 to PCH1 on base document references, which most users cannot do.

Price variances are discovered after posting

Unit price on the A/P Invoice can differ from the PO line, and Business One does not hard-block that by default. Buyers learn about it when the supplier statement does not reconcile.

Replenishment is rule of thumb

The MRP wizard produces recommendations, but small teams often skip it and buy from memory. Minimum stock, lead times on the item master and real receipt history rarely get compared.

Supplier performance lives in heads

Which supplier is consistently late, short-shipping or returning goods is visible only by reading Goods Return and receipt documents one at a time.

Where AI earns its place in SAP Business One

Each use case names the ERP objects it reads or writes, so your ERP team can judge the integration effort before anyone commits budget.

Open PO line status by supplier

A buyer asks which open PO lines are past their delivery date, grouped by supplier, and gets a ranked list with days late and open quantity.

Touches: OPOR, POR1 (DocDueDate, OpenQty, LineStatus), OCRD

Outcome: Replaces a daily manual scan of the Purchase Order list with a question answered in seconds.

Received-not-invoiced reconciliation

The assistant lists Goods Receipt PO lines with no matching A/P Invoice and groups them by supplier and age, with the base document chain shown.

Touches: OPDN, PDN1, OPCH, PCH1 (BaseType, BaseEntry, BaseLine)

Outcome: Shortens the month-end clearing-account review from an export-and-pivot exercise to a reviewed list.

Invoice-to-PO price variance review

Compares A/P Invoice line prices to the originating PO line and flags variances above a threshold the buyer sets.

Touches: PCH1.Price versus POR1.Price, OPCH, OPOR

Outcome: Catches price drift while the supplier conversation is still fresh instead of at statement reconciliation.

Draft purchase orders from low stock

An agent proposes draft purchase orders for items below minimum stock using item master lead times and preferred vendor, held as a draft for a buyer to approve.

Touches: OITW (MinStock, OnHand, OnOrder), OITM (CardCode, LeadTime), ODRF

Outcome: Moves routine replenishment from memory to a reviewed draft queue; the buyer still decides.

Supplier delivery and return scorecard

Summarizes on-time rate, short-ship frequency and Goods Return counts per supplier over a chosen period.

Touches: OPDN, PDN1, ORPD, RPD1, OCRD

Outcome: Gives negotiation and sourcing conversations evidence instead of impressions.

Landed cost review

Answers why an item's inventory cost moved after receipt by showing the Landed Costs document and the allocation to the goods receipt lines.

Touches: OIPF, IPF1, PDN1, OINM

Outcome: Removes a recurring accounting question that otherwise needs a consultant to trace.

Purchase approval backlog

Shows purchase drafts waiting in approval, who they are waiting on and for how long, so stalled requests are visible.

Touches: ODRF, OWDD, WDD1 (approval procedures)

Outcome: Reduces purchase requests that sit unnoticed because an approver was on leave.

Reference architecture

The AI layer reads Business One through its documented interfaces, not through direct writes. Queries run under a dedicated read-focused user so answers respect Business One authorizations, and any write goes through the Service Layer as a draft for human approval.

1
Business One connector
Uses the Service Layer (REST/OData, /b1s) for document-level reads and drafts, plus a read-only SQL or HANA view account for heavy aggregates. Session-based login uses a scoped Business One user, not SYSTEM or a superuser.
2
Semantic layer
Maps OPOR/POR1, OPDN/PDN1, OPCH/PCH1, OCRD, OITM and OITW to buyer vocabulary, including user-defined fields (the U_ columns) added by your partner, so late, open and received-not-invoiced mean what your team means.
3
Model serving
An open-weight model served with vLLM or Ollama on a single customer GPU server or in a private cloud, sized for a small buying team and modest concurrency.
4
Retrieval and agents
Answers are grounded in live queries with the generated query shown. Agents that propose purchase drafts or supplier emails stop at a draft in ODRF or an unsent message.
5
Governance and audit
Each question, query and answer is logged against the requesting Business One user; write-back requires an explicit approval step.

Integration notes for your ERP team

Service Layer is available on both SQL Server and HANA editions of Business One 10.0; the version and patch level determine which entities and OData features are exposed.
Use a dedicated Business One user with only the authorizations buyers have; avoid the manager account used for DI API jobs.
User-defined fields and tables (U_ and @ tables) are indexed from the data dictionary so partner customizations are queryable.
Heavy aggregates run on a read-only SQL account or HANA view; the Service Layer is used where document semantics matter.
Drafts are created as documents in draft state (ODRF), so Business One approval procedures still apply before posting.
Add-ons that write to the same documents (EDI, e-invoicing, barcode) are checked for overlapping behavior before enabling any write-back.

Deployment options

On-prem next to the Business One server

Manufacturers with SQL Server Business One who want supplier pricing to stay on the local network.

A small GPU server or workstation-class host runs the model and connector on the same network as the database and Service Layer, with no outbound path.

Private or sovereign cloud

Business One on HANA hosted by a partner, or companies with no data center room.

A dedicated tenant under your own cloud agreement, connected over a private link to the hosted Service Layer.

Hybrid for multiple subsidiaries

Groups with several Business One company databases across countries.

One governance layer, one connector per company database, with per-country data residency preserved.

Compliance and data control

How the architecture supports your obligations. Certification and accountability stay with your organisation; the design keeps the evidence straightforward.

SOX and internal controls

Read-only by default, drafts instead of postings, and a log of every question support segregation-of-duties arguments where Business One is a subsidiary ledger.

GDPR and regional data protection

Supplier contact data in OCRD and contact person tables stays inside infrastructure you control; no third-party AI API receives it.

Customer NDAs on cost and sourcing data

An architecture with no external data egress lets you show OEM customers where their part pricing is processed.

Export controls where parts are controlled

On-prem processing lets you keep controlled technical data in your own boundary; Netray does not itself hold ITAR registration, and your own compliance owner defines scope.

Where Netray fits

ERPray

Natural-language questions, dashboards and read-only agents over Business One documents are the core ERPray pattern, with the query shown for every answer.

Custom build

Teams with heavy add-on or user-defined field usage, or who want a supplier-email agent, often need a bespoke connector and approval flow.

How an engagement runs

Phase 1 . 1-2 weeks

Discovery

-Inventory of purchasing documents, UDFs and add-ons in use
-Authorization and user review
-Ranked list of buyer questions
-Deployment recommendation

Phase 2 . 4-6 weeks

Pilot

-Connector against a test company database
-Open PO and received-not-invoiced use cases live
-Accuracy review against known answers
-Logging validated

Phase 3 . 3-4 weeks

Production

-Cutover to production with a scoped read user
-Draft-only replenishment agent if approved
-Runbook for new questions
-Administrator handoff

Phase 4 . Ongoing

Scale

-Additional company databases
-Supplier scorecard and landed cost views
-Model refresh
-Quarterly accuracy review

Questions to ask any vendor, including us

A short list that separates real SAP Business One AI work from a chatbot demo.

Does the connector use the Service Layer and a scoped user, or a database login that bypasses Business One authorizations?
Where does the model run, and can you show the network path proving supplier prices do not leave our environment?
How are our user-defined fields and add-on tables mapped without a developer re-mapping each one?
What happens to the connector after a Business One patch or version upgrade?
Can every generated query be viewed and logged against the asking user?
Can we start read-only and add draft creation only after the answers are trusted?
What hardware is needed for a team of five buyers, and can it be a server we already own?
Which questions would you tell us a saved query handles better than AI?

Frequently asked questions

Can AI create purchase orders in SAP Business One?

It can prepare them, but not post them on its own. The agent creates a purchase document in draft state through the Service Layer, so Business One approval procedures still apply and a buyer confirms before anything is posted. Direct database writes are never used.

Does it work with both SQL Server and HANA editions?

Yes. The Service Layer is available on both, and heavy aggregates use a read-only SQL account or HANA view. The semantic layer is built on the same OPOR, POR1, OPDN and OPCH structures in either edition, though performance tuning differs.

Will it understand our user-defined fields?

It can, after the discovery step. User-defined fields appear as U_ columns, and the layer reads the data dictionary to map them to business terms. Meaning still has to be confirmed with your partner, because a field named U_Flag tells a model nothing by itself.

Is this better than a Crystal Report or saved query?

For a fixed, repeated question such as the open PO list, a saved query is cheaper and perfectly fine. AI earns its place on ad hoc cross-document questions where nobody built a report, and on summarizing results into a readable explanation.

Does supplier pricing leave our network?

Not in an on-prem or private-cloud deployment. The model and connector run on infrastructure you control, and there is no call to a public AI API. You should verify that by reviewing the network path and firewall rules during discovery.

How accurate are the answers?

Answers are grounded in a live query rather than model memory, and the query is shown so a buyer can check it. Accuracy depends on mapping quality, which is why the pilot compares answers against known-correct results before wider rollout.

What does it need from our Business One partner?

Mostly access and knowledge: a test company database, a scoped user, and a short walkthrough of customizations and add-ons. The partner does not need to rebuild anything, and can continue to own the Business One environment.

Related guides

SAP Business One + AI

AI for SAP Business One, Without Sending Your Books to a Public API

Add AI to SAP Business One without sending customer data or financials to a public API. Service Layer and DI API integration, sized for an SMB budget.

PO follow-up that does not wait on a person

AI Purchase Order Automation: Supplier Follow-Up and Expedite Without a Buyer Chasing Email

AI drafts routine PO confirmation and expedite messages from open ERP order data, with a buyer approving every message and change before it goes out.

AP automation + on-prem AI

AI for accounts payable invoice matching in your ERP

On-prem AI reads vendor invoices, runs 3-way match against your ERP PO and receipt, and routes only real exceptions to your AP team. No invoice data leaves your network.

Ask your ERP anything

Natural Language Query for ERP Data: Ask SAP, Infor, or Oracle a Question in Plain English

See how natural language query over SAP, Infor, Oracle, and NetSuite data works: grounded text-to-SQL, role-based permissions, and a visible audit trail.

RAG + SQL + permissions

A Private LLM Grounded on Your ERP Data

How a private LLM answers questions on your ERP data: RAG plus text-to-SQL, role-based permissions inherited from the ERP, and where each fits.

SAP B1 + machinery builders

AI on SAP Business One for Industrial Machinery Builders: Projects, Spares and Service in One Conversation

Private AI on SAP Business One for machinery builders: project and BOM status, spare parts, service contracts, warranty and CE documentation, on-prem and read-only.

Plan it with numbers

Free Tool

Talk it through with an engineer who knows SAP Business One

Bring one real question your team cannot answer from the ERP today. We will map the data path, the model, and where it runs, and tell you honestly if AI is the wrong tool for it.

Start the conversation See ERPray

All AI for ERP guides