Most firms we talk to ask the same thing first: which part of our work could agents actually do? The answer rarely depends on the industry. It depends on two properties of the job itself.
1. Is there a checkable right answer?
A claim code is right or wrong. A shipment classification is right or wrong. A reconciliation either balances or it doesn’t.
If a senior person on your team can look at the output and check it against rules, then agents plus a written rulebook can do most of the work. The rulebook records what “correct” means and every way the output has gone wrong before. Each mistake the agents make goes into it, so the same mistake doesn’t happen twice.
If the only test is “does the partner like it,” the job isn’t ready yet. That’s usually a sign the rules exist but live in someone’s head. Writing them down is the first piece of work.
2. Does it repeat?
Hundreds of similar jobs a month is what makes the rulebook worth writing. A job you do twice a year never pays back the effort of encoding it. A job you do four hundred times a month pays back within weeks.
Volume also makes the review step cheaper over time. At the start, your experts check every output. As the rulebook grows, more work goes straight out, and experts only see the cases that carry money, legal exposure or reputation.
What doesn’t qualify
Work that depends on one-off judgment keeps a person in charge. That includes a negotiation, a new kind of case or a call on a client relationship. For that work we build tools that prepare the decision: they gather the documents, flag the gaps and draft the options. They don’t make the call.
A quick test
Pick the job your team repeats most. Ask:
- Could a senior person tell you, in a minute or two, whether a finished job was done right?
- Does it happen at least a hundred times a month?
Two yeses make it a good candidate. One yes usually means there’s groundwork to do first, most often writing down the rules.