Our dev team worked on their tech feasibility research, exploring options and exchanging with machine learning engineers. One week later, we grasped what was feasible within that short timeline, and rough concept ideas started to emerge. I facilitated a co-creation workshop allowing more time for debate, sketching, and visualising ideas via fast prototyping (paper first, later SAP's BUILD tool). At the end of this design sprint, we narrowed down to three concept ideas:

1. Anomaly Detection: alerting managers automatically when cost center spending looks off, before mistaken bookings go undetected.

2. Smart Forecasting: predicting next year's expenses from historical data, reducing the manual, time-consuming bottom-up planning managers had to do line by line.

3. Smart Tagging: suggesting expense tag groups automatically based on patterns from other users, instead of managers setting them up by hand.

I designed the validation questionnaires, and on the next sprint I moderated expert interviews and stakeholder interviews with finance experts, controllers, and our new target group, managers, using thematic synthesis to identify patterns across sessions and present findings to the team. We gathered feedback from ten internal experts through 30-minute demo sessions, aimed specifically at desirability testing, measuring each concept's perceived value before touching usability. Smart Tagging ranked lower on both desirability and feasibility, narrowing our focus to Anomaly Detection and Smart Forecasting. With the design more mature by that stage, we then took it to SAPPHIRE/ASUG 2018, selected among hundreds of applicants, where we ran task-based usability testing with external experts to measure both desirability and real task completion. The results pointed toward both Smart Forecasting and Anomaly Detection, but Anomaly Detection won on technical feasibility and business viability, and went on to ship in the final product.