Next article
Back
Process intelligence takes LLMs to the next level – here’s how
It’s unlikely, but not impossible, to go viral at 70 years old. That’s what happened to Artificial Intelligence in late 2022 with the launch of ChatGPT. Seemingly overnight we all went from AI-adjacent to AI converts: testing out the tool’s ability to summarize content, write jokes, and come up with first-date ideas.
Suddenly teams everywhere tried to put AI to good use. So, what’s AI done for you lately at work? Likely answer is ‘not enough’. But with process intelligence, you get smarter large language models that make AI work better, faster, stronger.
Whether it’s suggesting optimal lead times to a materials planner or automating next steps for supply chain transformation, process intelligence empowers enterprises to make decisions based on data and take action knowing what value it will bring. It's AI that knows how your business flows. Every system, every department, every process.
What process intelligence can make possible – three examples
Tighter compliance
Process intelligence informs the large language model (LLM) about company rules so your conformance checking can be watertight. If teams are breaking SLAs like late deliveries, late invoice payment, or not closing tickets in set times, you can deploy process-intelligent AI to complete the task according to the rule – as well as suggest corrective actions to those responsible. The same applies to cases where compliance consequences may be greater, such as shipping goods without an invoice.
Fewer process breakdowns
Process intelligence makes the LLM aware of common causes of process breakdown – based on your company’s past activity as well as comparing it to industry best practices. With process intelligence, you can build AI solutions that tell you when the next activity in the process will occur and predict what is going to happen next. Beyond predictions, you can also get advice on how to prevent the breakdowns happening in the first place.
Work with counterfactuals
Pairing process intelligence with object-centric process mining, you could create models for counterfactual situations i.e. “How would my SLA improve if I increase the capacity of my Shared Services Center?’ Or, ‘If I remove unnecessary steps from my Order Management process, how much time will I free up for the team?’ With process intelligence, planners could build an LLM assistant that rapidly responds to changes and sends suggestions – as well as runs analyses of the potential impact of decisions.
Process intelligence puts typical LLM issues in the rear view mirror
Those breakthroughs are possible, in part, because process intelligence addresses common LLM flaws. Here’s how process intelligence and Celonis solve typical LLM issues:
Reliability - aka ‘hallucination’. Despite the hype of what AI can do for you, LLMs still confidently state incorrect information as fact. It can be as basic as watching ChatGPT fail at simple arithmetic. However if you prompt ChatGPT to a calculator it will provide the correct result. Our process knowledge is a little like the calculator – we combine AI with the right process data and knowledge to ground responses in the process context.
Staleness. LLMs take months to train and deploy. Fine-tuning is expensive and time-confusing so the model cannot provide responses based on recent events or information. We use a variety of techniques to enable the foundation models to provide responses that are based on customers’ up-to-date process data and knowledge.
Security. Foundational models pre-trained or fine-tuned using private data risk leaking this information to any of the model’s users. That could include company information. There’s no way to delete data from the model once it’s up and running. We ensure customer data is always kept separate to avoid data leaks.
Why Celonis and process intelligence
12 years, more than 1,300 customers, and 5,000 enterprise deployments. Turns out that’s what it takes to create the most advanced, sophisticated, and accessible process intelligence in the world today.
Like mathematics translating the rules of the universe or the Rosetta stone deciphering the hieroglyphs, process intelligence makes processes understandable to AI. By contextualizing our process knowledge with data, we’ve created process intelligence that can tell you an invoice is a duplicate, an order is likely to be delayed, and team members are missing out on cash discounts.
What happens when AI speaks process
Providers of process intelligence open up to their ecosystem a universe of possibilities. Check out two ways Celonis gets AI to work for business:
Connecting customers with expert LLM providers. Celonis will make it easy for customers to integrate with LLM providers like OpenAI, Microsoft Azure and Google PaLM. Combining these foundational models with our process intelligence, customers can create new generative AI solutions that improve their processes and, as a result, their business. For example, getting answers about their processes inside Microsoft Business Chat or deploying an LLM to help plan responses to stock shortages.
Domain-specific solutions. To build reliable and highly accurate solutions, we’ll tailor models specifically for the use cases that customers expect from us. For instance, to improve the way business users interact with process data in Celonis through conversational interfaces, and to automatically recommend process improvements.
Process-intelligent AI understands your business and how to drive its performance like no other technology. It did not appear overnight – here’s a taste of the products and innovations that got us here:
Machine Learning Workbench allows customers to build custom models and drive more intelligent decision-making.
Duplicate Checking app uses ML algorithms to detect duplicate invoices with fuzzy matching.
Check out a demo of our work with our partner Rollio, showing how customers can generate their own data analyses and visualizations by typing in questions in everyday language.
Tap into the superpower of AI
Enterprises everywhere want to up their AI game. Some want to simplify everyday tasks, others want to improve a process here and there – the best will unleash it on every task, every process, every decision, and every action that matters.
They will become process-intelligent enterprises where AI speaks fluently with processes to transform the business at speed and at scale. Ask those teams what AI has done for them lately and my guess is they’ll answer, ‘everything.’
