How Do You Run a Marketing Agency?
I run a marketing agency by identifying the repetitive tasks that burn the most hours and automating them first. Victor taught me this after years of running his own agency. We build AI skills for each workflow, test them on real client work, then scale what works.
Start with the bottleneck. Find the task your team does every week that follows the same steps and eats the most hours.
Then automate that one task. Not five. Not an entire agency OS because some kid on YouTube built one in five minutes. One task.
Test it. Fix what breaks. Get your team used to it. Then move to the next one.
Three Guys Beat the Enterprise
Vinod my co-founder works at a large company during the day. His team wanted to use Claude for a project but could not. A Microsoft enterprise agreement locks them in.
To switch to Anthropic they would need a master agreement. That takes months.
Victor my co-founder heard this and laughed. "That's why we can pack big companies now. They're so slow and so complex."
"Three guys can outpace them."
He was not bragging. That is what we live every day at Sucana.
Three founders building an AI analytics platform for lead gen agencies. No 50-person team, no procurement department.
When we want to try a new AI model, we try it that afternoon.
The speed difference is not about talent. It is about permission.
Big companies need six meetings to approve a tool. A Slack message is all we need.
That is the real advantage of building an AI-powered agency. Not the AI itself. The speed at which you can move when you are small and willing to change.
I wrote a full guide on how to build an AI-powered marketing agency from the ground up. This article is different. This is about the order that works, the mistakes that waste money, and what I learned building Sucana.
Where AI Fits (And Where It Fails)
The first mistake I see agencies make is looking at what AI can do and trying to find places to use it. I made that mistake myself.
What works is the opposite. Look at where your team wastes time. Then ask if AI can take that off their plate.
The best tasks for AI share four things. Weekly cadence. Same steps every time.
Data inputs, not judgment calls. And output the client sees.
Reporting fits all four. Creative strategy fits zero.
I started with reporting. Victor my co-founder spent two hours per client every Monday pulling numbers from Facebook Ads Manager into Google Sheets.
Five clients. Ten hours. Every single week.
That is the kind of task AI was built for. I wrote the full system for automating client reporting with AI. It cut the time to about 40 minutes per client.
The key was not replacing Victor. It was giving him back the hours he was wasting on data entry. That distinction matters -- the guide to the AI skills marketers need now shows it is the tasks that disappear, not the people.
The Order That Works
I have watched agencies add AI in every possible order. Most do it wrong.
Five tools on the same day. Performance Max turned on. All copy from ChatGPT.
Automated bidding across everything. Then they wonder why results got worse.
The order matters more than the tools.
Data first.
Before AI can analyze anything, your data needs to be clean. Consistent naming in your ad accounts.
UTM parameters that match your CRM. One source of truth for leads.
I spent two weeks cleaning up our naming before I let any AI touch the data. It felt like wasted time. It saved me months of wrong answers.
Analysis second.
Once your data is clean, feed it to AI for analysis. Not execution. Analysis.
Let the AI read your campaign data and tell you what happened. Phase transitions. Geographic spending patterns.
Creative fatigue signals. The kind of detail that takes hours to find manually.
This is where AI earns its keep. It reads a full quarter of data in seconds.
Victor saw this when our AI found a geographic scaling trap he missed. Spain's budget grew 37 times but leads only increased 9 times. Eight years of experience and he did not catch it.
Execution last.
Only after your data and analysis layers work do you add AI to execution. Ad copy and bid adjustments.
Creative testing at scale comes last.
The agencies that skip straight to execution are on Reddit complaining that "nothing is reliable." They are right. AI execution without clean data is a liability.
What It Costs (Honest Numbers)
Most articles about AI agencies skip the cost conversation. I will not.
The tools are cheap. Claude costs $20 a month, ChatGPT is the same.
Most automation platforms have free tiers.
The expensive part is setup time. That is the cost nobody talks about.
I spent four weeks building our first AI reporting workflow. Hitting walls, rebuilding prompts, fixing data formats, starting over. The frustration was the education.
After those four weeks, the workflow saves ten hours a week. The math works out fast. But the setup period is real.
The second hidden cost is human review. AI does not produce finished work. It produces drafts that need a human eye.
Every AI report still needs someone to check the numbers. Context the AI cannot know has to come from a person.
I budget about 30% of the time savings for review. If AI saves an hour, the review takes 20 minutes. Net savings are real but not as dramatic as the marketing says.
How to Pitch AI to Clients
This is the part most agencies get wrong. They lead with the technology.
"We use AI to run your campaigns." That sentence makes clients nervous. They picture a chatbot replacing their account manager.
I wrote a full guide on how to build an AI-powered marketing agency. The short version: lead with the result, not the tool.
"We catch campaign problems in three days instead of two weeks." That is the pitch. How you do it is the backstory.
Victor showed one of our AI reports to his client Pablo. The reaction was not about AI.
Pablo asked when the next report was coming. Depth of insight sold itself.
The agencies losing client trust are the ones explaining AI before showing results. Show the report first.
How to Get Your Team to Use AI
Buying AI tools is easy. Getting your team to use them is the hard part.
I wrote a full guide on marketing team AI adoption strategy. The biggest lesson: do not start with training.
Start with one person and one task. Pick the team member most frustrated with their workload.
Then identify the task they complain about every week.
Build one AI workflow for that task. Let them use it for 30 days.
When they tell the rest of the team it saved five hours a week, adoption happens on its own. Nobody needs a training deck when a colleague leaves the office two hours earlier.
The mistake is rolling out AI to everyone at once. That creates resistance.
One person, one task, one win. Expand from there.
How to Measure Whether AI Is Working
Most agencies add AI and never track the difference. They feel faster, think results improved.
Nobody puts a number on it.
I track three things.
First, time per task. Before AI versus after. Put those numbers on a spreadsheet and track monthly.
Second, error rate. Not "slightly off" but wrong enough that a client would notice. If that number is climbing, your prompts need work.
Third, client feedback. Not surveys, direct reactions.
Pablo asked for his next report unprompted. That was the signal.
If your AI tools do not move at least one of those three numbers, they are adding complexity without value.
I built a full system for proving marketing ROI with AI. The same principles apply to measuring AI inside your agency.
The Part Nobody Talks About
AI breaks. Often.
We opened Sucana for testing on March 2. Full team call.
Victor typed a message and the screen reset. Nothing.
Vinod checked the backend. Same error.
It was not us. Claude's API was down everywhere.
Every business built on that tool lost an hour of their day.
That moment taught me something. Building on someone else's technology is a dependency. You need a plan for when the AI goes down.
For us, Victor can still run his reports manually. The old process is slower, but it works. We parked the boats nearby instead of burning them.
The agencies that go all-in on AI with no manual fallback are one outage away from missing a client deadline.
What an AI-Powered Agency Looks Like in Practice
The daily reality is less dramatic than the articles make it sound.
Monday morning, I check what the AI analyzed over the weekend. Campaign performance, geographic spend patterns, creative fatigue signals.
Fifteen minutes instead of two hours.
Then I look at what the AI flagged. A CPL spike in one market, a creative running three weeks without a refresh.
Budget that shifted to Display when the client wanted Search.
The AI surfaces things I might miss. I make the calls.
That is the model. A layer of AI doing the heavy data work so humans can focus on judgment and client relationships.
Victor put it well. The AI does not just recommend the best-performing number.
It recommends the one you can iterate on without breaking your production budget.
That kind of thinking separates a useful AI workflow from a tool that sounds good in a demo.
Where to Start
If I had to start over today, I would do four things in the first month.
Week one: find the bottleneck. List every task your team does weekly. Pick the one that takes the most hours, follows the same steps, and produces output clients see. I walk through this exact process in my guide on where agencies should start with AI automation.
Week two: automate that one task. Use Claude, ChatGPT, or whatever tool fits. Build a simple workflow. It will not be perfect. That is fine.
Week three: test and fix. Run it alongside the old process. Compare the output. Fix what breaks. Let your team get comfortable with it.
Week four: measure. Put the old time and the new time side by side. If the numbers work, roll it out for real. If not, fix it before moving on.
One month. One task. One win.
Do not build an agency OS. Do not automate five things at once. Do not follow the YouTube kid who says he built it all in a weekend.
Start small. Get your people used to it. Then expand one task at a time. When you are ready to go further, I documented how I built a full AI brain for the agency using Claude Code — the architecture, the skills, and how it all connects.
Frequently Asked Questions
How do you build an AI-powered marketing agency from scratch?
Start with one repeatable task, not with AI tools. Identify the workflow that eats the most hours every week, usually reporting or data analysis.
Build one AI workflow for that task. Test it with one client for 30 days.
Measure the savings, then expand.
Most agencies use Claude or ChatGPT for campaign analysis and ad copy. Automation platforms like n8n or Make handle data connections between ad platforms and reporting tools.
The tools matter less than how you use them. A $20 Claude subscription with good prompts beats a $500 platform with generic ones.
How do you pitch AI services to agency clients without scaring them?
Never lead with the technology. Lead with the result. "We catch campaign problems in three days instead of two weeks" is a pitch.
Show clients a report with depth they have never seen. The reaction will be about the insight, not about what produced it.
How much does it cost to add AI to a marketing agency?
The tools cost $20 to $40 per month. Automation platforms often have free tiers for small volumes.
The real cost is setup time. Expect four to six weeks of building and testing your first workflow. Budget 30% of time savings for ongoing human review.
Will AI replace marketing agencies?
AI will replace tasks that involve pulling data, formatting reports, and generating ad copy at scale. Those are already faster with AI.
Strategy, client relationships, and judgment calls about what the data means are still human. The role changes shape, but agencies that add AI get stronger.
How do you automate client reporting with AI?
Pull campaign data into a structured format. Feed it to AI asking three questions: what happened, why, and what to do next.
The process takes about 40 minutes per client instead of four hours. AI handles data reading while you add context and judgment.
What tasks should agencies automate with AI first?
Start with reporting. It runs every week, follows the same format, and the output is visible to clients.
Stay away from creative strategy and client communication early on. AI needs patterns and structured inputs, and strategy gives it neither.
How do you train a marketing team to use AI?
Do not start with training. Start with one person and one task. Pick the team member who complains most about their workload.
Build one workflow. Let them use it for 30 days. When they share the time savings, the rest of the team follows.
How do you price AI-powered agency services?
Do not discount because AI does some of the work. The value to the client stays the same no matter what produced it.
Price based on the outcome. Deeper analysis, faster problem detection, better recommendations. Catching a budget leak in three days instead of fourteen is worth more, not less.
What is the ROI of AI for marketing agencies?
AI cuts reporting time by about 75% in our experience. What took two hours per client now takes about 40 minutes.
The real ROI is what you do with the saved time. Deeper analysis, faster problem detection, and catching issues before clients notice them.
What happens when the AI gives wrong answers?
It happens regularly. AI produces drafts, not finished work. Every output needs a human check.
Track error rates monthly. If wrong answers climb, your prompts need better context. Feed the AI your client's business details, not just campaign data.
Can a small agency compete with large agencies using AI?
Small agencies have the advantage. New tools get adopted the same day. Large agencies need months of procurement and approvals.
Three people with AI and clean data can produce analysis that matches a ten-person team doing everything by hand.
How long does it take to see results from AI in an agency?
Reporting improvements show up in the first week. Your first AI report will be faster and more detailed than your manual process.
Campaign performance improvements take 60 to 90 days as the AI spots patterns across enough data.
What is the biggest mistake agencies make when adding AI?
Automating everything at once. AI bidding, AI copy, AI reporting, and AI creative testing in the same week. Impossible to tell what is working.
Add one AI layer at a time. Start with reporting, measure the result.
Then add the next layer.
Do clients care whether their agency uses AI?
Most clients care about the result, not the tool.
When they see their own data explained at a depth they have never experienced, the AI question disappears. Results first, always.
