SAPERP PlatformEurope

SAP ByDesign + private AI

AI Integration for SAP Business ByDesign, Built on OData

Short answer

SAP Business ByDesign is a multi-tenant SaaS ERP, which means you cannot self-host the ERP itself, but you can build an AI layer that reads ByDesign through its OData and SOAP communication scenarios and runs on infrastructure you control. For a mid-market finance team, that means natural-language reporting, variance commentary drafting, and document search without financial data ever reaching a public model API.

ERP
SAP Business ByDesign
Industries
Manufacturing, Professional Services, Wholesale Distribution
Written for
Finance Director

SAP Business ByDesign is a good fit for a mid-market finance director precisely because it is fully managed: SAP hosts it, patches it, and upgrades it twice a year, and you are not carrying the infrastructure burden a larger on-premise SAP deployment would require. The trade-off is that ByDesign has no on-premise option and no direct database access, everything goes through its OData and SOAP web services, which shapes what an AI integration can and cannot do.

In practice, that limitation is less restrictive than it sounds. ByDesign's business configuration lets you activate specific communication scenarios and scope a technical communication user to exactly the entities, general ledger, cost centre reporting, sales quotes, business partner data, an AI layer actually needs, which is a reasonably clean integration surface once you know which scenarios to enable.

What a mid-market finance team is usually missing is not more data inside ByDesign, it is the ability to ask a plain-language question and get an answer, and the time saved not writing month-end variance commentary from scratch. A finance director with a lean team rarely has a dedicated BI resource to build that layer internally, and sending financial data to a public AI assistant is not something most finance leaders are comfortable approving, even when the underlying ERP is already SAP-hosted cloud.

This page describes an AI layer built specifically around ByDesign's OData integration model, deployed in a private cloud tenant or on your own infrastructure, so the model and the data it reasons over never leave an environment you control, and the outputs stay drafts a finance professional reviews rather than numbers that go straight into a board pack.

What usually gets in the way

The problems we hear most from finance director teams running SAP Business ByDesign.

ByDesign's own reporting requires knowing where to look

Answering a specific finance question, why did this cost centre run over budget this quarter, means navigating to the right report variant and filtering it correctly, a skill that sits with one or two people on a lean finance team.

Month-end variance commentary is written from scratch every cycle

Drafting the board-pack narrative explaining actual-versus-budget variance by cost centre is a recurring, largely repetitive writing task that still takes a finance professional several hours each close.

ByDesign has limited native extensibility for building custom AI features

The Partner Development Infrastructure supports lightweight extensions, but building a genuinely custom AI experience inside ByDesign itself is not realistic; any serious AI capability has to live in a layer alongside it, reading through OData.

A lean finance team has no dedicated BI resource

Mid-market finance directors rarely have a data analyst on staff to build and maintain natural-language reporting internally, which means this kind of capability either comes from a vendor or does not happen.

Sending financial data to a public AI assistant is a hard no for most finance leaders

Even though ByDesign itself is SAP-hosted cloud, routing general ledger or customer data through a public model API for AI-assisted reporting is a different risk decision, and most finance directors we talk to are not willing to make it.

Where AI earns its place in SAP Business ByDesign

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

Natural-language financial reporting

A finance user asks a plain-language question, for example a cost centre variance summary for the current quarter, and gets a grounded answer with the underlying report or journal entries cited.

Touches: General ledger and cost centre reporting OData services, journal entry vouchers

Outcome: Answers a routine finance question in the time it takes to type it, instead of a report-navigation exercise that only certain team members can do quickly.

AP/AR exception narrative

The agent summarizes open item and payment run data into a plain-language cash and aging position, flagging accounts that have moved significantly since the last check.

Touches: Accounts payable/receivable open items, payment run OData entities

Outcome: Gives a finance director a quick, readable cash position summary between formal reporting cycles rather than only at month-end.

Sales quote drafting from history

For a new customer quote, the agent finds pricing and terms from similar past quotes and drafts a first-pass proposal for the sales or finance team to adjust.

Touches: Customer quote OData entities, business partner and pricing data

Outcome: Speeds up quote turnaround by grounding the first draft in actual pricing precedent rather than starting blank.

Month-end variance commentary drafting

Using actual-versus-budget data by cost centre, the agent drafts a first-pass narrative explaining the largest variances, for the finance team to verify and finalize for the board pack.

Touches: Actual and budget/plan data by cost centre, general ledger account detail

Outcome: Cuts the writing time for month-end commentary meaningfully, since the numbers and likely drivers are already assembled rather than researched from scratch.

Business partner data quality check

Before month-end close, the agent flags likely duplicate or incomplete customer and vendor master records based on name, address, and tax ID similarity.

Touches: Business partner OData entities (customer and vendor master)

Outcome: Catches a class of master data issues before they distort AR/AP aging or duplicate payment risk reports.

Production or project order status Q&A

For manufacturing or project-services customers on ByDesign, the agent answers plain-language questions about production or project order status pulled from the relevant OData entities.

Touches: Production order or project task OData entities, depending on edition

Outcome: Gives a finance or operations user a quick status check without navigating to the underlying work centre view.

Contract and document semantic search

The agent indexes attachments linked to business objects, contracts, purchase orders, quotes, and lets a finance user search them in plain language rather than opening each record to find a specific clause or term.

Touches: Attachments linked to business partner, sales, and purchasing objects

Outcome: Turns document lookup from a manual search-and-open exercise into a direct answer to a specific question.

Reference architecture

Because ByDesign cannot be self-hosted, the AI layer connects entirely through its OData and SOAP communication scenarios and runs on infrastructure you control, with the model, retrieval index, and query log all sitting outside SAP's multi-tenant environment.

  1. 1

    ERP connectors

    Read access through ByDesign OData services and SOAP communication scenarios, each explicitly activated in business configuration and scoped to a technical communication user with defined, limited authorizations.

  2. 2

    Data and semantic layer

    General ledger, cost centre, business partner, and sales entities are mapped to finance-team language, and scheduled batch extraction is used for heavier analytics given ByDesign's OData call rate limits.

  3. 3

    Model serving

    An open-weight model served on your own infrastructure or in a private cloud tenant within the EU, sized for a mid-market finance and operations team's query volume.

  4. 4

    Retrieval and agents

    Retrieval-augmented Q&A over financial and business partner data, plus drafting workflows for variance commentary and quotes, all grounded in data pulled through the connectors above.

  5. 5

    Governance and audit

    Every answer and draft is logged with the OData entities it was based on. Access mirrors the business role assignment a user already has in ByDesign, and nothing writes back to ByDesign by default.

Integration notes for your ERP team

  • ByDesign only exposes data through OData and SOAP communication scenarios, there is no direct database access as with an on-premise SAP system, so every integration starts with activating and scoping the right scenarios in business configuration.
  • A dedicated technical communication user is created with authorizations limited to the entities the AI layer needs, mirroring the business role assignment a real finance user would have.
  • ByDesign's OData call rate limits mean scheduled, batch extraction is generally preferred over constant polling for heavier reporting and analytics use cases.
  • The Partner Development Infrastructure can support lightweight custom OData extensions where needed, but the integration should tolerate ByDesign's twice-yearly tenant-wide upgrades without breaking, since schema and API details can shift between releases.
  • No write-back to ByDesign happens by default; ByDesign's process guardrails are intentionally tight, and quotes, journal entries, and master data changes are drafted for review, not submitted automatically.
  • Attachments and documents linked to business objects are indexed for semantic search as retrieval context, not duplicated as a separate system of record.

Deployment options

Customer-controlled private cloud AI layer

Mid-market firms comfortable with cloud infrastructure generally but not with sending financial data to a public model API.

The model and retrieval index run in a private cloud tenant you control, connected to ByDesign over a secured OData channel, with no data sent to a third-party model provider.

On-prem AI layer over scheduled OData extracts

Firms wanting the model and data physically on their own hardware, even though ByDesign itself remains SAP-hosted.

A scheduled extract (hourly or nightly, depending on the use case) pulls OData entities into an on-prem semantic layer, so read-only natural-language Q&A does not require a live connection to ByDesign for every query.

Hybrid

Organizations wanting near-real-time operational Q&A alongside deeper periodic financial analysis.

Real-time OData queries handle operational questions like current order status, while a scheduled extract feeds the deeper financial analysis and variance commentary use cases.

Compliance and data control

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

GDPR / UK GDPR

Because the AI layer runs on infrastructure you control rather than a third-party model provider's, financial and business partner data used for reporting never leaves an environment covered by your own data processing agreement.

EU AI Act

Read-only, human-reviewed drafting and reporting outputs sit toward the limited-risk end of the Act's classification, and the governance layer's logging supports the transparency documentation a compliance review will ask for.

Data residency (ByDesign EU data centres)

Deploying the AI layer in the same EU region as your ByDesign tenant avoids an unnecessary cross-border data transfer between the ERP and the AI layer reading from it.

SOC 1 / SOC 2 (inherited from SAP's ByDesign hosting)

SAP publishes its own compliance posture for ByDesign hosting; the AI layer adds its own access logging and role mirroring on top rather than duplicating financial data into a separate, less controlled store.

Employee data handling

HR and payroll-related OData entities are excluded from the AI layer's scope by default and only enabled with explicit sign-off, consistent with how most European finance and HR teams want employee data handled around any new reporting tool.

How an engagement runs

Phase 1 . 2-3 weeks

Discovery

  • -Review of which OData and SOAP communication scenarios are currently active and which need enabling
  • -Business configuration and authorization scoping plan for a technical communication user
  • -Prioritized use case list, typically natural-language reporting and variance commentary first
  • -Deployment model decision (private cloud, on-prem extract, or hybrid)

Phase 2 . 5-7 weeks

Pilot

  • -Natural-language Q&A live over general ledger and cost centre reporting
  • -First-pass variance commentary drafted for one month-end close cycle
  • -Accuracy review against the finance team's own reports for the same period
  • -Go/no-go review with measured time saved on reporting and commentary drafting

Phase 3 . 6-8 weeks

Production

  • -Rollout to the full finance team and relevant operations users
  • -Quote drafting and business partner data quality checks added
  • -Documented runbook for handling ByDesign's twice-yearly upgrade cycle

Phase 4 . Ongoing

Scale

  • -Document and contract semantic search added
  • -Extension to additional business units or legal entities on the same ByDesign tenant
  • -Periodic review of OData scenario usage as ByDesign functionality evolves across releases

Questions to ask any vendor, including us

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

  1. Does any part of the integration send our financial data to a public model API, given ByDesign itself is already SAP-hosted cloud?
  2. Which specific OData and SOAP communication scenarios does the integration require us to activate, and how tightly are they scoped?
  3. How does the solution handle ByDesign's twice-yearly tenant upgrades without breaking the integration?
  4. Can the AI layer run in an EU region matching our ByDesign tenant's data residency?
  5. Does the system ever write back to ByDesign, submit a journal entry, change a master record, and if so how is that gated?
  6. How is variance commentary accuracy validated against our own finance team's judgment?
  7. Is HR or payroll data excluded from the AI layer's scope by default?
  8. What happens to OData rate limits if multiple users query heavily at month-end close?

Frequently asked questions

Can SAP Business ByDesign be run on-prem for AI purposes?

No, ByDesign itself is multi-tenant SaaS with no on-premise deployment option. What can run on-prem, or in a private cloud tenant you control, is the AI layer that reads ByDesign data through OData and reasons over it, which is the pattern described on this page.

Does this require custom development inside ByDesign?

Generally no. The integration activates and scopes existing OData and SOAP communication scenarios in ByDesign's business configuration. The Partner Development Infrastructure can support lightweight extensions where genuinely needed, but most use cases do not require them.

How does the AI layer handle ByDesign's twice-yearly upgrades?

The integration is built to tolerate schema and API drift between ByDesign releases rather than assuming a fixed structure, and the connector configuration is reviewed as part of ongoing support so an upgrade does not silently break a report or query.

Will this write journal entries or change master data automatically?

No, by default it does not. Quote drafts, variance commentary, and master data quality flags are all drafts a finance professional reviews and acts on through the normal ByDesign process; nothing writes back automatically.

Is our financial data safe from being used to train a public AI model?

Yes, that is the specific design goal. The model runs on infrastructure you control, private cloud or on-prem, and is never called as a public API, so your data is never sent anywhere a model provider could use it for training or anything else.

Can this work alongside other systems, not just ByDesign?

Yes, where a finance team wants to query ByDesign data alongside spreadsheets, a CRM, or other sources in one place, that broader mixed-source pattern is what DataRay is built for.

How long does a typical pilot take for a mid-market finance team?

A pilot covering natural-language reporting and one month-end variance commentary cycle typically runs five to seven weeks, timed to align with an actual close so the drafted commentary can be validated against real numbers.

Talk it through with an engineer who knows SAP Business ByDesign

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.