SAP B1 + chemicals
AI on SAP Business One for Chemical Manufacturers: Formulas, Batches and Safety Data in One Place
Short answer
AI for SAP Business One in a chemical company connects a private model to batch, formula and inventory data so formulators, QA and logistics can ask about lot attributes, formula costs, hazard classes and expiry in plain English. It reads Business One through the Service Layer and read-only views, retrieves SDS and REACH documents, and runs on-prem so formulations stay confidential.
- ERP
- SAP Business One
- Industries
- Chemicals, Specialty Chemicals, Coatings
- Written for
- Plant Manager
Chemical blenders, coatings makers and specialty formulators often choose Business One because it is affordable and covers finance, inventory and production for a mid-sized operation. But chemistry adds demands that general ERP screens handle only partly: formulas expressed in percentages, dual units of measure (kilograms and litres through density), potency or concentration adjustments, shelf-life dates and hazard classifications that drive how a drum can be stored or shipped.
Business One provides a production BOM (OITT, ITT1), batch master data with two free attribute fields plus manufacturing and expiry dates (OBTN), and warehouse-level quantities. What it lacks natively is a process-manufacturing layer: potency-based substitution, formula versioning with approvals, laboratory results against specification, and hazardous goods data. Companies close the gap with UDFs, add-ons, LIMS exports and spreadsheets.
The consequence is that crucial answers sit in different places. The formulator has the recipe in one file, QA has the certificate of analysis as a PDF, regulatory affairs holds the safety data sheet, and logistics has the dangerous goods classification in a freight forwarder portal. Anyone asking a cross-cutting question, such as which finished lots used a raw material whose SDS was just revised, has to assemble the picture manually.
This page explains how a private, read-only AI layer can connect those sources around Business One, where it genuinely saves time, and where it should stay out. Formulations are the core intellectual property of a chemical company, so keeping model and data on your own hardware is not a luxury here.
What usually gets in the way
The problems we hear most from plant manager teams running SAP Business One.
Formula cost is stale the day after you set it
Raw material prices change often, and the BOM cost in Business One is only as current as the last roll-up. Controllers rebuild margin views in Excel to see what a price change does to each formula.
Lab results and ERP batches are separate worlds
Certificates of analysis, assay values and viscosity or pH results live in a LIMS, PDFs or spreadsheets. Linking a result to the Business One batch (OBTN) and checking it against specification is manual.
SDS, REACH and CLP data is document-bound
Safety data sheets and registration information sit as files per supplier and product. Finding which products contain a substance of concern or an outdated SDS version means searching folders.
Unit of measure and density conversions cause errors
Buying in kilograms, blending in litres and selling in drums creates conversion risk. Wrong density assumptions produce inventory variances that are hard to trace after the fact.
Hazard and shipping data is not where planners look
UN numbers, packing groups and ADR or IMDG classes are held outside the item master or in UDFs of varying quality, so order release and carrier booking depend on tribal knowledge.
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.
Formula where-used and impact analysis
Ask which active formulas contain a raw material and what the cost impact is if its price rises. The agent walks multi-level BOMs and calculates impact per finished item.
Touches: OITT, ITT1, OITM, OITW, OPCH/PCH1 latest prices, OPLN
Outcome: Price-change impact reviews take minutes; the controller still decides about repricing.
Batch-to-lot genealogy
From a finished lot, list the raw material lots, suppliers and production order. In reverse, from a supplier lot with a quality problem, list affected finished lots and customers.
Touches: OBTN, OIBT, OITL/ITL1, OWOR, WOR1, ODLN, OCRD
Outcome: Faster containment decisions with a documented trace instead of manual reconstruction.
Certificate of analysis matching
Retrieval extracts assay and property values from CoA documents, compares them to item specifications held in UDFs or a spec document, and flags out-of-range lots for QA review.
Touches: OBTN attributes, UDFs for spec limits, attachments, LIMS export tables
Outcome: QA reviews exceptions instead of reading every certificate; release remains a QA decision.
SDS and substance lookup
Natural-language search across safety data sheets for items containing a named substance, with Business One item codes and on-hand quantities returned.
Touches: OITM, OITW, SDS document store, UDF CAS number fields
Outcome: Regulatory questions answered from a consistent index rather than ad hoc folder searches.
Expiry and retest-date watchlist
Lists batches approaching expiry or retest dates by warehouse, with open orders that could consume them first.
Touches: OBTN.ExpDate, OBTQ, ORDR/RDR1, OWOR
Outcome: Fewer expired drums and better first-expired-first-out discipline, depending on date capture quality.
Production variance by formula
Compare theoretical formula consumption with actual issues per production order and highlight over-additions and yield loss by line or operator shift.
Touches: OWOR, WOR1, OIGE, OIGN, OITT
Outcome: Targeted process reviews on the formulas that drift, rather than reviewing everything.
Dangerous goods readiness check
Before a sales order is shipped, check that each item has UN number, class and packing group captured and flag gaps.
Touches: ORDR/RDR1, OITM UDFs, ODLN
Outcome: Missing shipping data found before the carrier does.
Reference architecture
A private model and retrieval layer sit next to Business One. Structured questions go to curated read-only SQL views; document questions go to an index of SDS, CoA and specification files. The two are joined by item code and batch number, and every answer cites its sources.
- 1
Structured data access
Read-only views on OITT, ITT1, OBTN, OIBT, OWOR, WOR1 and warehouse tables for SQL Server or SAP HANA; Service Layer for authorized document reads.
- 2
Document ingestion
Parsers for SDS, CoA and specification PDFs, with extracted fields (CAS number, assay, revision date) stored beside the text and linked to item code and batch.
- 3
Semantic layer
Plant vocabulary mapped to schema: formula, intermediate, concentrate, potency, retest date, UDF names for hazard and spec data.
- 4
Model serving
Open-weight model served with vLLM or Ollama on local GPUs, with deterministic calculation tools for cost roll-ups so arithmetic is not left to the model.
- 5
Governance
Role-based access to formulas, with finer restriction for crown-jewel recipes, full audit logging, and approval for any write.
Integration notes for your ERP team
- Check that batch-managed items use the two attribute fields consistently, or move key properties into UDFs the semantic layer can read.
- Formula cost roll-ups should run in deterministic code or SQL, with the model only explaining the result.
- Handle dual UoM carefully: expose the UoM group (OUGP) and density UDF to the semantic layer so conversions are explicit.
- If a LIMS exists, take exports or a read-only API instead of querying its database directly.
- Restrict formula tables to named roles; use separate indexes for confidential formulations.
- Pin SDS documents by revision date so answers state which revision they used.
Deployment options
On-prem server in the plant or HQ
Formulators for whom recipes are the main trade secret
Everything, including document index and model, stays on the corporate network with no inbound or outbound internet requirement.
Private cloud tenancy
Companies hosting Business One at a regional provider
Dedicated virtual network next to the SAP host; private endpoints only; customer-managed keys for storage.
Air-gapped deployment
Defense, energetics or contract manufacturers under strict customer terms
Model weights and updates are delivered by controlled media; document ingestion happens inside the enclave.
Compliance and data control
How the architecture supports your obligations. Certification and accountability stay with your organisation; the design keeps the evidence straightforward.
REACH and CLP (EU)
Retrieval surfaces substance and classification data from your documents. Registration and classification decisions stay with regulatory staff.
TSCA (US)
Helps find inventory status notes and supplier statements already on file; it does not determine compliance status.
OSHA Hazard Communication and GHS
Supports locating current SDS revisions and flagging items with missing or old sheets.
ADR, IMDG and 49 CFR shipping rules
Checks that required UN data is captured on items, while classification and documentation remain a trained dangerous-goods advisor task.
ISO 9001 and customer quality agreements
Records every answer and source, supporting document control; the AI is not a controlled record itself.
Where Netray fits
ERPray
Grounded Q&A and read-only agents over formulas, batches and inventory with visible queries.
DataRay
Searching SDS, CoA and specification files and joining them to ERP data on-prem.
Custom build
Needed where LIMS integration, potency calculations or approval workflows are specific to your process.
How an engagement runs
Phase 1 . 1-2 weeks
Assessment
- -Formula and batch data review
- -Document source inventory
- -Confidentiality tiers for recipes
- -Use case shortlist
Phase 2 . 4-5 weeks
Pilot on formulas and batches
- -Where-used and cost impact agent
- -Batch genealogy trace
- -Evaluation set with formulator input
Phase 3 . 3-4 weeks
Document retrieval
- -SDS and CoA index
- -Spec comparison workflow
- -Revision-aware citations
Phase 4 . 2 weeks plus ongoing
Operate
- -Runbook
- -Access review
- -Question quality dashboard
Questions to ask any vendor, including us
A short list that separates real SAP Business One AI work from a chatbot demo.
- How are formulas isolated so only authorized roles can query them?
- Does cost roll-up use deterministic calculation or does the model do arithmetic?
- How are SDS revisions tracked and cited?
- What does the system do when density or UoM data is missing?
- Can it integrate with our LIMS read-only?
- Does anything leave our network during indexing or inference?
- How are answers evaluated against formulator expectations before go-live?
Frequently asked questions
Can SAP Business One handle chemical formulas?
It supports multi-level production BOMs and batch management, which many blenders use successfully. It does not provide a full process-manufacturing layer with potency adjustment, formula approval or lab integration, so companies use UDFs, add-ons or separate systems. AI can help read across those sources, but it does not replace a missing formula-control process.
Is it safe to put proprietary formulations in front of an AI model?
Only if the model runs where you control it. A private deployment keeps formulas on your own hardware, with role-based access and logging. Public cloud assistants are a different risk. We recommend separate indexes and tighter roles for crown-jewel recipes, and no use of your data to train shared models.
Can AI read safety data sheets reliably?
It can extract fields such as CAS numbers, hazard statements and revision dates and answer questions with citations, but SDS formats vary and scanned documents need care. We evaluate extraction accuracy on your own sheets and keep the original document linked so a regulatory person can verify.
Does this replace a LIMS?
No. A LIMS manages samples, methods and results. AI can read exported results or certificates and compare them to specifications in Business One, flagging exceptions for QA. Release decisions and lab records remain in your quality system.
How does it deal with density and unit conversions?
Conversions are done by deterministic code using the unit-of-measure groups and density values in your item data, not by the model's own arithmetic. When density is missing, the agent says so instead of guessing, which also gives you a list of data to clean.
What hardware is required?
A single GPU server with 24 to 48 GB memory handles a small team and a document index for a mid-sized chemical company. Larger document sets or many concurrent users need more. We size using your document count and expected users before you buy anything.
Is SAP's own AI enough for a chemical company?
For generic assistance, possibly. Vendor AI generally does not know your formula structure, UDFs or document sets, and it runs through vendor cloud services. If your needs are simple and cloud use is acceptable, try it first. If recipe confidentiality or custom data matter, a private layer fits better.
Related guides
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.
RAG + SQL + permissionsA 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 QM + on-prem AIAI for SAP Quality Management: Notifications, 8D, and CAPA Without the Manual Drafting
AI grounded on SAP QM data that triages notifications, drafts 8D and CAPA reports, and summarizes inspection trends, deployed on-prem for full traceability.
Agents + approval gatesAI Agents for ERP, Running On-Prem
A practical guide to on-prem AI agents for ERP: what they can safely automate, where human approval belongs, and how to design the guardrails.
SAP AI Buyer GuideSAP AI Consulting for On-Prem and Private Deployments
What to demand from an SAP AI consulting partner when data must stay on-prem or private: OData/BAPI mechanics, Joule vs. private LLM, and buyer questions.
SAP S/4HANA + private AIAI for SAP S/4HANA, Running On-Prem or in Your Private Cloud
Run AI on SAP S/4HANA without sending ERP data to a public API. On-prem and private-cloud architecture, CDS views, OData, and honest deployment trade-offs.
Plan it with numbers
RAG Infrastructure Sizing Calculator
Estimate vector storage, node RAM, generation GPUs, and monthly infrastructure cost for a retrieval-augmented generation deployment over your document corpus.
Free ToolOn-Prem LLM Total Cost of Ownership Calculator
Model the full multi-year cost of running LLMs on your own hardware, including GPU capex, power, cooling, support contracts, and operations staffing.
Free ToolGPU Sizing Calculator for LLM Inference
Work out how many GPUs you need to serve a given open-weight model to your user base, based on memory footprint and token throughput.
GuideEnterprise RAG Architecture: The Full 2026 Blueprint
A practitioner's blueprint for enterprise RAG in 2026: ingestion, chunking, embedding, retrieval, rerank, generation, and the eval loop that keeps it honest.
GuideA Data Governance Framework for Private AI
A practical data governance framework for private AI: classification, access controls, retention, and audit patterns that satisfy CMMC 2.0 and ITAR.
GuideSecurity Hardening for On-Prem AI Stacks
Security hardening for on-prem AI stacks: container isolation, model supply chain, prompt injection defenses, and audit logging for defense contractors.
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.