The role of “to-be” UX

The role of envisioning and documenting “to-be” state for your application’s UX in an AI implementation is extremely important. At a very basic level, it is important to know what the end state of your application would look like. But consider the variables involved, and the complexity shows its dangerous face. For one, the application will call a prediction rule. What if the prediction rule shows no prediction. What would the to-be state look like? Will it show an error message? Will it change the UX to take advantage of a deterministic rule? Now, what if the prediction does happen, but the confidence level is low? What kind of UX is required there? Let’s say the prediction rule works really good for certain subset of data, but not so well for another subset. Now should we provide UX options that are activated based on the confidence level of that particular prediction? Or should we provide UX options based on the subset of the data where prediction accuracy typically falls in a certain range? Another aspect of the UX is ”people” - a very important aspect. What can be done in UX to make people comfortable with the prediction? The answer could impact how the results are displayed, to providing supporting explanations for prediction, to even providing supporting analytics. Also, if we expect the accuracy to improve over time, should we plan for changing UX in phases? Then there is the aspect of telemetry. What info should I collect to make sure that I have enough inputs to either improve the model, set in data governance process, and of course plain old - to measure usage. In this blog I have touched upon only a few of such considerations. Needless to say, planning to-be UX upfront is an important step towards building practical AI. #abhayPracticalAI #artificialintelligence #ai