It is generally not helpful to try and get 100% of the Decision Service built in the first iteration. A more agile approach is preferred, where multiple iterations develop the functionality of the Decision Service. In general, two forms of iterative development are possible.
In the first, the number of transactions handled by the Decision Service gradually increases. This works well when the Decision Service makes a decision that would otherwise be made by a person but where the option for a manual decision still exists. The initial version of the Decision Service can thus handle relatively few transactions, leaving most for manual review. Reviewing the manual actions taken can define and shape the missing business rules. Candidate rules can be compared to the manual actions to see how close the rules are getting and to see if the rules outperform the manual decision-making. Over time, the number of rules or their sophistication can be increased so that the Decision Service handles a larger percentage of transactions. This approach is used in underwriting, for instance, where the Decision Service gradually evolves to handle more products and more customer situations with fewer applications being referred to an underwriter with each of the iterations.
The second approach is more appropriate when an automated decision is required. For instance, a Decision Service that must return an appropriate ad to display on a webpage does not have the option of deferring that decision to a person. It must make some kind of decision. Such a Decision Service can still be evolved over time. It needs to have a reasonable default response, perhaps a standard and non-controversial ad that can be used whenever it cannot make a more appropriate response. Initially, it might always make such a response, and, over time, more rules and analytic models can be added to allow it to make a more targeted response in specific circumstances. Each of the iterations decreases the percentage of the times when the generic response is made.
Besides developing the initial Decision Service in an iterative approach, most Decision Services repay an investment in a decision analysis infrastructure for ongoing analysis and refinement after deployment.