Automation without AI: when rules are enough
For recurring tasks with describable rules, a rule is the better solution: traceable, testable, cheaper to run. This text explains how to tell the cases apart.
What you will learn
What a rule achieves
A rule is a sentence you can write down: "If the request does not contain a customer number, it goes to case handling." Such sentences have three properties that are worth a great deal in operation.
The price is that someone has to formulate the rules. But that is exactly the work that is due anyway: anyone who cannot describe their process in sentences cannot improve it either.
Where a learning method has the advantage
There are tasks where no rule can be formulated, because the connection is not expressed in sentences:
What they all have in common: the result is an assessment, not a decision. That is why every one of these cases needs a step where a person checks — and an honest answer to the question of what a mistake costs.
An example from practice
The same task, two situations
Incoming orders are to be assigned to the right department. If they arrive via a form with a selection field, the assignment is a rule — one line, always correct, changeable at any time.
If they arrive as free text by email from a hundred different senders, no rule covers all the wordings. Then a learning method is justified — with a suggestion instead of a decision, a visible confidence indication and a route for the cases where it is unsure.
The better question in this example, however, is: can the form be introduced? A process that structures the input solves the problem at the root and costs less than either automation.
Four questions before the decision
Common misconceptions
About the author
Fabian Lücker is managing partner of PHP-Schmiede KG in Rheine and has worked in software development since 2007. He is responsible for development and design, and writes here about decisions that come up in real projects.