I have blogged before about Tom Davenport's articles on analytics, or at least about articles that quote him such as this one and this one. In an email exchange the other day, he pointed me to this article - Automated Decision Making Comes of Age (registration required). The article is great - I highly recommend it - but it requires registration and as such I won't quote from it extensively here. The summary provided for free, however, has some key points in it on which I will elaborate from my own experience and materials.
Futurists have long anticipated the day when computers would relieve managers and professionals of the need to make certain types of decisions. But for a variety of reasons — including management skepticism and concerns about solution complexity — automated decision making has been slow to materialize.
When considering the types of decisions suitable for decision automation using specialized technology like business rules I have previously proposed some rules of thumb. Firstly the decision is most likely to be operational, possibly tactical and probably not strategic. Not only do operational decisions have the kind of volume that justifies automation, they tend to be highly repeatable and there is typically value to consistency. Tactical decisions may also be good candidates if they are complex enough and reasonably high value but you will likely find you don't automate the whole decision so much as support or guide it. Within these kinds of decision:
-
There are lots of rules
Managing large numbers of inter-dependent rules is very difficult to do with manual or traditional development approaches -
There are rules that change often or must change quickly
This aspect of business agility was not one that Jeanne and Tom emphasized but it is often the key driver for decision automation. If my decision is constantly or rapidly changing then traditional approaches will not work well enough. -
The rules are complex
Approaches designed to externalize are manage business rules will work better when the rules are complex than will traditional approaches. The syntax is cleaner, the rules are atomic and the interactions are better managed. -
The rules require business expertise to understand
Implicit in much of the paper, this is a crucial consideration. When you are trying to enforce rules in an automated system and the rules require non-IT expertise to understand, then an approach that manages those rules as declarative components in a business-friendly syntax is likely to work much better than a traditional approach. -
Data-driven insight must be applied in a system
Automating decisions gives you a platform for applying the kinds of insights resulting from analytics to automated systems. Most BI approaches are aimed at helping people understand what the data is telling them. This will not work for telling a system what the data is telling it. You need predictive analytics that can be embedded - for instance, segmentation analysis that uses analytic techniques but results in business rules (in the form of a decision tree) can be a powerful way to improve an automated decision. If you have not automated the decision, the report on segmentation will be interesting but not truly useful.
Automated decision making is finally coming of age, the authors argue, and the new generation of applications differs substantially from prior decision-support systems. Today's applications are easier to create and manage than earlier systems. Rather than require people to identify the problems or to initiate the analysis, companies typically embed decision-making capabilities in the normal flow of work. Those systems then sense online data, apply codified knowledge or logic, and make decisions — all with minimal amounts of human intervention.
So what are the key reasons you can and should consider decision automation right now?
- Because maintenance by business users to deliver business agility is practical using the best-in-class tools
- Because the technology is production-ready - business rules management systems are faster than ever and embedding predictive analytics is getting easier
- Because the tools can be integrated into a wide range of systems and are easy to integrate with existing IT infrastructure
- Because you have enough data from your electronic backbone of information systems to make this practical
They can help businesses generate decisions that are more consistent than those made by people, and they can help managers move quickly from insight to decision to action.
One of the key concepts in Enterprise Decision Management is the idea that you can use business rules and predictive analytics to turn insight (both insight derived from data and insight derived from expertise and experience) into appropriate decisions and actions. Tom and Jeanne also make this point and, although they use different words, show a focus on several aspects of decision yield, notably consistency, precision, speed and cost. The one area that seemed to get less focus in the paper is one of agility - the fact that some of the technologies developed to automate decisions, especially business rules management systems, are also very robust in the face of change and so deliver great business agility.
This can help companies reduce labor costs, leverage scarce expertise, improve quality, enforce policies and respond to customers.
The paper talked about many benefits of decision automation like reduced staffing costs, more precise decisions, better risk management and so on. One additional area where decision automation can lead to a reduction in costs is from lower external reporting costs. If you think about processes that include buying external reports, like auto insurance underwriters and DMV reports, these reports cost money. A decision automation approach can be designed to request external reports only if the data in the report will actually make a difference. This can reduce costs dramatically in these kinds of processes. Another is in the area of exception handling. Most decision automation will not automate 100% of all decisions, some will still need to be manually reviewed. However, there is often still a lot of value in using rules to support this manual review. Instead of just getting a decision to process the reviewer will get information about why they are being expected to manually review it - what rules fired to cause it to be referred. This can dramatically speed the process of review.
As automating decisions becomes more feasible, organizations need to think about which decisions have to be made by people and which can be computerized.
Tom and Jeanne talk about how companies need to decide which decisions can be computerized. I would add that there are sometimes decisions that must be automated (so that responses to events are sufficiently timely for instance) and those that should be (so that customers can get the self-service options they want say). Decision automation works and is ready for prime time. Look for opportunities to exploit it not just places where it would be OK to use it.
In the paper Tom and Jeanne also talk about concerns with a potential lack of future workers to provide expertise when capturing and documenting rules. However, one trend I see is the use of decision automation to capture the knowledge of baby boomers retiring. Some companies have manual decisions being taken by groups of workers who are mostly going to retire soon and who the company either thinks they cannot replace or that they would prefer not to have to replace. The potential for capturing their knowledge into a decision automation system can be significant and the lack of future workers could be a reason for automation not a problem.
So, want some examples of people who are succeeding with this already? Check out the case studies. Want to pre-empt the excuses you will get? Check out my top 10 excuses.
Read the paper, it's cool.
