Innovation in action: real-world solutions from Celonis AI Lab in the Bay Area

Last year, six teams of Celonis customers gathered in the Bay Area for AI Lab–our ongoing series of one-day AI hackathons. Here, they worked hands-on with Celonis solution experts to build impactful prototypes for their most pressing use-cases. Teams leveraged Celonis products–like Annotation Builder, Orchestration Engine, and Prediction Builder–to build solutions using their own unique process data and business context.

At the end of a busy day of solution building, each team had five minutes to showcase their solution. During their presentation, they explained the business opportunity they addressed, the potential impact of their solution, and most importantly, how they used AI to achieve it. This hackathon was more than problem solving–it was about collaboration, competition, and innovation.

Here’s a round-up of what our customers built in the Bay Area:

American Supplies Company: Smart Delivery Predictions in Action

Opportunity: Late deliveries are typically caused by warehousing defects, transportation bottlenecks, and other external logistical factors. Traditional predictive tools operate with an extensive data lag and manual reporting, which limits the ability to perform interventions proactively. Identifying “at-risk” or “projected late” orders also require manual investigation across different systems and multiple teams, leading to delays.

Solution: Using Prediction Builder and Annotation Builder, the team designed a solution that transforms their current delivery process into a proactive operations model. Prediction Builder identifies orders likely to be late, enabling timely intervention to prevent further delays. With Annotation Builder, specific actions are recommended to influence order delivery performance, enabling faster, smarter decisions to execute necessary adjustments to ensure on-time delivery.

Multinational Digital Communications Corporation: Agent Dwell Time

Opportunity: Managing customer and partner issues can be time consuming and tedious. A solution that delivers consistent and repeatable service to hundreds of support agents of various capabilities can be envisioned, ensuring every customer interaction meets a unified standard of efficiency and accuracy.

Solution: By leveraging Annotation Builder, a primary analysis of past email exchanges was performed to categorize and identify specific interventions that would accelerate the case resolution. This foundation enabled a Retrieval-Augmented Generation (RAG) solution that analyzes open cases and active email threads. By surfacing similar cases, recent interactions, and relevant documentation, the system recommends specific actions to drive faster outcomes. These solutions were framed out during the AI Lab and are in various stages of development.

IT Service Management Company: Non-PO Invoice Coding Assistant

Opportunity: Processing non-PO invoices under $2,000 typically requires manual effort from the company’s Accounts Payable team, which rely on repetitive back-and-forth actions with buyers and requesters. The manual assignment of cost centers and spend categories also leads the Accounting team into a cycle of re-reviews, causing further delays in payment execution.

Solution: The team implemented Annotation Builder to automate the classification of non-PO invoices based on invoice memos, vendor, requester, and historical patterns for similar invoices, generating actionable recommendations. These recommendations are delivered directly to the Accounts Payable processors within Celonis, allowing them to instantly accept or manually override the suggestions when needed, creating a streamlined validation process.

IT Service Management Company: Late Payments Prediction

Opportunity: Diverse arrays of bottlenecks like approvals, exceptions, and holds typically lead to late payments and other delays. To improve financial performance and vendor relations, the organization sought to reduce the Accounts Payable team’s late payment rate, and better identify friction points and address them.

Solution: To identify why an invoice is likely to be late, the team leveraged Annotation Builder which provides transparent reasoning, whether due to potential approval bottlenecks, to exceptions and historical vendor trends. Annotation Builder is also used to prioritize invoices based on due date and risk level–creating a view to monitor to-be late invoices. The team utilized predictive modeling to uncover hidden patterns to further forecast the likelihood that an invoice will be late.

System Software Company: Deal Slip Predictor

Opportunity: Sales leaders face a visibility gap regarding opportunities at risk of slipping into future quarters–obscuring the true volume of Annual Contract Value (ACV). Forecasting relies on Deal Managers manually updating each opportunity with a risk level and sales representative intuition–leading to unreliable close dates and last-minute surprises. Revenue planning, cash flow timing, and investor guidance accuracy can become distorted when deals slip.

Solution: Leveraging Prediction Builder, the team introduced an objective, data-driven layer to the forecasting process. By analyzing real-time deal signals, like performance history, quote volume, and legal redline status, this solution predicts the probability of a slip for every open opportunity. It provides visibility into ACV at risk and enables real-time prioritization and proactive deal coaching for sales leaders–ensuring resources are strategically deployed to secure commitments and stabilize revenue performance.

Digital Infrastructure Company: Incident Categorization & Automation Monitor

Opportunity: Incident pipelines are currently constrained by a high volume of tickets that require manual intervention–resulting in incorrect assignments and poor customer satisfaction due to long resolution times. Many of these tickets are Tier 1 and are ideal candidates for automation or self-service, but there is limited visibility regarding incident types and the development of effective automated workflows.

Solution: The team used Annotation Builder to automatically categorize incoming tickets based on descriptions and historical closure notes to determine whether manual categorization can be replaced and improved. By analyzing these trends, the solution further assigns an automation and self-service potential score to every incident type–pinpointing exactly which requests are most eligible for full automation.