AI for operations teams

Why operations first

Not because the work is less valuable — because of its shape. Ops work tends to repeat, arrive as text, and have an obvious right answer that someone can verify in seconds. Those three properties are exactly what current AI handles well, and they are much rarer in strategy, design or sales relationships.

There is a second reason, less often stated: operations people are usually the ones who know where the time goes. They can name the workflow that costs the most without a discovery exercise.

What moves

Inbox and enquiry triage

Sorting what arrives into what needs a person, what can be answered from existing material, and what is noise — then drafting the answerable ones. The volume is high, the cost of a mistake is low because a person still sends, and the time back is immediate. Usually the first deployment: inbox triage and follow-ups.

Data movement between systems

The enquiry into the CRM, the order into the spreadsheet, the ticket into the project tool. Nobody was hired to do this and everybody does it. Worth noting: where the path is fully deterministic, a workflow tool is cheaper — automation vs agents. Agents earn their place when the input varies enough that a flowchart would need a branch per case.

Coordination

Proposing times, chasing confirmations, rescheduling, sending reminders. Genuinely tedious, entirely rules-plus-judgement, and the kind of thing that decays quietly when a team is busy.

Recurring reporting

The weekly summary that happens fortnightly. An agent on a schedule produces it every time, which is worth more than producing it better. This is also the cleanest demonstration of scheduled work, where most of the compounding lives.

Supplier, order and exception chasing

Following up on what has not arrived, flagging what is late, noticing what changed. Monitoring is the class of work humans are worst at sustaining and machines are unbothered by.

What does not move

Being specific matters more here than anywhere, because overreach in ops is how a team loses confidence:

• Judgement about people. Which supplier to trust, which customer needs a call from a human, when to make an exception.
• Negotiation. An agent can prepare the position. It should not hold it.
• Anything where the right answer depends on context nobody wrote down. Until it is written down, which is a project of its own.
• Final approval on anything that leaves. Not because the agent cannot draft well, but because accountability sits with a person.

Where the boundary goes

For an ops deployment the useful default is three tiers:

• Autonomous: reading, searching, summarising, drafting, updating internal records that are easy to correct.
• Gated at first, widened later: creating records other systems depend on, scheduling with third parties, internal notifications.
• Always gated: anything leaving the company, anything financial, anything that creates a commitment.

Start everything one tier stricter than you think necessary. Widening is a two-minute change; rebuilding trust after an incident is not.

What it feels like when it works

Not "the team is smaller". The honest description from teams running this is that the floor rises: the things that used to fall through on a busy week stop falling through, the weekly report exists every week, enquiries get answered at 11pm, and the ops person spends their day on the exceptions rather than the queue.

That is capacity, not headcount, and it is the framing worth taking to whoever approves the budget — see agent team vs human team.