What is AI product management?

1. Ambiguity of outcomes

• In a traditional product management system, product behavior is usually binary, with predetermined results.
• For example, a simple website’s submit button is either blue or not blue. Or when the customer launches on an e-commerce website, and if they are logged in the computer would say “Hello Tom”. 
• But now with ML, the outcomes can no longer be clearly determined. Product managers have to deal with probabilities. For example, AI and ML might predict that the probability of a customer buying a product is, let’s say, 80%. 
• Let us take a real life example. Cancer and other healthcare industries heavily use AI and ML. Looking at an X-ray film, AI and ML can make a prediction that there is a 95% likelihood that the customer may have Tuberculosis. This is answered only in terms of a percentage and not a final yes or no. These variable amounts generated by using AI have to be interpreted and navigated by the product managers.

2. Explainability of outcomes

• ML algorithms often become black boxes, often making it difficult for product managers to understand the reasoning behind their predictions. 
• Let us take another example. When a credit card transaction is declined, it is quite difficult to understand the rationale behind this outcome. This turns the situation into a data science problem for product managers. 
• This later on becomes even more difficult for them to justify the AI-based product decisions to stakeholders.

3. Fairness, bias and data imbalance

• The data that you use may not be very balanced. It may be skewing towards one segment of people. Machine learning models trained on such data are usually biased towards the majority class.
• This usually happens when models try to minimize the overall error rate and, in the process, tend to ignore the minority class.

4. New infrastructure/processes/tools

• In the AI and ML world we live in, the AI ops and infrastructure, processes and tools that we use are very different from traditional product management.
• This requires product managers to become experts in understanding and utilizing the new AI based tools to solve product problems.

5. Identifying the right problems to solve creating intelligent experiences

• This still remains a very big issue for the product managers and hence they have become so much more relevant. 
• They should know how to find the right ML problem from the business problem that they have, and also know how to translate that into a ML problem so that the data scientists exactly know what to model for.
• These problems are not easy to go to the stakeholders and pitch the return of investments because AI products take a long time to materialize.