Vin Vashishta

Bottom-up prioritization suffers from a lot of bias. Frontline teams and customers all feel their needs should be priority 1. Data and AI Product Managers need to use a value-based prioritization framework. Expectations must be managed. Internally, we can align on core business goals. With external customers, we must emphasize what we can deliver and have already delivered. We should also explain that priorities are constantly reevaluated.

Prioritization is a value-based decision that comes after opportunity discovery, so bottom-up prioritization is the estimation phase of an opportunity that's surfaced by working with frontline teams and leadership. Building an accurate estimate of costs and potential returns benefits from working with domain experts who own the workflow that the product/feature targets or interact with the customers it will target. They have the best connection with the need. For data and AI products, frontline employees help define functional and reliability requirements. We need both for the feasibility and cost assessment.

Bottom-up prioritization connects products with domain expertise. The people involved are as close as we can get to the internal workflow or customers. Experts provide the most accurate starting point possible, even though it's imperfect and incomplete. For data and AI products, they help us determine the baseline. They provide input on what data points are important and what outcomes we should focus on. All of these points make opportunity estimation more accurate so we can prioritize based on value.

With NN training, garbage in = garbage out. Preprocessing should target finding and fixing the root causes of data issues. It's too expensive and often infeasible to clean your way to quality data, so it's critical to gather it properly to begin with.

There are 2 sides to every user research story: what the user tells us and what they actually do with the product. Designing products to capture usage and compare the two stories is possible. I have worked with internal teams who raved about a data product. I looked at the user login history data to find the product’s power users; no one had logged in for over a month. Screen for people who will give you honest feedback and use qualifying questions during the selection process.