SAP · Automation, AI & the Intelligent Enterprise
Automation that runs inside the process, and intelligence that has to earn its place there.
Noble applies automation and AI where an SAP process actually runs: workflows and process automation on BTP that orchestrate standard transactions, document intelligence that reads an invoice or a delivery note before a person has to, SAP’s AI services and Joule applied to a task rather than demonstrated, and machine learning on the operational record — each with a defined boundary between what the system recommends and what a person approves.
Where automation belongs, and where it does not yet
The intelligent enterprise is SAP’s phrase for a system that does not wait for a person to notice. Most of what makes an organisation slow inside SAP is not the transaction — it is the approval that waits in an inbox, the invoice that is keyed by hand from a PDF, the exception that sits in a queue until somebody remembers it. Workflow and document intelligence remove that waiting, and they are deterministic enough to be trusted to act.
A model that predicts — which supplier will deliver late, which invoice is a duplicate, which asset is about to fail — is a different kind of thing. It is probably right, and the organisation has to decide what "probably" is allowed to do. Noble’s rule is that a prediction is advisory until its record in production says otherwise: it suggests, a person approves, and the approvals are measured until the boundary can be moved with evidence. That boundary is designed, not defaulted, and it is the first decision in every engagement on this page.
The advisory boundary
A loop, because it never finishes: the model suggests, a person decides, the decision is recorded, and the record is what eventually justifies letting the model act alone on the cases it has proved itself on. The attended stage is the one a person holds.
- 01Event in the processInvoice, order, reading
- 02Model suggestsWith its confidence
- 03Person decidesApprove, correct, reject
- 04Transaction postsStandard SAP
- 05Decision recordedFeeds the next review
What the service covers
Workflow and process automation
Approval, exception and escalation flows built on SAP Build Process Automation and the workflow capabilities of BTP, orchestrating standard transactions rather than replacing them. A flow is designed against the process as it is actually done, with the delegation of authority encoded once and the inbox it lands in chosen by the person who has to act.
- Approval and exception workflows on SAP Build Process Automation
- Delegation of authority encoded once, reused everywhere
- Escalation with a clock and an owner
- Task inboxes designed for the person acting
Robotic automation, used sparingly
Screen-level automation for the steps that cannot yet be reached through an interface — a legacy system with no API, a portal that only has a browser. It is built where it removes real keying, and it is retired the moment a proper interface exists, because a robot on a screen is a debt with a schedule.
- Screen automation only where no interface exists
- Every robot registered with its retirement condition
- Replaced by an API or event as soon as one is available
Document intelligence at the point of entry
Invoices, delivery notes, purchase orders and forms read by a model before a person has to — with SAP Document AI where the process is inside SAP, and with the AI service’s document pipeline where it is not — and posted to the standard transaction with the extraction confidence attached. Low-confidence fields go to a person; the person’s correction goes back to the model.
- Invoice and document extraction with SAP Document AI
- Confidence carried to the posting, low confidence routed to a person
- Arabic and English documents, in one pipeline
- Corrections returned to improve the next read
SAP’s AI services and Joule
SAP’s own AI — Joule where it is available in the applications the organisation runs, and the AI services on BTP for what a process needs — applied to a specific task with a specific owner. Noble’s work is to find the task where SAP’s AI is genuinely the right tool, configure it to the process, and ground it in the organisation’s data through the governed layer rather than through a copy.
- Joule enabled and scoped in the applications that carry it
- AI services on BTP applied to a named task
- Grounded in governed data, not in an extract
- Data residency and access kept inside the SAP boundary
Machine learning on the operational record
Prediction built on the reconciled record: late delivery, duplicate payment, demand by material, the asset likely to fail next. Built as advisory first, with the suggestion, the confidence and the person’s decision recorded so the boundary can be moved on evidence. The model side — training, evaluation, monitoring — is the AI service’s work, done with this one.
- Prediction on procurement, finance, supply chain and asset data
- Advisory by default, with the decision recorded
- Served into the SAP screen where the decision is made
- Trained and monitored with the AI service
The boundary, and who holds it
Every automation on this page has a written answer to three questions: what it may do without a person, what it must show a person first, and who is accountable when it is wrong. The answers are configured into the workflow, logged with the transaction, and reviewed on the same cycle as the authorisation model — because an automation that can post is a user, and is governed as one.
- Automations governed as users, with roles and review
- Every automated posting attributable to its automation
- A person named as accountable for each boundary
Where this connects
- BTP, Integration & ExtensionsWorkflow, document intelligence and AI reach the core through the released interfaces and the BTP services described there.
- Data, Analytics & PlanningPrediction starts from a reconciled, governed record. The layer that page builds is what the models here are trained on.
- AI for Enterprise SystemsThe model side of this subject — retrieval, agents that act through APIs, evaluation and monitoring — is the AI service’s page. The two are written once, from opposite sides.
The services alongside
- AI & Machine LearningModel development, evaluation, monitoring and the engineering that keeps a model honest in production. The SAP work here is applied; the model work is theirs.
- Digital TransformationProcess automation across the estate, beyond SAP — where the same workflow crosses a system SAP does not own.
SAP BTP, SAP Build Process Automation, SAP Document AI and Joule are products of SAP SE, named to identify what the work is done on. All marks belong to their owners.
Name the queue.
The place where work waits — an inbox, an exception list, a stack of PDFs — is where automation earns its place first. Bring the queue, and the decision it is waiting for.