Barring a directive from on high, i.e. the Executive Floor so to speak, it can be difficult deciding which new projects, or old ones for that matter, will benefit from automation of business decision making. Selecting the right project candidates is critical. Obviously sophisticated automation of unimportant decisions is a waste of time, however interesting the technology might be.
This article follows on from the first in the series entitled “Building Automated Decision Making into BI System Design”. I proposed a set of five project selection and implementation Phases for decision automation projects. This version gives more detail on the first three of these. A later piece will deal with the last two.
This discussion is a work in progress. I’ve built several automated decision systems, but have not attempted to define a project management methodology for them. Any further suggestions are welcome, Dear Reader.
The Phases suggested in my last piece have been modified in the light of comments and suggestions from many readers. They currently are:
- Identify the controllable business variables in your business environment
- Determine the business processes, and relevant decisions, that impact those controllable variables
- Identify the BI systems that support the business processes and decision contexts selected in Phase 2
- Design the business analytics that are the basis of the decision automation: business rules, predictive analyses, time series analyses wherever Phase 3 indicates potential value
- Evaluate feasibility and profitability of implementing the analytics created in Phase 4
Note: Phases 4 and 5 will be iterative.
Discussion on Phase 1
Phase 1 is, I believe, a novel idea in this context. Most authors writing on Decision Automation focus on decisions and business processes as their starting point, but I think this can result in poor project selection. Just because a business process is important to the enterprise does not mean that automating decisions that are critical to success is the right thing to do. Empowering managers to perform better with improved conventional BI systems may be an optimal answer.
As I see it, controllable variables are likely to be the least common denominators for automated BI. Decisions that can be automated must look to adjust, or consciously not-adjust, one or more of these. Unless it is feasible to monitor and adjust a particular controllable variable there is no point in making a decision that requires this; there’s nothing to do. I might decide that sailing a 50 metre yacht around the world is desirable, but without the funds to purchase it, or ability to sail it effectively, the decision is rather pointless. I may as well decide to fly to the moon.
Or in a business context, deciding to raise product prices when all our customers are on fixed price contracts would be equally useless. Therefore, a portfolio of identified, automatically adjustable, control variables is an essential commodity in the business of decision making automation.
These adjustable variables are limited in number, because of the nature of business. Below I offer some examples of such control levers, drawn from a product manufacture and/or distribution business context. I chose this sector because it is somewhat different from the insurance or financial services examples that are usually presented in decision automation discourses.