The first question every operator asks us is some version of "can your agents do X." It is the wrong question, and answering it directly is the fastest way to deliver something that does not work in production.
The right question is whether X is what the team actually does. In our experience the answer is almost never a clean yes. The work happens in detours, in the email thread the SOP does not mention, in the spreadsheet a senior controller maintains because the ERP cannot, in the five minute conversation that resolves the exception nobody documented. An agent built from the org chart description of the work will pass a demo and then fail on its first real cycle.
So we observe first.
Two weeks, one cycle
For two weeks our agents sit alongside a real operations team and watch a complete cycle of work. Not a sample. The whole month-end close, the whole AP run, the whole order-to-cash loop. We capture what is opened, what is typed, what is copied between systems, where the handoffs happen, and where the exceptions land.
The output of those two weeks is not a report. It is a map of the work as it is actually performed, with timing, frequency, and the cost of each step. Often this map is the first time anyone has seen the operation written down at this level of granularity.
Then we scope
With the map in hand the conversation changes. Instead of "can your agents do AP," it becomes "of the eleven tasks in this AP cycle, which three account for sixty percent of the time, and which of those three are amenable to an agent today." That is a question with an answer.
The work that gets automated first is rarely the work that looked most automatable from the outside.
This is why the observation phase is not a deliverable we can skip. It is the part that makes the rest of the engagement possible.
What this is not
We are not building a process mining product. We are not selling dashboards to a buyer who will then go figure out what to do with them. The map exists so we can build the agents. The agents exist so we can run the work. The whole thing is one engagement with one outcome.
If you are running operations in a business that cannot hire an AI team, and the consulting firms quoted you a six month engagement and a deck, we should talk.