Enterprises are deploying AI agents faster than they can teach them how the business actually works.
An agent assigned to approve a purchase order, screen a customer, or close the books doesn’t fail because the model isn’t capable. It fails because it doesn’t know your rules: the approval threshold for EMEA orders above €100k, the exception that requires board sign-off, the compliance step that is never optional. That knowledge rarely lives anywhere an agent can reach it. It is scattered across 500-page policy documents, SharePoint sites, Confluence pages, email threads, and the heads of the three people who have been there long enough to remember why the rule exists.
So teams do the only thing they can: they hand-curate a bundle of documents and paste it into each agent’s prompt. It works for a demo. Then a rule changes, the bundle goes stale, two departments encode the same policy differently, and nobody discovers the drift until an audit does. Every new agent starts the same context-gathering from scratch. The promise of the agentic enterprise stalls, not on the agents, but on the fractured, static knowledge underneath them.
This is the gap SAP Company Memory preview is built to close.
A governed knowledge layer, not another document store
SAP Company Memory is a governed knowledge layer that delivers an organization’s business rules, policies, and institutional knowledge to AI agents and to people at the moment of decision.
The difference from a wiki or knowledge base is structural. Instead of storing documents and hoping consumers can find the right passage, SAP Company Memory turns scattered rules into atoms: small, consistent, retrievable units. Each atom carries three things:
- Purpose: why the rule exists
- Scope: when it applies (region, role, process, system)
- Rule: the exact governing statement
Every atom keeps a citation back to its source document and page, so any answer can be traced to its origin. Atoms carry hierarchy too. A global sanctions rule, a company-wide policy, a regional exception, a customer-specific term can all coexist, and the right one surfaces for the right context, the way a well-briefed employee would apply it.
Because the knowledge is structured once and governed centrally, it can be consumed many ways:
- Agents connect at runtime over MCP and retrieve exactly the rules relevant to the decision in front of them, with no manual SOP-to-prompt translation and no vendor lock-in.
- Developers query and export structured rule sets through a CLI at build time.
- People ask questions in plain language and get a specific, cited answer back.
And when rules collide, for example “orders below €20 need approval” in one document and “orders below €20 don’t” in another, SAP Company Memory scans the whole corpus, surfaces the conflict, and walks an owner through resolving it before any agent ever consumes a contradiction.
The same layer that governs agents answers your people’s questions too
Here is the part that surprises teams: the knowledge you structure to make agents reliable turns out to be exactly what your people have been missing.
The rules buried in that 500-page document weren’t only invisible to agents. They were invisible to the new hire on day three, to the process owner trying to work out what breaks if a policy changes, to the employee who asks a colleague and still gets it wrong. Make the knowledge explicit, governed, and queryable, and the human side lights up:
- Instant, cited answers to policy questions. “What’s our approval threshold for EMEA orders above €100k?” returns a specific atom with its source, not a link to a PDF to go read.
- Onboarding in days, not months. New hires query SAP Company Memory instead of shadowing colleagues to absorb tribal knowledge.
- Impact analysis before a change. “What breaks if we change this rule?” is answered in seconds, not weeks of manual cross-referencing.
- Authoring by the people who own the rules. Business owners write and maintain rules in natural language, rather than filing a ticket with a technical team.
- Knowledge that survives people leaving. Institutional memory becomes explicit and retrievable instead of walking out the door.
The two audiences reinforce each other. Humans author and validate the rules; agents consume them at runtime; what we learn from execution feeds back to the people maintaining them. The knowledge layer gets more valuable the more it is used.
The foundation is already in place
A business-rules layer for agents is only as good as the rules that feed it. The hard part is not building the retrieval layer; it is cold-starting a corpus of real, validated enterprise rules that agents can actually rely on.
SAP enters this with a foundation already assembled:
- Process models that specify how a business should run, built from two decades of enterprise process design in SAP Signavio.
- Execution data from business process intelligence and S/4HANA that shows how it actually runs, so rules can be grounded in reality, not just asserted.
- Domain expertise packaged across thousands of SAP reference models and content packages.
- The business owners themselves, already working inside the tools where this knowledge lives.
Where we are, and where we’re headed
SAP Company Memory is in the hands of selected design partners today, through an early-access program running on SAP Signavio infrastructure. We are deliberately building this with customers rather than announcing it finished; the beta partners are shaping ingestion, governance, and the human query experience as we go.
Broader availability follows as that validation matures. We will share specifics on timing as we get closer; for now, the direction is clear and the foundation is in customers’ hands.
What we are building toward is a self-enriching organizational memory: a knowledge layer that humans trust for answers, agents rely on to execute correctly, and systems draw on for consistency, one that gets sharper every time it is used.
Want to follow along, or get involved?
If your organization is deploying AI agents into business processes and running into the knowledge gap described here, we’d like to talk. We’re expanding the design-partner program and are especially interested in teams working through agent governance, onboarding, and compliance.
Want to know more? Feel free to reach out!
Further readings:
- SAP Company Memory
- Process atoms
- How Process atoms differ from business rules
- Process atoms whitepaper
Read More Technology Blog Posts by SAP articles
#abap