The desirable economics associated with any investment is easily enough stated – the benefits should exceed the costs. In an immature domain such as predictive analytics the economics are not so easily established, and this is made more difficult by the hype, over-expectation, inexperience and general confusion surrounding the topic. Predictive analytics and big data are everywhere, and we’ve become so enamoured with the buzzwords and hype that the economics seem like a rather dull topic to address.
The first thing we need to establish is why a business would want to use predictive technologies. I shall ignore big data simply because it’s a non-topic, despite the hype. It’s plumbing for data – bigger pipes, more of them and different shapes – but just plumbing. So back to predictive analytics. The reason your organisation should be interested in these new technologies is because they enable a second wave of business automation. The first wave lasted over four decades and was primarily concerned with the automation and efficiency of business processes. The swan song of this era was actually the Enterprise Resource Planning suite. It integrated transactional systems and made them much easier to manage – in theory at least.
So welcome to the second wave of automation – decision automation. This is the real reason why predictive technologies are so important – they allow us to automate (completely or partially) the decision making process in our businesses. As always the real pioneers are to be found in financial services. Loan approval, fraud detection, customer targeting and so on, almost always involve the use of predictive models. They not only automate the decision process, but they make it more effective. It’s actually a much more powerful use of technology than the process automation that went before it.
This background is necessary to understand the benefits that might be derived from predictive analytics. Provided the models are accurate, that the data is available and of good quality, that the models can be deployed into a production environment, and that the effects of the models can be measured and managed, we have a formula for a very significant lift in business efficiency and efficacy. And what is more it is measurable – given the will and the tools to measure.
So somewhat surprisingly the difficult part of the equation – the benefits – is actually easier to measure that we might first have thought. Next we need to consider the costs, and thrown into this we need to consider risks. The costs are fairly easily listed – data, technology, skills, management overhead, training etc. etc. The risks however are harder to quantify, because unlike process automation decision automation has more profound risks associated with it. These can be roughly categorised as:
- Faulty models due to poor data quality and low skill levels.
- Poor integration with production systems.
- Unmanageable complexity and poor monitoring and management.
The immutable law of risk and return means that a technology that might deliver significant gains might also deliver significant risk. A well conceived and executed predictive model might easily revolutionise some part of a business. A sloppily developed and managed model might do a business considerable harm. A loan application model that says ‘no’ to the majority of applicants is hardly going to help a business – but hopefully it would be quickly spotted. But when the number of predictive models used is counted in the hundreds, an errant model (or models) may be much harder to spot. The solution, as always, is model management and reporting; although such is the immaturity of predictive analytics that only a few suppliers provide any such capability.
To summarise. Predictive technologies are a key enabler in the automation of business decisions. The potential benefits are not only more efficient processes, but also more effective ones. The costs associated with decision automation (the reason we use predictive technologies) are easily listed. The risks however need more careful consideration, and frankly need to be taken very seriously – if you want greater gains you inevitably have to take greater risks.
This in a nutshell is the economics of predictive analytics – a key enabling technology in decision automation. Those who get it right will start to lap those who haven’t got a clue – and it will show.