Every analytics project teaches you something a textbook cannot. The data is messier than the case study. The stakeholders need different things than they say they need. The metric that seemed obvious in the brief turns out to be the wrong thing to measure. After completing three healthcare analytics projects spanning telehealth, digital health, and public health, here are the lessons that stayed with us.
The first lesson came from the telehealth project. We were asked to build a dashboard to track patient adherence. Simple enough — until we asked the care team to define what adherence actually meant in their context. They had never been asked. It turned out that the word meant different things to different clinicians, and none of those definitions matched what was being captured in the data. Before we wrote a single line of SQL, we spent two sessions clarifying terminology. The dashboard we eventually built was only possible because of that work. Clarity of definition is not a soft skill. It is the foundation of every reliable metric.
The second lesson came from the sleep health prediction project. The model worked. The accuracy was solid, the feature importance was actionable, and the product team was pleased. But when we presented the findings, the most important moment was not the model output — it was explaining what the model could not do. Predictive models built on cross-sectional data have real limits. Knowing where the analysis stops and where the speculation starts is not something the model tells you. It requires domain knowledge to communicate responsibly. Analytics without intellectual honesty is not analytics. It is noise with confidence intervals.
The third lesson came from the YOHAN Africa public health project. We had 399 questionnaires and three research questions. The temptation in any analysis is to go looking for interesting findings — to let the data lead you somewhere unexpected and call it insight. We resisted it. Every question we asked of the data mapped back to a decision the NGO could actually make. The analysis was constrained, and the output was sharper because of it. Analytical rigour is not about complexity. It is about discipline — knowing what question you are answering and refusing to answer a different one instead.
These are not lessons about tools or techniques. They are lessons about how real healthcare analytics work actually happens — in the gap between what is asked, what is measured, and what actually matters for patient care. They are also the reason we build every part of the HDIQ programme around real project work rather than synthetic exercises. The textbook version of healthcare analytics is clean and manageable. The real version is where the learning actually happens.