What could AI actually do in your job?
What AI could do in operations
Six kinds of thing, in order. The first three you can start this week with the tool you already pay for. The next two need building. The last you should buy.
Basic promptsprompt ops
Ask the chat tool you already have in plain words and check what comes back. No set-up. It works today.
What it takes. A subscription, ten minutes of practice and your own judgement on every answer.
- A recorded walkthrough turned into a written procedure
- A long supplier or client thread summarised into a status line and a next step
- Meeting recordings turned into actions with owners and dates
- And hundreds more. Anything you'd ask a bright new starter to take a first pass at.
Advanced promptscontext ops
Give it what it needs before you ask: the brief, the examples, the rules, the data. This is context engineering, and the quality comes from what you feed it rather than how you word it.
What it takes. The right material gathered in one place and a clear idea of what good looks like.
- A procedure checked against your own standards and the last incident report
- The weekly status pack drafted from the updates people already send, in your template
- Policy and contract wording checked against your own rules before it goes out
- And hundreds more. Anything where you already hold the material and the judgement.
Skillsskill ops
A job written down once so anyone on the team gets the same result every time: the steps, the rules, the examples, saved and shared.
What it takes. Someone to write it and own it, and training so the whole team uses it rather than the keen few.
- A procedure-writing skill so every process doc has the same shape
- An incident write-up skill: paste what happened and get the same five-section report
- A status report skill everyone uses so the pack builds from the same inputs
- And plenty more. We run over 180 of our own.
Automationsauto ops
The same job running on a trigger with nobody typing a prompt: when a call ends, when a lead lands, every Monday at eight. Connected to the tools you already run.
What it takes. The process written down step by step, test cases, data you're allowed to connect to, and a build.
- The weekly status pack built on a trigger from the systems with nobody chasing
- New jobs or tickets enriched and categorised the moment they land
- Records kept current from emails and calls with nobody typing
- And more. We've built over 50 of them.
Agentsagent ops
Software that makes a bounded decision inside a workflow that already runs, with a written job, tests and a person at the gate. The last thing you add rather than the first.
What it takes. The automations underneath it already running, a fence round what it may never do, and someone watching it.
- Work routed to the right person without anyone looking, with the exceptions flagged
- Risks called before they become incidents with a person deciding the response
- Supplier changes judged for impact and escalated when they matter
- And a few more. We've built them and every one is custom to the business. Because an agent is left to run on its own, there are fewer good use cases than you'd think. For now.
Tech built for the jobbuy ops
Some jobs already have a product that does them well. Buy it, plug it in, and keep the build budget for the things nobody sells.
What it takes. A subscription and an hour to set it up. Check what it does with your data first.
- A call recorder that writes the notes
- A document tool that reads invoices and contracts into your system
- A ticketing tool with AI triage built in
- And more every month. If a product already does the job well, this is the list it belongs on.
If the ones you want sit below the line, that's a build, and it's the conversation we're set up for. If they sit above it, start this week and come back when the list gets longer.