Will AI replace marketing agencies? For most clients, not on the numbers. AI replaces tasks: set-up, QA, assembly, summarising. It does not replace accountability, relationships or judgement, and once you price those in, the "cheaper" in-house AI option often costs more. Where AI does win is against agencies that sell admin as expertise. That distinction also answers the quieter question underneath: whether AI makes you more creative or simply redundant.
The debate got a proper costing this week. Writing in Mumbrella, AZK Media founder Azadeh Williams takes on the claim that an AI platform plus an in-house hire is cheaper than a PR agency. Her illustrative comparison sets a $10,000-a-month retainer, about $120,000 a year, against a $20,000 AI subscription plus an $80,000 to $100,000 PR hire before employment costs. Add media databases, monitoring and specialist support, and she argues a sophisticated in-house set-up can "easily reach north of $220,000."
What does the maths actually say?
The trap is comparing the price of a tool with the cost of a team. As Williams puts it, "you can't compare the price of one software subscription with the cost of an entire agency and call it a saving." The subscription is the cheapest line on the spreadsheet. The expensive lines are the person who runs it, the person who checks it, the tools around it and the senior judgement nobody budgeted for.
Run the sum yourself. The subscription plus one hire comes to roughly $100,000 to $120,000 a year before on-costs, extra tools or management time. That is already level with the retainer, and the client has swapped a team's range for one person's. If that person is off sick, resigns or simply does not know the press in a new market, the saving evaporates.
Williams is writing about PR, but the logic travels. Swap "PR hire" for "performance marketer" or "social lead" and the shape of the sum looks much the same.
Where does the maths work?
Williams is not defending every agency. She argues that agencies charging high fees for generic releases, recycled pitches and clippings-based reporting are vulnerable, and that "clients shouldn't be paying premium fees for administrative inefficiency." That is the honest answer to the headline question. AI will replace agencies whose value was mostly process.
A Forbes Agency Council panel of agency leaders lands in a similar place. Multiply's Matt Jayson singles out campaign set-up and QA, loading and checking ads inside the platforms, as the agency work that gets automated: "For one high-growth client, that's easily 20 hours a week of invisible work." Pierce Kafka of Kafka Media Group argues that as reports become automatic, clients will need "human interpretation that explains why the data matters, what it reveals and what to do next."
Now look at the chores most agencies quietly loathe: timesheets, monthly reports, status meetings, budget tracking, manual resizing, invoice chasing, competitor audits. None of them is why a client hired you. All of them are why your people go home late.
Why don't clients trust AI on its own?
Psychologists who study trust tend to break it into three questions. Is the other party able? Do they mean well? Will they be straight with me? Trust Theory, in plain words, says we only rely on someone when all three answers are yes.
An algorithm can demonstrate ability. It struggles with the other two. It cannot mean well, and it cannot be held to account when it is wrong. That is why a client will happily let AI draft a hundred headlines but wants a named person to sign off the one that runs. It is also why Williams argues that AI can help find journalists and analyse coverage, while only people build the trust that wins editorial opportunities. Trust is bred in conversation, not in the data of algorithms.
The same holds inside the agency. Teams distrust AI outputs they cannot question. They start trusting them when they can interrogate the data like a colleague: ask a follow-up, challenge a conclusion, see the posts behind a claim. That is the thinking behind Defy MCP, which lets teams question client and category data inside Claude, prompting it like a co-worker rather than learning another dashboard.
Which agency tasks should AI take, and which should it leave alone?
| Agency work | AI's role | Why |
|---|---|---|
| Campaign set-up and QA | Take it | Repetitive and rule-based; Jayson puts it at 20 hours a week for one high-growth client |
| Monthly report assembly | Take it | Gathering and formatting data adds no judgement |
| Competitor audits and monitoring | Take it | Machines read a whole category faster and more consistently |
| First-draft copy and variants | Assist | Useful for volume; people choose and refine |
| Strategy and the creative idea | Assist with evidence | AI supplies the research; people make the leap |
| Media and stakeholder relationships | Leave alone | Trust is built person to person |
| Regulated claims sign-off | Leave alone | Someone accountable has to be on record |
The pattern is simple. The more a task depends on rules, the more of it AI should take. The more it depends on trust, the less.
Can AI make you more creative, not just faster?
Yes, but only if you decide what the time is for. Saved hours have a habit of turning into extra meetings. Here is a straightforward way to make sure they turn into better work.
- Log a fortnight of real work. Every task, every person, in half-hour blocks.
- Sort each task into chore, craft or relationship. Be honest. A surprising amount of "strategy" turns out to be formatting.
- Automate one chore end to end. The monthly report is usually the best first candidate.
- Ring-fence the hours you get back. Assign them before you save them: creative development, client workshops, testing new ideas.
- Keep a named person on everything client-facing. AI drafts; people sign.
- Make outputs checkable. If the team cannot see where a conclusion came from, they will not trust it, and neither will the client.
- Tell the client what changed. Show them the hours that moved from chores to thinking. It is the strongest answer to "couldn't AI do this cheaper?"
What should you say when a client asks if AI could replace you?
Show them the maths, then show them your chore list. Agree that AI should do the repetitive work, because it should. Then be specific about what they are paying you for: judgement, relationships, accountability and ideas. If you cannot name those clearly, the client's question is a fair one.
Williams closes with a line worth borrowing: "The future isn't AI versus PR agencies, it's AI-powered PR agencies versus outdated PR models." Replace PR with your own discipline. That is the real contest, and it is one agencies can win. It is also the split behind the Defy platform: machines take the chores, people keep the judgement.
