Reading Customer Mindset From Product Telemetry
Behavioural mindset indicators translated from product interactions; proactive lens for responsible engagement of at-risk users.
Theme Customer Experience & Personalization · Also Risk
In brief
Situation. Customers do not tell you how they feel. They tell you how they feel by what they do, by how fast they click, how often they reload, what they abandon, what they return to. Every product generates this signal continuously.
Complication. Translating raw interaction telemetry into emotional and behavioural indicators, anxiety, frustration, joy, engagement, eagerness, satisfaction, requires both deep data modelling and careful collaboration with the people responsible for customer welfare.
Resolution. Building on the Timeline of Facts work, this initiative used advanced data modelling to translate product interactions into mindset indicators. The work was scoped with close collaboration from Customer Experience, Product Development, and Responsible Customer Engagement to ensure the analytics aligned with business objectives and with the company’s duty of care.
Impact. Better understanding of customer mindsets, leading to improved product development, more personalised experiences, and, importantly, a proactive lens on responsible behaviour for at-risk users.
The longer story
Here is a thing that took me far too long to fully internalise: human beings are very bad at telling you why they did something, and very good at telling you something that sounds plausible.
The data is the opposite. The data is honest in a way the survey is not. If a customer clicks the same button seven times in two seconds, they are not “engaged.” They are frustrated. The system has failed them, and they have communicated this in the only language available, by hitting the button harder.
The interesting design question in a gaming product is that the same telemetry that tells you “this customer is loving it” also tells you “this customer is in trouble.” Joy and compulsion are not the same thing, and a responsible operator needs to be able to tell them apart.
The model we built does not replace human judgment; it lights up the cases that need it. That is, I think, the right ethical posture: AI as the early-warning system, humans as the responders.