The useful test is not whether a model can do a task, it is whether somebody can verify the output in seconds. Where verification is fast, automation pays immediately. Where verification takes as long as doing the work, it costs more than it saves and introduces errors nobody catches.
Research and enrichment, which is the clearest win
Finding which companies match a profile, what they recently announced, who holds a relevant role and what they use already is slow manual work with a checkable answer. A wrong result is obvious in a second.
This is where most of the realised value sits today, and it is unglamorous: the reason a team gets through four hundred accounts instead of eighty is not a cleverer email, it is that the list was built in an afternoon.
Summarising and recording, which nobody wants to do
Call notes, meeting summaries and CRM updates are the tasks that do not get done, and their absence is the reason most pipelines are untrustworthy. Automating the record is more valuable than automating the outreach, because it fixes the data everything else depends on.
The check is fast: the person who was on the call reads four lines and corrects one. That is a workable loop.
Reporting, once the definitions are fixed
Pulling the same numbers from four systems into one view every month is mechanical, and the monthly rebuild by hand is why the answer always arrives a week after it was useful.
The prerequisite is that the definitions are stable. Automating a report built on a metric that three people define differently produces a confident wrong answer faster, which is worse than the slow version.
Where it should not go yet
Fully automated outreach written and sent without a human reading it is the clearest mistake. The failure is not grammatical, it is that the message is plausible and wrong about the company, and the recipient cannot tell the difference between that and contempt.
The same applies to anything that makes a judgement with a slow check: forecasting, deal scoring that nobody can interrogate, pricing. Use it to prepare the judgement, not to make it.
Where to start
Pick the task your team complains about most that has an obviously right answer, and automate that one. In most B2B teams it is list building or the CRM record, not the writing.
Then measure the time it gave back rather than the technology. If nobody can name what they now do with the hour, the automation has not landed, whatever it is producing.