In my experience, data serves three distinct functions in business. Understanding each one: the value it creates, the pitfall it carries, and how to execute it correctly: is the difference between analytics that works and analytics that wastes.
Defensive analytics
Defensive analytics is activated when something has already gone wrong. A decision needs justification, a course of action needs defending, or accountability is in question.
The value, when approached rigorously, is significant: an honest, evidence-based account of what happened and why. That matters both internally and with external stakeholders.
The pitfall, and this is the one no organisation talks about openly: is that defensive analytics is where the integrity of your data practice is most at risk. Two failure modes are worth naming.
The first is motivated analysis. The data does not support what someone needs it to say, and the analyst is quietly pressured to work backwards from a predetermined conclusion. This happens more often than most organisations admit, and it corrodes the credibility of every subsequent analysis produced.
The second is retrospective justification. The decision is already made. Analytics is deployed not to seek truth but to build a case: the data becomes a lawyer, not a judge.
In practice, defensive analysis is almost always the most urgently requested. When something goes wrong the demand for data is immediate, and almost always politically loaded.
Protective analytics
Protective analytics is about building continuous visibility into performance before problems become crises. Think of it as early warning infrastructure.
The value is operational clarity: monitoring product expiries before they become write-offs, tracking sales performance to know whether your team is generating real returns or whether payroll is simply running ahead of revenue. When designed well, you know your business is at risk before it feels like it.
When a measure becomes a target, it ceases to be a good measure.
That is Goodhart's Law, and it is the pitfall here. The moment people understand what is being tracked and rewarded, they begin optimising for the metric rather than the outcome. In practice this shows up in two predictable ways: staff who learn to hit the numbers without doing the underlying work, and sales teams who concentrate on large, visible accounts while neglecting the small and medium clients who are often the backbone of long-term revenue.
Protective analytics works when it is designed with careful consideration of the inputs, outputs and outcomes being measured. The choice of what to measure, and the reasoning behind that choice: is the most consequential design decision you will make.
Offensive analytics
Offensive analytics is about using data to proactively extract value: to grow, optimise and win: rather than simply respond to events. The value is strategic: smarter decisions about where to invest, how to price, which customers to prioritise, and when to act.
The pitfall is where confidence becomes carelessness.
Consider an FMCG company running a targeted sales promotion. Customers are segmented and discounts applied on certain products to drive sales in adjacent categories. Unit sales volume is up 20%: a win, on the surface. But if the discount depth exceeds what the volume gain recovers, net profit can still be negative. The promotion boosted sales volume while bleeding margin. Without measuring the right metric, you would never know.
The same principle applies to predictive inventory management. A business forecasting stock levels to improve service levels may succeed on that metric while quietly accumulating holding costs that erode every gain. Offensive analytics demands the most rigorous measurement design precisely because results can look good long before you have looked closely enough.
What ties all three together
The risk is the same across all three: measure the wrong thing, or measure correctly and draw the wrong conclusions, and you produce knowledge that harms decisions rather than improving them. Data is only as valuable as it is correctly applied, and when it is not, it will have you chasing your tail with full confidence.
What I have consistently found is that any analytics endeavour must be grounded in a testable, defensible hypothesis before the work begins. That means structured conversations with the key people in a business unit to understand what truly drives outcomes: the real inputs and outputs, not just the ones that are easy to measure. And before any of that, it means grounding your thinking in existing literature: understanding how similar problems have been approached, which frameworks have been tested, and what the evidence actually supports.
As Newton put it, you are standing on the shoulders of giants. You are not starting from zero. You are building on what is already known and applying it to the specific context in front of you. That is the difference between analytics that informs and analytics that misleads. Not the tool. The rigour behind the question.