AI on display in Columbus: Five real-world use cases from Celonis AI Lab

Last year, Celonis kicked off the month of October in Columbus, Ohio with six teams from five Celonis customers gathering together for our one-day AI Lab. Customers worked hands-on to build impactful solution prototypes for some of their most pressing business challenges with our latest AI products–like Process Copilots, Annotation Builder, and Prediction Builder.

The participating teams had five minutes to showcase their projects at the end of an intense day of solution building. Each presented the business problem they aimed to solve, highlighted the potential impact of their solution, and demonstrated their creative use of AI and Celonis products.

Here’s a round-up of what our teams built:

American Electric Power: AI Governance on Descriptive Spend

Opportunity: The AEP team sought to reduce the company’s annual descriptive spend—material purchases that are not linked to the master catalog (a list of approved goods and services with negotiated prices). By using AI to link materials referenced in purchasing descriptions with the master catalog, AEP could improve catalog conformance, minimize contract leakage, and ensure orders undergo crucial price validations—thereby realizing significant cost savings.

Solution: The team designed a solution using an agent to automatically review key purchasing data–including PO line and header descriptions, internal/external notes, and the Material Master Name. The agent then makes a recommendation of the top five matching catalog IDs for the descriptive item. These precise recommendations are then pushed directly to an operational dashboard, giving procurement agents a streamlined interface for review and immediate action.

American Electric Power: AI Helpdesk

Opportunity: The current dependence on siloed systems necessitates substantial manual effort, requiring six full-time employees (FTEs) to field and resolve all inquiries. This reliance on time-intensive, manual data retrieval creates processing delays and a notable lack of transparency, a condition which can strain vendor relationships.

Solution: Upon receiving a vendor inquiry, an agent is triggered automatically that categorizes the request, assigns both a severity level, and identifies the appropriate resolution team. The agent then drafts comprehensive response options for the resolution teams, automating time-consuming manual searches, enabling FTEs to review and finalize the agent’s pre-written responses. This helps to speed up resolution times and significantly improve transparency for vendors.

Tarkett: Spend Under Management Improvement Analyst

Opportunity: Category Managers currently face a critical visibility gap: they know contract leakage exists, yet they are unable to effectively quantify it, making it difficult to identify their true spend under contract. This limited visibility prevents the identification of significant opportunities to negotiate new, more favorable contracts or maximize the utilization of existing agreements.

Solution: Using Annotation Builder, each purchase order item is automatically analyzed–using category, vendor, and free-text description–to determine if a contract was used, if a new one is needed, or if an alternative vendor should be leveraged. This solution also provides automated backward-looking reports for the C-suite's Spend Under Management initiative and identifies contractual opportunities based on purchasing patterns and flags misclassifications via free-text parsing, preventing the need to reroute items to Category Managers.

American Healthcare Services Company: AI Powered Chargeback Prediction & Resolution Operationalization

Opportunity: This organization processes an average of hundreds of thousands of partner communication lines every day. Of those, the daily rejection rate–ranging between one and two percent–translates to several thousand lines that the chargeback team must manually resolve to ensure timely payment. The use case centers on efficiently determining the root cause of vendor rejections, predicting the likelihood of future rejections, and providing coordinators with data-driven guidance on dispositioning chargeback lines.

Solution: The team developed a solution designed to stop rejections before they even occur. Using the Prediction Builder, it can be forecasted whether chargebacks are likely to be rejected before the notification is received. It leverages past Process Insights to flag these potential issues instantly. Next, with Annotation Builder, the reasoning behind the prediction is provided, offering targeted, user-friendly disposition recommendations to coordinators.

American Medical Equipment Supplier: Vendor Confirmation Orchestration

Opportunity: The process of obtaining purchase order (PO) confirmations from vendors can be time-intensive, owing to the necessity of repeated vendor contact to secure final confirmation and the requirement to manually update all confirmed information. This cumbersome process often results in a high rate of missed confirmations, thereby reducing planning reliability due to unconfirmed delivery dates.

Solution: Using Orchestration Engine, outstanding confirmations are monitored. When triggered, it creates a pre-populated form and sends an automated email directly to the vendor. If the vendor confirms via the secure form, the data is instantly written back to Celonis. For confirmations received via email with a PDF, the solution leverages a Large Language Model (LLM) to extract relevant confirmation data, ensuring it is accurately written back into Celonis. This eliminates repeated manual contact and enhances real-time planning reliability.

American Insurance and Financial Service Company: Claims Management Letter Compliance

Opportunity: All fifty states maintain unique regulations regarding required letters to the insured. Failure to comply directly results in negative customer satisfaction, operational inefficiencies, and the risk of substantial fines. Previously, regulatory compliance was analyzed through manual, random-sampling methods, which revealed crucial compliance gaps that were previously blind spots.

Solution: The team configured Annotation Builder to read existing internal documentation on letter requirements, allowing it to instantly determine which letters are mandatory for any particular open claim. Then, the system presents all potential letter requirements for the appropriate state directly to the claim adjuster. The solution also highlights any letters that need to be sent, but are currently missing, removing the administrative burden of manually checking compliance documents.