Why this matters now
Marketing chiefs looking at what they pay agencies are "still not seeing the cost savings in terms of fee" from AI. That's Gartner's Jay Wilson, speaking to Glossy in April 2026.
Every hour AI saves is an hour you can't bill. If you do nothing, the savings slowly go to the client, one renewal at a time.
Name the work clients pay you most for. Count the hours that go into it. Then decide, in writing, what to do with the time you get back.
The method
Try it on one team or one client first. Each step gives you something you can show your partners.
Move one
Write down two weeks of real work as a list of tasks. Everyone wants to skip this part. It's also the part only you can do.
Be specific
"Account management" is a department. "Draft the weekly status note for the client" is a task. You can time it, and you can make a decision about it.
Aim for 25 to 40 tasks per role. Fewer than 20 usually means you're still listing departments.
The one column that matters
That number drives every step after this one. It's also the proof your partners will trust at the end.
Include the hidden work too: formatting, chasing approvals, hunting for the deck you already made. That's where unbilled hours hide.
Move two
Ask three questions about each task. If any answer is yes, a person may need to keep it. Click a task to see where it lands.
Question 1 · Judgment
Does it mean deciding what the story should be? Just writing it up doesn't count.
Deciding versus drafting.
Question 2 · Relationship
Does it only work because a client or reporter trusts you personally?
Would it be worth less coming from someone else?
Question 3 · Accountability
Do you make the call yourself, on the spot? Signing off on someone else's work doesn't count.
Your call, or your sign-off?
Senior people tend to say yes to almost everything. So for each role, rank the yes answers by how much they matter to the client, and keep the top five. Everything else, plus the prep behind those five, is fair game for AI.
The one idea to remember
Almost every task you keep has hours of prep behind it. Clients pay for the hour you spend in front of them. The prep is where your profit leaks.
The client meeting is yours. The prep behind it doesn't have to be.
This keeps the whole thing honest. Without it, "keep" becomes an excuse to change nothing. With it, the work you keep becomes the best place to start, because it has the most prep behind it.
Move three
There are three kinds of help you can set up. The right one depends on how often the task happens. Drag the slider to see.
52 times a year
Turn this on when a step can't be written as simple instructions. For example: "decide which of these 40 articles matter" instead of "copy this into that."
The rule
Match the tool to how often the task happens. If it happens a lot, set up autopilot, because the setup pays for itself. If it's rare, just ask AI each time, because setting something up would cost more than the task ever will.
Step-by-step instructions a computer follows, with no AI. When an email lands in this inbox, copy these details and post them in the team chat. Takes time to set up, costs almost nothing to run, works the same every time.
Autopilot with AI added to one step that needs reading. Summarize this. Flag which of these matter. Medium cost to set up and to run.
You tell AI the goal and it figures out the steps. Nothing to set up. But you pay every time you use it.
Bars show the total cost over a year, setup plus running. They're rough comparisons, not real prices. The shortest bar wins.
A real example
A team set up an AI assistant to watch the news for their clients. It read every alert and sent the important ones to the team chat. It worked, but the AI bill ran far higher than planned. The fix was simple autopilot reading one inbox. Same job, more reliable, a fraction of the cost.
They paid AI to think through a job that ran a thousand times a day.
You'll want to hand off the task that annoys you. Start with the one that repeats.
Move four
Set up one thing, not three. Teams that try three at once usually finish none. Then count honestly, including the two costs people forget: checking the AI's work and setting it up.
Count it honestly
Hours saved = hours now − (hours after + checking + setup ÷ months). Value = hours saved × 12 × cost per hour.
Want this for every task on your list?
Before you set anything up
Write it down, with a number. An agency has three honest choices.
Change how you charge so clients pay for results instead of hours. Then the savings stay with you.
Use the time for senior advice, measuring results, or the service clients keep asking for that you keep turning down.
Give your people real time away from billing to learn this. They can only build the skill on real client work, and a full billing target leaves no room for it.
Hours you don't claim get claimed for you.
Usually at renewal, by the client's purchasing team.
For whoever owns the budget
Ask the three questions across a team and you get a clear list. It shows which work is routine and which is the advice clients pay top dollar for. That's exactly what you need when you sit down to talk about price.
Agencies that skip this step tend to find their answer in the client's cost-cutting slides. Agencies that do it get to set the terms: here's what our people still do, here's what AI now handles, and here's what that's worth to you.
For whoever runs the agency
One more question for whoever runs the place: where are people stuck, and what can't you afford to get wrong? Use this chart to spot risk. Use the hours from step one to decide what to tackle first.
High stakes · Many can do it
A named person signs off
AI helps, and one named person approves it. A real name, instead of "the team checked it."
High stakes · Only one person can do it
Your biggest risk
Only one person can make these calls, and one bad call could cost you the client. You need a backup before you need AI, because this risk walks out the door when your best people leave.
Low stakes · Many can do it
Hand it off
Low risk, and lots of people can do it. Set it up once and stop debating it.
Low stakes · One person can do it
Hand off to free someone up
A senior person is stuck on routine work because only they know the steps. Freeing them up is your fastest win, and usually the easiest to get approved.
This builds on two old ideas. Economists have been breaking jobs into tasks for years, and AI companies have long said to use the simplest tool that works. What's new is applying both to a real agency week.
The question that's coming
Someone is going to run these numbers on your retainer. The leader who shows up with the list sorted, the hours counted and a plan for the saved time is the one who sets the terms.
The guide includes the full worksheet, a chart for deciding what to set up first, and notes for running it with your team. Back to all resources
One issue a week. One AI development that matters for comms, broken down into what it means for your budget, your team, and your next board conversation.