In my last article, I wrote about what automation and AI look like inside a creative agency, and why the basics matter more than the tools. A few people came back with the obvious next question. That all makes sense, but what do we actually use it for?
So this week I'm getting practical. I'm not going to recommend specific tools, because they change every few weeks. What lasts is knowing which jobs in an agency suit automation, AI and agents, and which ones don't. I'm an operations consultant, not a technical specialist, so this is about the jobs and the checks around them rather than how the tools work.
One thing I notice in the AI conversation in agencies is how much of it is about creative work. The admin that runs the agency gets far less attention. That's where a lot of the margin is, so it's where I'd start.
Automation, AI and agents - how they fit together
I split automation from AI last week. There's now a third word being thrown around, so it's worth keeping all three straight.
- Automation - a rule makes something happen without anyone chasing it. A quote gets written, goes to a director for approval, and once it's approved it goes to the client with a reminder set to follow up in a week. A timesheet isn't in by Friday, so the person gets a reminder, and if it's still missing on Monday it goes to their line manager.
- AI - adds some thinking. It reads, summarises, drafts and suggests. You get a draft or an answer and a person decides what to do with it.
- Agents - AI that is given a goal rather than a set of steps. It works out what to do, which of your systems to look at and what to do next based on what it finds. Ask it which jobs are heading for an overrun and who needs to know, and it looks across your project and timesheet systems, works out the answer and drafts the messages. You decide the goal and what it's allowed to touch, and it decides the route.
I see these as steps on a ladder. Automation comes first because it's predictable and easy to check. AI comes next, once the process is clear. Agents come last, because the more a tool does on its own, the more you need a clear process, good information and someone checking the result.
If your process isn't agreed and your data is patchy, an agent will happily carry out a bad process for you, over and over. Get the basics right, automate what repeats, use AI for the drafting and look at agents for jobs that are well understood and low risk.
Look for jobs with lots of words and lots of repetition
The best fit is a task that involves reading, summarising, drafting or sorting information, happens often and can be checked by a person in a few minutes. Most agency admin ticks all of those boxes.
The poor fit is anything that relies on judgement, relationships or a final decision. What to charge, how to handle a difficult client and whether someone is coping with their workload should stay with a person.
Here's where I'd look.
Briefs and scoping
After a client call, someone has to turn a page of notes into a brief. AI can produce a first draft and point out what's missing. No deadline, no budget, unclear approvals, no definition of done.
The account manager still owns the brief and still checks it with the client and the creative team. But starting with a draft and a list of gaps is a lot quicker than a blank page.
Estimating and quoting
A first draft of the scope, assumptions and exclusions can come straight from the brief. If your past jobs and timesheets are in good shape, AI can also help compare a new job to similar ones you've done.
That last part is only as good as your data. If your timesheets are patchy, the comparison will be too. Approving the quote is automation. Setting the price is a decision for a person.
Meetings, status updates and handovers
Meeting note takers are one of the easiest places to start. They join the call, record it and produce a summary and a list of actions. After a client call, that gives you a record of what was agreed that you can go back to, instead of whatever people remember.
They come with a few things to sort out. Tell everyone on the call it's being recorded and get their agreement, especially clients, and check what your contracts say. And someone still needs to read the actions before they go out, because an owner or a date will occasionally be wrong.
The same approach works for a weekly status summary drafted from what's already in your project system, or a handover note when someone goes on holiday. Nobody enjoys writing these and they're often done in a rush. A draft that someone corrects in two minutes beats a note that never gets written.
Further down the ladder, an agent could work out which jobs need an update, pull what it needs from your systems and draft each one, but I'd get the draft version working first.
Resource planning
AI can help summarise who looks overbooked, prepare the overview for your weekly resource meeting and draft the messages that need to go out.
A few weeks ago I wrote that resource management isn't a desk job, and that still holds. Let the tools take the admin so your resource manager has more time to walk the studio and talk to people.
Timesheets
Reminders and approval routing are automation, and most agencies should have them in place already. AI can then help flag entries that look unusual so someone can take a look.
Some tools now try to build a timesheet from where the work happened. That's agent territory, and I'd be careful. Talk to the team first, because people don't always like feeling watched, and check the result before you invoice from it.
Scope creep and change requests
Most scope creep starts with a small request and someone saying yes to be helpful. By the time anyone adds it up, the job has overrun.
Some tools claim to read incoming requests, compare them to what was agreed and flag a likely change in scope, and some will draft the change request too. The account manager still decides, and still has the conversation with the client.
This only works if your scope is written down clearly and you have a change process the team actually follows. Another one where the basics come first.
Finance and reporting
Most founders don't need more numbers. They need someone to explain them. AI can draft the commentary on your monthly figures, such as why margin moved, which clients overran and which jobs are heading that way.
Invoice triggers and reminders for late payments are automation. Whoever knows the numbers should check anything AI drafts before anyone relies on it.
Process and documentation
In a lot of agencies, the process lives in people's heads. Last week I said to write yours down, and this is where AI can help.
Get someone to talk through how a job works, record it and let AI turn it into a first draft of written steps. Then the team corrects it. The same approach works for onboarding guides, checklists and internal FAQs.
Finding things
Agencies build up years of decisions, documents and past work. Being able to ask a question like "what did we agree with this client about amends last year?" and get a useful answer saves a lot of digging.
How well this works depends on the tools you have and how organised your files are.
Where I'd be careful
- Client confidentiality - check your contracts and what your clients allow before putting their information into any tool. Confidentiality clauses were mostly written before AI, and they rarely cover a tool receiving client information.
- Check everything - a draft is a draft. Someone who knows the subject needs to read it before it goes anywhere.
- Have a simple policy - even one page with a list of approved tools. It's the same thinking as the shiny object questions from last week.
- Creative work - the creative team should have a say in where AI fits, and it's worth agreeing with clients up front so nobody is surprised later.
- Connections - when you connect a tool to your systems, check what it can see and what it can change. Start with read only access and keep a person approving anything that goes out.
- Cost - many tools charge by how much they're used. Something that looks cheap for one person can cost a lot more once the whole agency is using it.
How to pick your first one
Run each idea through a few questions.
- Does the team do it every week?
- Is it mostly words or information?
- Could someone check the result in a few minutes?
- Is it safe with the information involved?
- Is there a clear process behind it?
- Do we know what it will cost if everyone uses it?
If most answers are yes, you've found a good first job. Try it with one team, note how long it took before and check again afterwards. It's the same approach as last week and it's how you find out whether it gave any time back.
This is not the full list
These are the places I'd start, because they're common in agencies, easy to check and closely tied to margin. They're not the only ones. What can be automated, and where AI can help, is pretty much endless, and it grows every month. Use this as a starting point and ask your own team what they'd add.
There’s no finish line with this, so keep looking at what your agency needs, and always have a structured plan for putting it in place.
