Writing a playbook a machine will not misread
The rules you keep in your head are more conditional than you think. Getting them down is most of the work.
Ask an experienced buyer for their rules and you will get something like: if frequency goes over three, refresh the creative. Clean, memorable, and wrong about half the time in a way the person saying it already knows and did not mention.
They did not mention it because the exceptions are not rules to them. They are judgement. The gap between the rule as stated and the rule as practised is where every automation project runs aground.
Unpacking one rule
Push on that frequency rule and the conditions come out one at a time. Not during a launch week, because frequency is meaningless before delivery stabilises. Not on retargeting, where three is normal and seven is fine. Not if the ad set is under a few hundred dollars a day, because the fix costs more attention than the problem. Not if there is already a refresh in review. Not on the brand campaign, ever, because that one is measured differently.
One sentence became six conditions, and the person holding them could not have listed them cold. They surface when a situation triggers them, which is exactly what makes them hard to write down and exactly why the writing down is worth doing.
Four questions that pull the conditions out
The reliable way to get a rule out of someone's head is to ask about the boundaries rather than the rule.
- When did you last not do this, and what made that case different?
- What would have to be true for you to wait another week instead?
- Which accounts or campaign types does this simply not apply to?
- What do you check before acting, that you would not describe as part of the rule?
The fourth one surfaces the most. Almost every rule has a silent precondition, usually a glance at something on another screen, and the glance never makes it into the written version.
Write thresholds, not adjectives
A rule that says to act when frequency is high cannot be applied by anyone but its author. High is a number that person is holding and has not said. The same goes for stable, expensive, recent, and significant.
Every one of those has a value behind it. Recent means the last fourteen days. Expensive means over the acquisition cost ceiling for that campaign. Getting the values out is uncomfortable, because a written threshold can be wrong in a way an adjective cannot, which is precisely the property that makes it useful.
Say what the rule does when it is unsure
The last piece is the one most playbooks omit. A rule needs a behaviour for the case where its conditions are only partly met, and the default should almost never be to act.
Three defaults cover nearly everything: do nothing and say so, re-read in a stated number of hours and decide then, or draft the change and wait for a person. The third is the right default for anything touching budget, and a playbook that does not name one has effectively chosen the first by accident.
A playbook written this way is longer than the version in your head and duller to read. It is also the first version that survives being handed to someone else, which is the only real test of whether the rule was ever a rule.