Automated decisions can stop your losses

Late decisions, lost opportunities

In a former post on this blog I argued why Mr. Ferguson, Managing Director of Intelligent Business Strategies Limited, might have recommended a dangerous path for business intelligence, the path of Automated Decisions. However, at the network dinner later that day I ended up at the same table as Mr. Ferguson and thus had the opportunity to let him elaborate on the subject.

Mr. Ferguson argues that some business decisions in their nature take so long to carry out manually that it may inflict losses or otherwise put the organization at risk. For example, in a bank it may be determined in one department that a customer is unable to pay interest on his real estate loan. Thus, the bank should check its other engagements with this customer to make sure that his money shortage is taken into consideration when evaluating the overall risk. The customer may for instance have a credit card that must be blocked or somehow hindered from being further credited.

In one particular bank this evaluation process took six weeks. Thus, the bank would allow a customer to increase his debts several weeks after it was known that the customer might be unable to pay his debts.


Mike Ferguson at the Network Dinner in restaurant Påfuglen in Tivoli, Copenhagen

Advanced statistical algorithms

In that same discussion, Mr. Ferguson pointed out that by using a range of information about the customer such as historic transactions, current account statuses and demographics it is possible - thanks to advanced statistical algorithms - to make computers calculate probabilities of various scenarios or outcomes of the relationship. The algorithm will calculate the probability of a certain hypothesis, so if for example an employee in the bank needs to investigate whether certain conditions are met on a customer meaning his credit account should be blocked, then the algorithm can make a calculation of the probability that the employee will come to one conclusion or another - given the information available to the algorithm.

For example the algorithm could find that there is a 96% probability that a certain customer meets these criteria and the account should be blocked. Thus, it makes sense to allow the computer to make that decision if it can say with a certain confidence that the criteria are met.

Obviously there is a small risk that the criteria are not met and that the computer makes a wrong decision but this should be compared with the risk of human error. In fact, if an organization can document the level of human error then the algorithm can be programmed to make better decisions than humans. It is just a choice of the confidence level of the algorithm.
In cases where the algorithm cannot make a decision with the required confidence level the case will be left for manual handling by an employee.

Leave to computers what they can do better

Mr. Ferguson's point is that we should allow computers to make decisions that do not require human attention or that computers can do with a higher quality/faster than humans. When it comes to manufacturing it has long been accepted that machines usually produce a better quality than humans but maybe we yet have to accept that this idea can also be applied to decision making. A lot of science fiction has dealt with this idea and any one of us can probably mention at least one movie where this scenario runs out of control - think Terminator or Matrix.

But it is really ourselves who control what we leave to computers to decide. And in our constant race for improving production efficiency and service levels while maintaining and improving our quality of living it is important that we take the stress off humans by letting computers and other machines to do their part.

Automation and Lean BI

So, is automation of decisions "LEAN"? The immediate answer to this is "yes" because automation relieves humans of trivial work. Well, not trivial in the sense that it is easy to carry out manually but trivial in the sense that it can be automated and carried out by machines more efficiently than by humans.

However, there are also arguments against automation as a LEAN principle. If the number of decisions to be made is low then the effort of programming decision algorithms may not be justified by the benefits. This is fairly obvious. What may not be so obvious is that any type of automation makes the organization more "stiff" or fixed in its operations for two reasons:

1. automation takes longer to implement and change than manual processes
   
2. automation lowers the organizational awareness of the processes that are handled by machines. If all instances of a process are handled by a machine then there will be few humans who are familiar with the process. Thus, if the process breaks down or needs to be modified then there may be very few humans who actually know how to handle the processes manually. This will prolong the change process which means the organisation will react more slowly.
   

I shall not go deeper into this subject; automation and LEAN is also discussed elsewhere. I shall only conclude that automation is indeed LEAN if it can be justified by economics of scale and if it does not limit the organisation's flexibility severely.