Bill Fair, one of the founders of Fair Isaac, once said that to succeed with analytics you had to “grab the decision by the throat and don’t let go”. As Big Data and analytics become ever more central to organizations, and as more and more money is spent on analytics, this advice seems particularly timely.
As I have said before, it’s easy to spend money on data infrastructure, especially big data infrastructure, and on analytics without seeing much of a return. Too many companies assume that if they just collect enough data, hire enough data scientists, build enough analytics – spend enough money – that somehow their business results will improve.
I have bad news – they won’t.
Data and analytics, even big data and advanced analytics, cannot improve your business results. At least not directly. What they can do is allow you to improve your decision-making. Improve your decision-making and your business results will improve – pay fewer fraudulent claims by more accurately deciding which claims are fraudulent, manage risk better by accurately deciding how much risk a loan or supplier represents, retain more customers by deciding who’s at risk and what will stop them churning and on and on. Improved decision-making is that turns your data and analytic investment into better business results.
Which brings us back to Bill Fair and his pithy phrase:
- Unless you know which decisions you need to improve, and what better decisions look like, you can’t improve them. Which metrics matter and which decisions make a difference is critical to identifying where to apply analytics.
- If you can’t separate the decision you make from the process that acts on it or the system that stores the data it needs, you can’t change and evolve the decision to see which analytics work and which does not. Changing the analytics you use to decide which claims are fraudulent should not – must not – involve a change to your claims process as well.
- If you don’t have a clear sense of the decision and how it impacts your business it’s hard to identify the kind of analytics that will help. If you are making decisions about how to save someone who is threatening to churn right now you need different analytics than if you are trying to decide how to proactively stop someone churning next month, even though your objective is the same.
- Only if you have a clear sense of the business and legal constraints on a decision can you design appropriate experiments to adapt and improve your decision-making. Your data may “speak” but the regulations yell and its not generally practical just to turn the machine learning on and leave it to figure out what’s best.
- Outcomes are not the same as decisions and analytic decision-making will not always result in good outcomes – it just results in more good outcomes overall. To analytically improve decisions you need to analyze the decisions you made and how they worked out. Yet most organizations only track outcomes and don’t record how they made decisions.
So if you want to succeed with analytics, grab the decision by the throat and don’t let go.