Operationalize AI by Blending Deterministic and Dynamic Process Orchestration

It’s clear that, like the introduction of the Internet and mobile phones, AI will have a profound impact on how organizations:

• Serve and interact with their customers and users
• Design and deliver products and services
• Collaborate with their partners and suppliers
• Enable and empower their employees

What isn’t changing is the importance of process. Business processes define how you deliver products and services to your customers; how you work with your suppliers and partners; and how your employees get things done. AI will change the way we build, run, and improve business processes, but the processes themselves won’t disappear.

That’s why process orchestration is a critical component of AI adoption and strategy. The steps that make up a business process are executed by people, systems (including AI), and devices—we refer to these as endpoints. Process orchestration ensures that even the most complex series of steps are executed correctly across all endpoints. In turn, that ensures customers, suppliers, partners, and employees have a smooth experience when they interact with or are affected by processes.

Implementing AI-powered automation without making it part of a process orchestration strategy means you run the risk of creating disconnected automations that lead to operational inefficiencies and compounding technical debt. According to Accenture:

Generative AI and AI are now the highest contributors to a company’s tech debt along with enterprise applications, according to our digital core research. This trend will likely exacerbate: in our Pulse of Change survey, 52% of organizations said they plan to allocate more funds toward generative AI, heading into 2025.

– Accenture, Build your tech and balance your debt, 22 October 2024

Also, organizations in highly regulated industries are naturally concerned that the use of AI will disrupt carefully designed policies and procedures that ensure business processes comply with government and industry regulations.

The research that Camunda commissioned for the 2025 State of Process Orchestration and Automation Report shows that 84% of IT and business leaders are looking to add more AI capabilities over the next three years. In addition, 93% of respondents agree that AI applications and services will need to be orchestrated across business processes in order to get the maximum benefit from investment in AI.

These leaders aren’t taking a “wait and see” approach to AI adoption, but they’re aware of the dangers of a siloed AI strategy.

Let’s take a look at how process orchestration can help you leverage AI today, prepare for the AI innovations of tomorrow, and manage the potential risks of AI adoption.

Deterministic process orchestration

Deterministic process orchestration has been the dominant approach to process automation for many years. Deterministic means that what will happen each time an instance of the process is executed is determined ahead of time, usually by a business process model defined in a language such as Business Process Model and Notation (BPMN). Given the same inputs, configuration, environment, and conditions, the process will always follow the same sequence of steps and yield the same results.

There are many benefits to deterministic process orchestration:

• The process is predictable and reproducible, and can easily be audited.
• The use of a process model makes it easy for teams to visualize, explain, and collaborate on every aspect of a process’s design.
• If the process model is defined in BPMN, it’s self-documenting: what you see is exactly what will be executed.
• The cost of executing the process (for example, measured in hardware resource consumption) is relatively low, and execution is typically quite fast,

However, there are also downsides to deterministic process orchestration:

• To build a process model, you have to truly understand and agree on what should happen in a business process. Sounds simple, but many organizations struggle with inconsistent or poorly documented processes.
• Building the process model itself takes time and requires testing.
• It’s difficult to deliver highly personalized customer experiences if doing so means maintaining many different variations of a particular model.

Dynamic process orchestration

The recent explosion of widely available AI tools and services has many process specialists wondering, Is there another way? Can AI get us to a world of dynamic process orchestration?

Dynamic, also sometimes referred to in this context as non-deterministic, means that there isn’t a process model or any other artifact that defines exactly what will happen each time a process instance is executed. Instead, what happens is determined as the process instance is executed, leveraging AI models and runtime data.

Dynamic process orchestration isn’t common—yet—but we can already imagine some of the benefits it could offer:

• Faster time to value because the process design and testing phase would be dramatically reduced
• Better enablement of business stakeholders who want to automate processes but lack the skillset or IT resources to do so
• Easier delivery of personalized customer experiences based on data about a particular customer’s preferences and needs

In a deterministic world, every part of a process must be planned and modeled. If there are aspects of a process that can’t be determined beforehand, organizations rely on knowledge workers to step in and handle the work.

For example, in case management scenarios such as insurance claim fraud investigation and customer complaint handling, it’s hard to foresee exactly what will need to be done to resolve the case. You typically need a qualified person to look at the case, figure out what needs to be done, and sometimes, actually do the work. Dynamic process orchestration offers a way to increase the level of automation in these hard-to-define processes, freeing up knowledge workers’ time for other tasks.

But dynamic process orchestration has its downsides, too:

• There isn’t a process model that you can point to as the source of truth for what will happen in a business process.
• Auditing processes is more difficult, especially if you don’t have a record of the reasons an AI took certain actions.
• It’s hard to predict what results an AI will produce, even when given the same inputs, configuration, environment, and conditions.
• The cost of executing the process is relatively high and can be vulnerable to latency issues because the AI model is constantly being invoked.

Let’s talk about AI agents

You may have heard terms such as AI agents or agentic AI. Understanding agents is key to understanding the future of dynamic process orchestration.

According to Google’s recent Agents whitepaper:

… a generative AI agent can be defined as an application that attempts to achieve a goal by observing the world and acting upon it using the tools that it has at its disposal. Agents are autonomous and can act independently of human intervention, especially when provided with proper goals or objectives they are meant to achieve. Agents can also be proactive in their approach to reaching their goals. Even in the absence of explicit instruction sets from a human, an agent can reason about what it should do next to achieve its ultimate goal.

– Julia Wiesinger, Patrick Marlow, and Vladimir Vuskovic, Agents, September 2024

The autonomous nature of agents makes them ideal for executing a business process in which the process logic isn’t predefined. However, as my colleague Niall wrote in his blog post about AI agents and orchestration, the issue with agents is that they don’t produce predictable, repeatable results, which is what most organizations need them to do if they’re going to rely on agents to replace deterministically defined processes.

The best of both worlds: blending deterministic and dynamic orchestration

How can you gain the benefits of dynamic process orchestration without sacrificing the benefits of deterministic orchestration and—more important—without putting customer experiences and regulatory compliance at risk? The answer is blending deterministic and dynamic approaches in the same process, which is exactly what process orchestration enables you to do. When the dynamic parts of a process are executed by one or more AI agents, we call this agentic orchestration.

Imagine a common business process, such as a Know Your Customer (KYC) process. It involves a very well-known and highly regulated series of steps like:

• Collecting information such as the potential customer’s full name, date of birth, and home address
• Verifying that information using government-issued IDs and databases
• Checking whether the person appears on global or local sanction lists or is a known financial criminal
• Assessing the risk level associated with the person and, in some cases, performing additional due diligence
• Conducting ongoing monitoring of the person’s financial transactions and reporting suspicious activity to the relevant authorities

AI technology, such as optical character recognition (OCR), is already common in KYC processes—for example, it’s often used to extract data from photos or PDFs of ID cards and passports. This saves time and money, but OCR tools aren’t capable of making decisions or being proactive in the absence of human intervention like AI agents are.

An AI agent can enhance a KYC process by taking proactive actions such as:

• Guiding a person through the KYC process, responding to their questions, and dynamically determining how to solve issues such as an unreadable ID or a mismatch in personal data
• Monitoring for KYC policy changes, updating the KYC rules that human knowledge workers use, and changing the way the agent executes the process accordingly
• Automatically adjusting a person’s risk level based on ongoing monitoring and taking action such as restricting their bank account activity or dynamically adjusting their spending limits

A realistic outlook

Have AI agents reached a level of maturity where agentic orchestration is realistic for the processes that are most fundamental and important for your business? And is your organization ready to embrace agentic orchestration?

Forrester provides a realistic outlook in their Predictions 2025: Automation report, which says that in the coming year:

GenAI will orchestrate less than 1% of core business processes. GenAI will affect process design, development, and data integration, reducing design and development time as well as the need for desktop and mobile interfaces. Business users will develop initial workflows, create forms, and visualize the process. But this genAI efficiency still leaves current digital and robotic process automation platforms orchestrating the core process, subject to their deterministic and rule-driven models. To prepare for 2025, recognize that deterministic automation will remain in control of the core long-running process, while AI models will support bursts of insight and efficiency.

— Forrester, Predictions 2025: Automation, 22 October 2024

Now is the right time to ensure your process orchestration strategy is well-defined and your most critical processes are being orchestrated by a platform that allows you to blend deterministic and dynamic execution.

Keep in mind that you don’t have to wait for new use cases or business processes to surface before you think about how to enhance them with agentic AI. Take a look at your existing processes and consider how you might blend deterministic and non-determinist orchestration to achieve a better outcome. Doing so will enable you to experiment with agentic AI today and be prepared to leverage AI increasingly in the future.

Learn more

To learn more about the future of process orchestration, check out our blog post, Prepare for Autonomous Automation with Agentic AI, which breaks down “Gartner Predicts 2025: The Future of Automation is Autonomous,” and where you can get your own free copy.