Can AI be used effectively for business process automation?

On the surface, artificial intelligence (AI) and automation might seem like one of these magical pairings. After all, if you can integrate AI and machine learning into your business to automate processes, you should be able to improve productivity, reduce errors, optimize resources and drive business outcomes. And who doesn’t want that?

At Celonis we agree that AI and automation make a great partnership. But there’s one element missing from the equation—a secret ingredient that can turn AI and automation into the super-powered duo of business process management.

AI and automation: A match made in process heaven

A whole industry is springing up around combining artificial intelligence and automation in the form of Intelligent Process Automation (IPA). The IPA market is expanding rapidly, at a predicted CAGR of over 13%, and is expected to be worth around $37 billion by 2030.

IPA involves the integration of AI with traditional automation techniques so that, instead of simply automating routine, repetitive tasks, businesses can bring cognitive technologies into the mix and explore automations that adapt to business needs. As well as following set instructions, IPA can learn, adapt and improve over time, essentially thinking for itself. This makes it a step up from Robotic Process Automation (RPA), where bots can’t think beyond the boundaries of their set task.

So what is the missing piece of the AI process automation puzzle that IPA needs to be really effective? We call it process intelligence.

Process Intelligence: The missing ingredient

Process intelligence allows AI to speak the language of your business. It combines data from innovative technologies like process mining with standardized process knowledge gained through years of process optimization experience, so AI has the intelligence it needs to understand your business processes end-to-end. The situational awareness enabled by process intelligence gives AI the power to adapt to business needs and drive smarter process automation.

As Etienne Kneschke, Executive Director Business Process Management at KARL STORZ SE & Co KG, recently explained: “Process intelligence is the foundational enabler of generative AI-based automation in enterprises, as it feeds LLMs with the essential contextual understanding they need, in order to know how the business runs across systems, departments, and regions.”

Without the context provided by process intelligence, AI and automation can be a risky combination. No matter how advanced the machine learning algorithms are, if they’re learning from the wrong data they can end up automating sub-optimal processes and ultimately making the situation worse. And, without understanding operations end-to-end, automating one business process with new AI tools can have unintended consequences elsewhere in the organization.

Celonis delivers process intelligence through the Process Intelligence Graph, which extracts data from all your business systems so you can see how objects and events interact, how processes are interconnected, and how your business runs. It layers in standardized process knowledge and AI to form a connective tissue for your business.

Read more about how Celonis is applying standard process knowledge to power AI automations across the ecosystem.

Powering AI process automation

When AI is fed with data that has been contextualized with process knowledge, it can power intelligent automation using a variety of tools. The Process Intelligence Graph, for instance, is tech agnostic and can trigger automations in whatever tools are already available in the customer’s tech ecosystem. For Microsoft users this might be Power BI AI and Power Automate, while for others it might be an RPA tool like UIPath.

In addition to amplifying the effectiveness of any existing automation technology, Celonis has its own AI capabilities and automation solutions:

• Action Flows are used to define and orchestrate process automations. They are fully connected to the Process Intelligence Graph and can be set up in minutes through an intuitive, low-code interface with premade connectors.
• Machine Learning Workbench (MLWB) provides integrated and embedded machine learning capabilities. Fully hosted and managed, MLWB enables Celonis customers to create custom models.
• LLM for PQL Generation is a generative AI (Gen AI) tool that turns user queries into Process Query Language (PQL), the language used to turn process data into process intelligence.
• Process Copilot is an evolution of LLM for PQL Generation that uses Natural Language Processing (NLP) and conversational AI to make interacting with the Celonis platform as simple as chatting to a colleague.

AI and automation in action

Process intelligence, AI and automation can work together to drive new levels of observability and actionability. At present, intelligent automation use cases tend to be reactive, detecting and resolving issues in real time.

In the future, they’re likely to be more proactive, helping businesses take action to avoid or resolve problems before they even arise. This will be supported by technologies that help businesses see how processes operate and interact across the organization, such as object-centric process mining (OCPM).

Examples of reactive and proactive use cases are outlined below.

An example of reactive intelligent automation

Through process intelligence, a business discovers it is paying a significant number of invoices twice. It uses AI reporting tools to detect invoices that are exact or approximate duplicates. It then triggers an automation to block the duplicate payment before the money goes out the door, and alerts a member of the accounts payable team to investigate the situation.

This approach has been highly effective for global software company Autodesk. Operational dashboards for its accounts payable function are focused on fraud detection and use both the Machine Learning Workbench and the Celonis Duplicate Invoice Checker App to detect and resolve compliance issues, such as employees paying themselves.

With reactive use cases, AI can trigger automation scenarios to react faster than any human could – 24 hours a day – which is ideal for delivering enhanced customer service as well as detecting fraud and combating human error.

An example of proactive intelligent automation

Talking to Acceleration Economy at Celosphere 2023, Celonis lead transformation evangelist Rudy Kuhn outlined a potential use case for proactive automation. He explored how a high-end automotive brand that manufactures vehicles to order, might act if delivery of the leather for a car’s seats is delayed by a week.

Instead of simply informing the customer that their car will be ready a week late, which would be a reactive use case, the manufacturer could automatically re-plan and reprioritize the entire manufacturing process to ensure the vehicle is available on the agreed date. An AI-supported system can enable this type of proactivity by learning from past situations, predicting what will happen next, and triggering action to ensure the delivery date is achieved.