A Baton in-house innovation: User experience monitoring

To help inform the user experience design of our new On-demand simulations, the Baton technical team (lots of PhD’s here) decided to have some fun and built on BTP, two complementary User Experience monitoring capabilities. The first one, User Interactivity report, is something we may soon make available to instructors.

User interactivity report

DAS can now capture every single user action (time spent in app, change in prices, etc.) in a simulation for each participant (A1, B3, C2). This provides us with useful data as to what role a user focused on, and how often they clicked an app, or completed an action. Essentially, what part of the sim they access the most and how often they engage with the system.

Eye tracking and emotion recognition

User interactivity information is helpful but does not always tell us how participants felt, what report in a dashboard they really preferred (as opposed to being confused by), or their degree of happiness (or frustration!) during the game. All of which is important data when designing a game with no instructor. How do we capture this data? When testing one of our On-demand versions, we ask testers to activate their computer camera. Eye tracking data, correlated with the interactivity metrics, tell our designers what report in a Fiori Overview tile is really being used. In addition, a real-time emotion recognition Machine Learning algorithm analyzes facial landmark positions (e.g., eyebrows, end of nose, mouth) and provides emotional engagement data on 7 dimensions. When emotions and activity reach a pre-determined threshold, we automatically take a screen capture, providing a snapshot of hard data for our design teams to use. The emotion data captured is entirely anonymous.

The developers with whom we shared these new capabilities all saw value for their UI design. The fact it’s built entirely on BTP, therefore deployable like DAS on any SAP software, might lead us to bring it to the SAP Store. One thing is clear: this capability opens the door to truly intelligent user assistance, where the application senses the user need, let it be business insights or navigation assistance, and provides just-in-time help.

Innovations around our ERPsim Simulations

Innovations around Machine Learning Bots

Can I hire this bot? This is a  frequent request from participants exposed for the first time to the power of Machine Learning in an integrated enterprise system. Soon you will be. Baton‘s SimBots, present in all of our scenarios, leverage machine learning algorithms from the SAP HANA Predictive Analytics Library (PAL), running on the SAP Business Technology Platform. As they use the standard SAP cloud architecture, connecting to a customer’s production environment is doable. Need to add here a list of bots:

• Use of the Auto Regressive Integrated Moving Average (ARIMA) time series algorithm in SAP HANA PAL to forecast material demand in the field:
  • Predict future demand for each material based on past values;
  • Calculates, each day, the material valuation of the remaining inventory and the remaining budget to determine how fast it can spend;
  • Calculate a correction ratio (between required and available budget) to apply on the forecasted demand to obtain the allocation quantities to be made.
  • Forecast of the demand drives all business decisions to be taken such as:
    • Allocation quantities to be made based on forecasted demand and fund/material valuation;
    • Procurement of materials according to allocated quantities and remaining inventory;
    • Fund transfer is to be executed based on the required budget to cover the procurement to be made.

• Use of the Additive Model Time Series Analysis algorithm in SAP HANA PAL to forecast sensor readings and predict the number of days before equipment breakdown. Non-linear time series analysis that can handle data with strong seasonal (periodic up and down) effects;
  • Bot as Maintenance Engineer:
    • In Round 1, the bot only reacts to equipment breakdown by converting breakdown notifications to corrective maintenance orders (just like the other players);
    • In Round 2, the bot uses its sensor forecasts to determine the number of days before an equipment breakdown and depending on the mean time to repair KPI (learned from round 1 data), will generate preventive maintenance notifications just-in-time, to fix the equipment before breakdown occurs;
    • In Round 3, from a sensor forecast, the bot sets the threshold for each sensor so that the SAP IAM component triggers a preventive maintenance notification when a sensor reading exceeds a threshold value.
  • Bot as Resource Manager:
    • bot assigns operations to work centers to minimize maintenance costs based on the following metrics:
      • Work center costs per planned hour;
      • Equipment downtime cost (per day);
      • Work center efficiency (ratio between planned vs actual work);
      • Effective work center capacity (available capacity vs required capacity)  adjusted with the work center capacity ratio;
      • The bot sorts work centers by estimated cost to perform the work:
      • Costs per planned hour + equipment breakdown costs (considering the estimated time the work center will take based on its efficiency);
      • The bot optimizes, in sequence, maintenance operations-based work centers available capacity.
  • Bot as Inventory Specialist:
    • Based on predictive maintenance and related tasks list to be performed, required material components are ordered beforehand by considering different delivery delays associated by different suppliers.
    • This will avoid delays due to missing components, reduce time to repair the equipment and allow more efficient inventory management in the warehouse