David Proctor of Experian gave the next session I attended on Combining the Power of Analytics and Business Rules to Drive Enterprise Decisioning Solutions.
He started by talking about the criticality of business rules and the value of managing them. In particular the ability to being experts into a collaborative approach. David then gave a case study about a telco moving to simplify its change management of business rules while also handling more complex cross and upsell and making pricing tables easier for local management. Got some improvement through business rules management but found that the customers were not being treated right and so bad debt was growing to over 25%. In other words they did not focus on what the data would have helped them figure out.
Their view is that your business rules are only as good as the data, the analysis of that data so it becomes actionable insight and the overall decision made using it. This focus on predictive analytics to integrate with business rules is very much the view we took in the book and is something Experian has focused on for a while.
Data complexity is on the rise – more data sources with more data. In particular there are many external data sources that can be combined with a company’s own customer data. Few organizations, in fact, use all of their data. The data is scattered and hard to integrate and often only accessible through third parties. Companies need to start focusing on how to bring this data together to create usable facts. This helps improve decision performance (getting the most from your data and infusing models with insight) and can empower consumers to act more intelligently. The kinds of facts they mean are things like average number of minutes over plan each month, number of large cash withdrawals in a period. This drawing out of facts or attributes from data is the critical first step in developing powerful predictive models.
Building these facts into a predictive model turns this into real insight. Insight into likely future risk of a customer, insight into future lifetime value, insight into how a decision for a particular customer affects your overall profitability and so on. For instance, start with data about a customer. Derive some facts or attributes from this like number of times they exceeded their plan, number of times they were late. These new facts can be mined and used to produce predictions and segmentation. These models and analyses can be represented as business rules and not just business rules but mathematically valid business rules. Rules that, for instance, divide customers into segments that correlate strongly with certain kinds of behavior or outcome. It is much more useful, for instance, to have rules that divide customers up based on future profitability combined with risk than solely on the attributes you might store in your database. The sense of expanding the data your rules have available by integrating derived facts and predictions is very powerful and something far too few teams do in my opinion.
Returning to his example he discussed how analytics were used to improve the portfolio of customers by segmenting them more effectively and making different offers and taking different actions for different segments. Halved their bad debt, for instance. He had a second story about a customer who moved to business rules but also tried to expand the amount of data being used. For instance their old system only used about 10% of the data they had to drive actions. Using rules to get control of their logic they also used data analysis and analytics to bring over 60% of their data into their business rules. In other words much of this data could not usefully be accessed in business rules as it stands in the database. By using data mining and analytics to derive significant facts from this data they were able to bring much more of the data to bear in the the rules. Great point. While I had always thought about data mining and analytics are able to extend the data available, it had not occurred to me before that it could also make it possible to use some of the data you had previously ignored by turning into usable facts.
The final step David talked about, like Steve and like me, was decision optimization. How do you trade off the various facts and trends and predictions so as to maximize your profitability given your constraints. These kind of optimization techniques and tools allow me to optimize a whole set of decisions (as a set) so that each is as good as possible given the constraints on each AND the constraints on the overall set of decisions. Very similar mindset to that discussed by Steve Hendrick previously. He also pointed out that the benefit can come not from costs, which might rise, but from top-line growth. This is what drives the discussion on multiple dimensions of decisions in Decision Yield – precision, consistency, agility, speed and cost.
The overall approach runs from:
- Data – finding and getting access to all the right kind of data
- Analytics – extracting the right facts and predictions from this data
- Decision Optimization – to manage tradeoffs
- Deployment in a BRMS
- Evaluation for constant improvement – adaptive control
David also pointed out that Enterprise Decisioning really means treating decisioning as something the enterprise should manage, treating decisions as corporate asset. It does not necessarily mean managing all decisions within an enterprise and certainly does not mean implementing this in one go.
I liked their focus on the ability of data mining and analytics to make data accessible to business rules in addition to adding new data to the data you had.