Guest post from David Pugh, Program Director, Product Management, IBM Business Analytics
I really love speaking to customers who are pushing our software to its limits. Those customers who are at the bleeding edge – the innovators and early adopters – regularly have great input to the creation of the products of the future.
Why? Because they are using the products in anger – pushing the software up to and beyond its intended use in order to drive competitive advantage and increased revenue.
I wanted to share a story about how IBM SPSS Decision Management came into being from working with such visionary customers, as a follow up from last week’s blog post.
The story starts five years ago when I went on a “world tour” to spend time with customers with some of my product management colleagues.
In particular, we were interested in customers who were actively deploying and using the results of predictive analytics in their day to day business operations, such as:
· Auto insurance companies using predictive modeling to determine on the fly – while the customer is on the phone describing the accident – whether the claim was possibly fraudulent
· Retail banks providing personalized marketing offers to their online banking customers, trying to sell them additional products
· Mobile telecommunications providers scoring millions upon millions of customers every night looking for any signs that they were going to defect to a competitor
As I said, the customers we visited were using our products in anger, like a coach demanding the most from his/her players.
For instance, the customers were using Decision Management in conjunction with other applications (e.g. CRM, call center, websites and campaign management); had stringent performance requirements; and, all had invented their own methodology for managing / updating the predictive models that were being deployed into these front-end, operational environments.
It became apparent that the processes used to create, deploy and manage predictive models were eerily similar.
In fact, thanks to our customers, we were able to develop a list of best practices to easily create predictive models and inject the results of the analysis directly into business processes to improve outcomes.
The Seven Steps to Analytics Deployment
1) Acquire the Data. Customers use a mix of data including transactional, demographic, call center notes, social media, and attitudinal data from customer surveys as input to the modeling process.
2) Identify the Audience Determine the population for whom the outcome of the decision is valid. For example, with regards to insurance fraud, the customer may want to exclude any insurance claims that are caused by natural disasters (and process them a different way).
3) Define the Desired Outcomes This is the heart of “The Seven Steps” where the customer determines the range of “Decisions” that could be delivered into their operational environment.
For insurance fraud, the desired outcomes that would be ideally delivered as a “Decision” to the call center agent processing the claim could be:
· Fast track the claim – low risk of fraud and low cost
· Push through standard processing – low / medium risk of fraud
· Refer the claim to the special investigations unit – possible fraud.
4) Enlist Business Rules and Predictive Analytics to Determine the Ideal Outcome. If Step 3 is the heart, Steps 4 and 5 are the brains. Staying with the insurance fraud example, there may be a number of policy (business) rules that need to be applied to the decision, such as “All claims made within two weeks of setting up the policy must be investigated for fraud.”
Customers were also building predictive models that determined – based on historical examples of fraud – the liklihood that this particular claimant was behaving fraudulently.
5) Optimize the Outcome. What if the business rules output says “Refer the claim to the Special Investigation Unit” and the predictive model says “Push through standard processing”? The user may decide whether rules override models or vice-versa.
For marketing applications it is here that a user could optimize which of the five valid marketing offers would be made based on factors such as liklihood to respond, revenue and cost.
6) Deploy, Deploy, Deploy! Take the intelligence defined in steps 1-5 and deliver it to the appropriate business process. Once the IT configuration has been completed this is typically a one-click process.
7) Report and Monitor. Watch the performance of the deployed application and ensure that it continues to perform well, as well as share the results across the organization in easy to understand reports and dashboards.
Business users typically need to update rules, predictive models or the way in which the result is optimized on an ongoing basis. Automated techniques such as “Champion-Challenger” modeling are used to ensure the best models are always deployed.
If you’re a customer using our products in anger, please get in touch. Your input will help us build the next generation of IBM Business Analytics software.
For more information:
· Read the whitepaper on how Decision Management creates a closed-loop system that continually incorporates valuable feedback into your decision-making processes.
· Watch the video of industry analyst James Taylor discussing the importance of Decision Management.