Composable, process‑aware AI key to enterprise ROI
Effective AI systems are more than high-powered question and answer machines. While the ability of some AI applications may seem to be almost magical, the reality is often far different. An AI tool, whether that's a chatbot or a generative tool, that doesn't understand the context of the questions it's asked and the data it uses, will deliver poor outcomes.
Rudy Kuhn, the Lead Evangelist for leading process intelligence company Celonis, says that composability in enterprise AI is critical for ensuring scalable, governable and resilient AI.
"Composability in enterprise AI means building systems from modular interoperable components rather than relying on monolithic end-to-end black boxes. In practical terms, it means separating data and process context, separating AI models and agents, orchestration and policy layers, and governance and monitoring. Instead of embedding intelligence into one massive system, you compose it."
While this may add complexity to the initial build and deployment of an AI system or application, Rudy likens it to building a model with Lego. On one hand, opening a box with hundreds of bricks of different shapes and sizes may be complex. But it also offers greater flexibility and the capacity to change and move away from the original plan as circumstances change.
Rudy adds that the speed of geopolitical and business change and their unpredictability mean that monolithic systems are not fit for today's purpose. Flexibility is crucial in his view.
"Many organisations deploy AI without grounding it in how work happens. Without process, context, governance, and measurable consequences, AI becomes an expensive experiment rather than a value engine," Rudy says.
The consequences of not using a composable AI approach can be significant. A recent presentation by market analyst firm Forrester looked at the risks created by AI projects and the consequences of failed AI projects. One of their key findings was that a lack of situational awareness was at the root of many failed projects. But they said over a third of those failed projects could be rescued by injecting process intelligence into those projects.
By building AI applications through the lens of composability, it's possible to tackle the challenges of governance and accountability at the design stage rather than focussing purely on the outputs of the system.
"In a composable architecture, governance becomes embedded. Policies, constraints, identity management, and performance metrics are integrated into the orchestration layer," explains Rudy. "Autonomy must be bounded. In a composable system, accountability is not an afterthought. It's structurally built into context, constraints, and consequences."
This is why process intelligence is so critical according to Rudy. Process intelligence evolved from process mining where all the data used in a process is analysed and understood. Process intelligence takes this further by adding context and other supporting information to the data. This, in turn, can be used in the development of AI systems to ensure creator contextual awareness of inputs and outputs.
AI needs to have context of where users are in a process and have constraints. It needs rules and it requires consequence. The three Cs of context, constraints, and consequence are critical. Rudy stresses that process intelligence provides all three.
"Without process intelligence, AI operates in isolation," he says.
For organisations running older systems, Rudy says it's wise to avoid the temptation to avoid a 'black box' approach to AI. By separating decision logic from orchestration enables organisations to leverage legacy systems while adding policy‑enforcement and process‑intelligence layers. He says real‑time "situation awareness" and continuous monitoring can alert when performance falters or anomalies arise. This approach is auditable – an essential feature for regulated sectors such as financial services where transparency and traceability are crucial for building customer trust and avoiding regulatory issues.
To bring this all together, Rudy says a manageable, composable AI system hinges on five design principles.
"You must separate the data, model, orchestration, and governance layers so each evolves independently. Adopt a process‑aware architecture that embeds business context into every AI interaction and add a deterministic control layer that enforces policies and constraints at runtime, preventing black‑box autonomy. Ensure you have continuous monitoring and feedback loops that capture telemetry and audit trails for ongoing refinement and link model performance directly to process metrics and ROI."
With many organisations looking to ensure they deploy responsible AI, Rudy says this model ensures every step of the AI journey, from conception to execution, can be shown to deliver value and not create security issues or produce incorrect outputs.
"A composable AI system distributes intelligence while centralising governance. It makes every decision traceable to process context, policy constraints, and measurable outcomes. Responsible AI is not about ethics statements. It's about architectural choices," he says.
Rudy's interest in AI extends beyond typical enterprise use cases. His song "Love at Second Sight", performed with an AI-generated voice, appears to be a classic love story. In reality, it reflects a central idea of his work: intelligence alone is not enough. Without context, even AI can misunderstand what truly matters.
One of his colleagues brought it to life with a music video and vocalist.
Composability in AI that leverages process intelligence is the foundation for successful AI projects. Organisations cannot hand over the keys to their AI kingdom to an inscrutable black box. By layering each component it's possible to deliver ROI and support good governance and transparency.