There's a whole class of problems I call "small data" - under about 100 things (note: often your sample set is a year lookback, so for example new customers might be > 100 but only 100 for this year, still small data). Anything that looks like that is often able to be handled manually and probably should be until it's sufficiently useful to automate...things that look like this often are: customers, employees, infrastructure.

In general, put off automation for anything that's not every day or month that you can't just do in a few hours , e.g.:

* Parts of performance assessment & compensation tooling

* Sophisticated recruitment analytics (especially important because with small data it's often not precise or accurate/needs manual attention)

* Provisioning new stuff that doesn't happen often: databases, clusters..etc.

* Maybe the biggest bucket of stuff in general is analytics. Often times people try to get whiz-bang end result numbers with data sets of like 100. It's almost always the wrong move to have crazy analytics abstraction layers and automation when the numbers are small.

A potential simple heuristic might be something like: if it take a day or longer to automate, you need to save at least 5 days of time in the next 6 months for it to be worth it.