Everyone Now Sells a Semantic Layer for Processes. Who Makes It Portable?

Image: Celonis, from the Context Model launch press release (May 2026).

For: BPM, enterprise architecture, process mining and transformation leaders Window: 6 to 25 October 2026 Data: every signal behind this briefing, with links to the original sources, is on our BPM Trends page

The question used to be whether processes need a semantic layer for AI. In October the market answered it. SAP, Celonis, ARIS and Microsoft each now present their own layer as the place where AI agents learn how your business works. The open question has moved: how does that knowledge stay consistent across the several platforms most enterprises run, and who governs it?

Our BPM Pioneer agent has collected more than 3,000 signals from 23 publishers since summer, 1,351 of them reported by two or more sources. "Semantic Layer & AI Context Infrastructure" is now one of the largest topics we track, with 215 signals, and total weekly volume roughly doubled in October: 260 signals in the week of 12 October, against 100 to 120 per week through September. The charts, cluster radar and full feed are on BPM Trends.

1. Process context now has a number

The 2026 Process Context Study by The Hackett Group and ARIS surveyed more than 200 senior leaders at Global 2000 companies. Organisations with strong process context were five times more likely to report successful AI outcomes. 86% said AI agents cannot be deployed reliably without process context, yet only 22% have real-time, end-to-end visibility of their process flows, and 59% describe their visibility as fragmented across systems (AI Magazine summary).

Read it as directional: it is a vendor-co-sponsored survey of self-reported success. But the direction matches what we see in projects. The bottleneck for agents is rarely the model. It is whether your processes, roles, controls and systems are captured in a form an agent can reason over.

2. Four vendors, four context layers

• SAP grounds its Joule agents in the SAP Knowledge Graph, together with process context from Signavio and landscape context from LeanIX (TechTimes on SAP's Q3 and Joule). On 6 October SAP also agreed to acquire TechWolf, whose "context graph for work" maps tasks and skills and will become a core of SuccessFactors (Nasdaq, CIO).
• Celonis launched the Context Model in May as a "context layer" between data platforms and agent platforms, built on its Process Intelligence Graph. In October it is visibly scaling go-to-market around it. Celonis describes the model as open through APIs and MCP.
• ARIS announced Process Context for Enterprise AI on 20 October, available from 26 October. Its MCP interface gives agents access to processes, roles, systems, rules, risks and controls held in ARIS (AI Magazine).
• Microsoft made an ontology the core of Fabric IQ: entities, relationships, rules and actions that agents use as their shared business language, now also reachable via MCP (Enterprise DNA).

The data platforms are moving the same way, from Databricks to a widely cited Gartner prediction that universal semantic layers will be critical infrastructure by 2030.

The practitioner read: each of these layers is strong at what its platform sees, and each is "open" in the sense that agents can call it. That is good for customers. But calling a layer is not the same as carrying its meaning elsewhere: MCP access lets an agent ask; it does not let you take the definitions, rules and cross-tool identities to another platform unchanged. Most enterprises run several of these platforms at once, often during a multi-year transformation, so the knowledge that matters most is the knowledge that connects them.

3. The open standards are ready, for part of the problem

• Graph queries. GQL (ISO/IEC 39075) was published in April 2024 as ISO's first new database language since SQL in 1987. What changed this autumn is adoption: Neo4j's Cypher is converging on it, which lowers the cost of switching graph engines.
• Semantic interchange. The Open Semantic Interchange project was accepted into the Apache Incubator as Apache Ossie, backed by more than 50 organisations including Snowflake, Salesforce, Databricks and dbt Labs. It now calls itself a specification for "semantic layer and ontology", with working groups for metric language, catalog and ontology (Dremio on the Apache move).

Image: Apache Software Foundation, Apache Ossie.

Ossie's vocabulary today is metrics, dimensions and business concepts for data and BI. It has no notion yet of an end-to-end scenario, a role in a lane, a control governing an activity, or the application an activity runs in. That process and architecture layer is where the gap is.

• Meaning and rules. The W3C standards that can fill that gap are not new: RDF for exchange, OWL for meaning, SHACL for validation. They are mature, vendor-neutral and readable by many tools. They have simply rarely been applied to BPM and EA repositories.
• Execution data. On the process mining side, object-centric event data (OCEL 2.0) and event knowledge graphs keep maturing, giving executed processes a graph shape that can connect to the designed ones.

4. Continuity matters in a moving market

The Munich I Regional Court ruled that one feature of SAP Signavio Process Intelligence infringes a Celonis patent and barred SAP from offering it in Germany (CIO, Computerwoche). SAP stresses that only a single feature is affected, that the product remains available, and that it will appeal; it is also challenging the patent's validity. Customers do not need to panic. The broader lesson is about continuity: product features, licences and roadmaps change, sometimes for reasons nobody planned. The same is true one layer down, where graph databases are shifting to source-available licences (SurrealDB example). Business knowledge that is documented once, in a form that survives such changes, is cheap insurance.

Five questions before you commit to a context layer

1. Where is meaning defined? In files you own and version, or only inside one platform?
2. Can rules be checked outside the tool? Can you prove that your repositories meet your modelling conventions?
3. Does it span your landscape? ARIS or Signavio, LeanIX, Cloud ALM, Jira and Celonis together, not only one platform's estate.
4. Who approves what agents write? Agents should propose changes that pass the same rules and the same steward as a human modeller.
5. What happens in a migration? Do owners, approvals, control links and history survive the move?

Our take: not another tool

We work with ARIS, SAP Signavio, SAP LeanIX, SAP Cloud ALM and Celonis every day. Their context layers will make AI agents better inside each platform, and we help customers get the most out of them. What we keep seeing in transformation programs is a different gap: the knowledge between the platforms is unconnected and ungoverned.

Business Flows is how we close that gap. It is not a sixth platform. It is our method, reference content and a shared knowledge core that the customer owns, built on open W3C standards and connected to the tools already in place. Where it pays off first:

• Transformation impact analysis. Replacing an application, changing an SAP process or introducing a new operating model? See every affected process, role, requirement, control and test across all repositories in one view. Fewer missed dependencies, faster change planning. This is usually where we start.
• Governed business context for AI agents. One set of approved definitions that agents on any platform can use, with every agent-proposed change reviewed like a human's. Less duplicated integration, more reliable answers.
• Repository migration and coexistence. ARIS to Signavio, a merger, or several repositories running side by side: identities, mappings and governance metadata are preserved instead of rebuilt.
• Continuous repository quality. Inconsistent ownership, obsolete references, conflicting application assignments and missing controls are found automatically, with source and owner.

We publish our approach openly, down to the governance rules and the design trade-offs, because a knowledge core the customer owns should be one the customer can inspect.

Read the deep dive: Your Business Knowledge, Connected. It walks through the four use cases, explains how we deliver them alongside your existing tools, and, for readers who want the engineering, how OWL and SHACL make it work.

Facing a live change, a migration or an AI rollout? Write to sales@bpexperts.de, and we will start with the question already on your desk.

This is a market-signals briefing compiled from monitored public sources; the underlying signals are listed on BPM Trends. It is shared for the BPM and transformation community and is not investment, legal or vendor-selection advice.