If you’ve been watching what’s happening in Advanced Analytics recently then (apart from Big Data which we’ll come back to) Deployment, Automation and Decisions are 3 of the hot topics at the moment.
At a high level this is all about 3 types of automation and governance:
- Collaboration. Helping analysts share resources e.g. analytical jobs and models with each other and with others e.g. IT Operations, Testers, Business Users and Researchers.
- Deployment. Making it easier to take models through a lifecycle e.g. development > testing > staging > production > refresh > retirement.
- This often involves making it easier to plug models into operational systems e.g. a web site or a call centre.
- Decisions. Supporting Business and Research decisions with models. For example having a model predict whether an ATM transaction is likely to be fraud but link that to a decision rule like “if the model score is over 40 and the transaction value is over $100 then refer the decline and refer the customer to a call centre”. This is where business users get to use models to support the decisions they make and where they can do What-If? analysis to simulate the impact of those decisions.
Analytics Vendors have developed platforms ready to manage and deploy models in a more sophisticated way then ever before:
- SAS has its Model Manager supported by SAS Workflow Studio and SAS Real Time Decision Manager for Decision Management.
- IBM/SPSS lines up with Collaboration and Deployment Service (C&DS) and Decision Management.
- KXEN has Scorer and Factory.
- This is one area in which open source clearly lags behind the commercial vendors. Though for R, Revolution Analytics – who are a commercial vendor – have RevoDeployR.
And so on…
The trouble is that that many of us have signed up to use these deployment and decisioning frameworks.
At technoloy is not the issue. Most organisations are simply not ready. A senior manager in a major high street retailer – recognised for using analytics to transform its business – tells us they don’t have enough models to justify purchasing one of the deployment platforms. In one of our recent engagements with another leading UK retailer we mined a number of insights that the business is deploying operationally but only one predictive model came out of several weeks of analysis.
The truth is we need to do more groundwork around people and processes before we can plug the technology in. This includes defining workflows between modelers, IT and business users. This has to start with a “selling” the value of them … these platforms are not inexpensive but, as with predictive analytics as a whole, they should more than pay for themselves.
We’ve been looking at some of the platforms and they are impressive. This isn’t just a case of technology looking for a business problem. We’ll write in more detail about or views on the specific platforms later. If you can’t wait James Taylor has looked at many of them already. Here are his views on SAS Model Manager and IBM/SPSS Decision Management.
So when we do get our collective acts together. The technology really is ready and waiting.
