James Taylor's Decision Management

One of the common questions I get from people once they understand the potential value of attacking decisioning problems is "how do I get started". Here's a quick summary of the approach I take that seems to work well:

1. Remove decision process from production applications and processes
   If you let the decision stay embedded in the system or the process then you will never manage to improve it. This does not mean disconnecting it from where it is used or ignoring how it is used by systems or processes, it just means identifying the decisions as unique aspects of a system, decision services if you will, and focusing on them as something that can be managed and improved separately.
2. Analyze decisions' potential for improvement
   Not all decisions are equally ready to be automated.By and large there needs to be some upside to doing it better or some downside to doing it worse - a risk or reward. In general these decisions will have:
   - Lots of rules
   - Rules that change often
   - Rules that are complex or interact in complex ways
   - Rules that require domain expertise to understand and manipulate effectively
3. Automate the high-volume decisions
   There is not much value in automating low volume decisions. Focus on applying automating and decisioning technology on those "blue-collar" decisions that are part of your day to day business and so come in relatively high volume. These are well suited because small improvements in the decision can have a big impact as you make so many of them.
4. Give business users control of how the decision is made but keep the user interface simple
   Business users don't want to maintain rules any more than they want to write code. If you want them to take control of these decisions, and you do, then you will need to keep the interface simple (to them - it can be full of jargon and abbreviations for example as long as they are your business users' jargon and abbreviations). Indeed as the complexity of the science used in making the decision goes up (see below), you will need to focus more and more on keeping the interface simple for those that control the business decision.
5. Apply predictive analytics to improve the decision
   Using the data you have or that you start to capture once you have automated the decision to create predictive analytics can make a huge difference to the effectiveness of the decision. You might be able to develop analytically-derived rules or add predictions that allow for completely new kinds of rules (treat people likely to churn differently for example). This is the real "operational BI" you are looking for.
6. Design for production
   Once you automate these transactions they will need to meet your operational/transactional standards. This is not offline reporting but front-line transactional stuff. Design it that way.
7. Provide change management and evolutionary guidance
   Automating the decision helps but getting good at evolving and improving it is what really delivers benefits. Don't think you are done when you first get the automation in, that's just the beginning. Make sure your plan includes how you will assess performance, how you will make changes and how you will create a constant improvement cycle.

This works. Really.