- Home
- Our services
- Enterprise AI · SAP
Data, Analytics & AI
Enterprise AI in the SAP ecosystem.
No generic AI: use cases that read your documents, understand your SAP master data and act within your transactions, under supervision proportionate to the risk. Our Casablanca AI Center of Excellence designs, measures, governs and industrializes them. And we do not stop at slides: we have built a demonstrator we can show you and adapt to your processes.
The problem
POCs everywhere.
Production, almost nowhere.
Generative AI demos impress, then stall. Without a connection to the core business system, without reliable data or a governance framework, a prototype stays a prototype. The AI that matters reads your real documents, understands your SAP master data and acts within your transactions, with controls in place. That work is as much engineering as it is models, and that is where we put our teams.
By process
The use cases worth the effort, domain by domain.
Enterprise AI is judged within a process, never in a catalog of tools. For each domain, the two or three use cases we know how to build, measure and integrate into the SAP core.
- Finance
Extraction and matching of supplier invoices (Document AI) · close support: detection of atypical entries and variance commentary · assisted clearing of incoming payments.
- Procurement
Compliance checks of purchase requisitions against the contract · spend classification and detection of off-contract purchases · drafting assistant for specifications, built from Ariba master data.
- Supply chain
Stock-out and supplier delay alerts with a recommended action · demand forecasting enriched with external signals · safety stock optimization by product family.
- Customer service
Qualification and routing of incoming requests · assisted responses based on customer history and S/4HANA data · detection of disputes at risk.
- IT & SAP run
Incident analysis and proposed resolution from the knowledge base · assistance with ABAP and BTP development and code review · monitoring of integration flows.
Joule & agents
Where an agent makes sense. Where classic automation is enough. Where nothing should be automated.
SAP now positions Joule and its agents on contextualized business workflows. The useful question is not “can we put an agent here?” but “is an agent needed here?”. We answer it before we build.
Relevant
When the situation varies from case to case, the decision requires cross-referencing several sources and the action remains reversible or supervisable: exception handling, incident resolution, purchasing or replenishment recommendations.
Classic, and sufficient
When the rule is stable and known: a workflow, a validation rule or a standard integration does the job better, cheaper and more safely than an agent. Fixed-rule matching and scheduled reminders fall into this category.
Automate nothing
When the stakes are irreversible, regulatory or human: approval of a payment above threshold, a credit decision, a labor-relations arbitration. AI prepares the file; the decision remains entirely yours.
Our accelerators
From assistant to agent.
Three families of accelerators, designed to anchor in your SAP core, from conversational answers to autonomous action, with document processing in between.
Conversational assistants
SAP Joule and Business AI Advisors grounded in your data: business users query the system in natural language (a figure, a status, an explanation) and get an answer founded on your master data.
Agents
Agents that detect an anomaly, recommend the action and trigger the transaction in S/4HANA, under explicit governance rules and a level of supervision suited to the use case.
Document processing
Automatic extraction, control and matching of business documents: supplier invoices, purchase orders, accounting records, all the way to posting in your SAP processes.
What we have built
A demonstrator, not a slide.
Our Casablanca AI Center of Excellence brings LLM, machine learning, RPA and SAP development skills together in a single team. It is the team that built our demonstrator: the extraction and matching of supplier invoices by Document AI, integrated into SAP S/4HANA. That last-mile mastery (S/4HANA integration, data quality, MLOps, change management) is what separates a mock-up from a use case that would hold up in production.
Under the hood
How it is built.
The pattern behind our demonstrator, described at architecture level. The mechanics are worth sharing, ready to adapt to your processes.
- The foundation
SAP S/4HANA (Cloud ERP Private or Public), with side-by-side extensions on SAP BTP: the business core stays standard, the intelligence lives alongside it and plugs in cleanly.
- The data
No reliable AI without data governance. SAP positions Business Data Cloud as the unified foundation for SAP and third-party data and for agentic AI; together with Datasphere, it is the foundation on which we anchor the use cases.
- The standard building blocks
SAP Joule and Document AI wherever the standard is enough. Reinventing a component the vendor maintains adds debt instead of capability.
- The custom engineering
Where the standard stops (agent orchestration, the business logic of detection), our teams develop, version and document, as with any other asset of your IT landscape.
- Human supervision
The level of human supervision is proportionate to the risk, the materiality and the confidence level of the use case: systematic validation on actions that commit the company, sample-based control where confidence is established, framed autonomy on reversible tasks.
- Traceability
Every agent decision is logged: what was seen, what was proposed, what was done, and by whom.
The field
What was hard.
Three workstreams took more than expected. First, making supplier data reliable: no model ever makes up for neglected master data. Then, setting confidence thresholds, that balance point between the agent that cries wolf and the one that lets things slip. Finally, keeping model drift in check over time, because an AI pipeline is maintained like a business system: continuously. Telling you this is part of the consulting.
Our approach
From prototype to industrialization.
Identify
Use cases with measurable value, prioritized with the business teams from your real processes — and the answer to the prior question: agent, classic automation, or nothing.
Prototype
A scoped prototype, on your data, with a quantified success criterion and a short deadline: feasibility is demonstrated in a few weeks.
Measure
Accuracy, straight-through processing rate, time saved, errors avoided: the value produced against the initial criterion, before any extension.
Govern
Supervision level, traceability, compliance, control of risks and drift: the framework that makes adoption possible.
Industrialize
MLOps, integration with the SAP core, data security, extension to real volumes, entities and languages, team training: a service that holds up under load over the long run.
The technology
The building blocks we put to work.
The demonstrator
Come see the system at work.
A demonstration by appointment, on the pattern described above, then a short workshop with our Casablanca AI Center of Excellence to identify your two or three highest-value use cases — and the ones better left unautomated.
Book a demonstration →