Start with the work about work: weekly client status report drafts, routing new leads and requests to the right person, and handoffs between project stages. Skip strategy, complex analysis and anything that needs judgment, plus any automation nobody owns or watches. Build one at a time, in your company's name, with alerts on.
Key takeaways
- Start with the work about work: client status report drafts, lead and request routing, and handoffs between project stages.
- Skip strategy, complex analysis, final creative and anything client-facing that sends without a person reviewing it.
- Every automation needs a named owner, an account in the company's name and a check for runs that succeed but produce nothing.
- Build one automation at a time, time the task before and after, and move on only when it saves hours.
If you run a small agency, you have heard that AI and automation will give your team hours back every week. Some of that is true. The hard part is deciding what to automate first in a marketing agency, because a bad first build eats more time than it saves and can embarrass you in front of a client. This post is my opinionated shortlist: the three automations most agencies of 3 to 25 people should build first, the ones to skip, and the habits that keep any of it from breaking without anyone noticing.
Some context on where this comes from. I run marketing for a global consumer brand and build the AI systems I use every day. I have spent more than seven years in paid media, analysis and marketing ops. The list below comes from doing the work, not from a tool vendor's feature page.
Why the order matters more than the tool
Most teams start with the tool. They sign up for Make or Zapier, open a blank scenario and look for something to connect. That is backwards. Start with where the hours go.
Asana's Anatomy of Work Index 2021 surveyed more than 13,000 knowledge workers and found they spent about 60% of their time on "work about work" instead of the skilled work they were hired to do. That is 2021 data, and your team will not match it exactly. In my experience, though, the pattern holds in agencies: status updates, chasing approvals, copying details from one app to another, and asking who owns what.
That work is the best place to start because it is routine. The second reason is about what AI does well and what it does not. In a Harvard Business School and BCG study of 758 BCG consultants doing 18 realistic consulting tasks, researchers described AI's uneven skills as a "jagged technological frontier." Inside it, AI helped. Outside it, output could be wrong and could make people do worse. On one complex task picked to fall outside that frontier, people using AI were 19% less likely to get the right answer than people working without it.
So my rule is simple. Automate routine, repeatable work where a mistake is easy to spot. Keep people on the work that needs judgment.
What to automate first in a marketing agency: the shortlist
Here is the order I recommend for most small agencies. Copy the table and fill in the owner column before you build anything.
| Order | Automation | Why it goes first | Human checkpoint | Owner |
|---|---|---|---|---|
| 1 | Weekly client status report drafts | Same structure every week, easy to check, saves time on every account | Account manager reviews before it is sent | |
| 2 | Routing new leads and requests | Dropped leads cost money and the rules are simple | Backup person gets it if the owner has not acted in a set time | |
| 3 | Handoffs between project stages | Handoffs are where details get lost | Project lead confirms the kickoff checklist |
1. Weekly client status report drafts
Every account manager writes some version of the same update each week: what shipped, what is in progress, what we need from the client, and how the numbers moved. The automation pulls tasks from your project tool and key metrics from your reporting source, then has Claude or ChatGPT draft the update in your agency's voice. The draft lands in a doc or a Slack message for the account manager to edit and send.
Note the last step. The automation drafts, a person sends. A status report with a wrong number goes straight to the client, and the trust cost is much bigger than the time saved.
An illustrative example, with made-up numbers: if four account managers each spend 90 minutes a week on updates and the draft cuts that to 30 minutes, that is four hours a week back. Your numbers will differ, so time it for two weeks before and after.
2. Routing new leads and requests
When a form comes in, a new client email arrives or someone asks for a change, the work is the same every time. Create a record in the CRM or project tool, tell the right person, and send an acknowledgment. In Make, that is a short scenario. AI can help sort the request by type, such as new business, a change to a live campaign, or billing, so it reaches the right inbox.
This goes second because the rules are clear and the cost of a miss is high. A lead that sits for three days because it landed in a shared inbox is money left on the table. Add one human checkpoint: if the assigned owner has not touched it in a set time, it goes to a backup person.
3. Handoffs between project stages
The moment a proposal is signed, a set of things should happen: a project created from a template, a shared folder, an intake form to the client, a kickoff checklist and a calendar hold. The same pattern repeats when copy is approved and moves to design, or when creative is approved and moves to launch.
Handoffs are where agencies lose details. Nobody forgets on purpose. The person who closed the deal assumes the project lead saw the email. Automating the handoff makes the next step happen the same way every time.
What not to automate yet
The list of things to skip matters as much as the list to build. These are the ones I tell teams to leave alone, at least for now.
- Strategy and recommendations. AI can help you think, but the recommendation a client pays for should come from a person who knows the account. This is the work most likely to fall outside the jagged frontier.
- Complex analysis and attribution calls. Pulling the data is fine. Deciding why performance dropped and what to do about it is judgment. A confident wrong answer here is worse than a slow right one.
- Anything client-facing that sends with no review. Drafts, yes. Auto-sent emails, reports or replies to clients, no.
- Processes that change every month. If the steps are not stable, an automation locks in a version that will be wrong soon. Fix the process first.
- Tasks you do once a quarter. The build and upkeep cost more than the time saved.
- Final creative. Use AI for first drafts and variations. Keep a person on what ships.
If one of these feels like the obvious place to start, that is a sign to look at the whole workflow before building. My post on how to run an AI workflow audit walks through how to do that step by step.
The automation nobody watches
The worst automation is the one that breaks and nobody notices. I would rather have no automation than one the team trusts and nobody checks.
The tools do not make this obvious. In Make, the incomplete executions setting saves a failed run so you can fix and finish it without losing data, but it is off by default. In Zapier, when replay is on, it will not send error emails or trigger Zapier Manager alerts until the last replay attempt fails, and Zapier says it cannot guarantee replays will succeed. Those are sensible features. They also mean a broken automation can stay quiet for a while.
Then there is the harder case: the automation that runs fine and produces nothing. A filter is wrong, a field got renamed, a connected app returns an empty list. No error fires. The status report goes out blank, or no leads get routed for a week, and the run history says everything succeeded.
Two habits fix most of this. First, every automation has a named owner and a check, even a simple one, such as "if zero leads were routed today, alert the owner." Second, every account and automation lives in your company's name, not a freelancer's or one employee's. In my experience, the person who built an automation is often the only one who hears when it fails. When that person leaves, the alerts go with them.
A checklist before you build anything
Run each candidate through this list. If it fails two or more items, move it down the order.
- The task happens at least weekly.
- The steps are the same every time, and you can write them down in under ten lines.
- A mistake would be easy for a person to spot.
- There is a human checkpoint before anything reaches a client.
- One named person owns it, and a backup knows how it works.
- The account it runs on belongs to the company.
- Error handling is turned on, such as incomplete executions in Make.
- There is a check for the silent failure: the run that succeeds and produces nothing.
- You timed the task before building, so you can tell if it saved anything.
Build one automation, run it for two weeks, and confirm it saves time before you start the next. Three automations that work beat ten that half work.
Doing it yourself versus getting help
You can do all of this yourself. Pick the first item on the shortlist, run it through the checklist, build it in Make and give it a named owner. In my experience, plenty of teams get their first win that way.
Help makes sense when you are not sure what to build first, when the team is too busy to map its own workflows, or when you already have a pile of automations nobody fully understands. That is the work Keeplight does. The AI Workflow Audit includes interviews with the owner and the team, a review of every recurring workflow, and a written report with the top three automations to build first, what to leave alone, tool cleanup and one quick win. You can see a sample report to judge whether it fits. After that, Keeplight can set up a shared Brand Brain so everyone gets the same quality from AI, and Care and Build plans keep automations monitored, including the ones that run and produce nothing.
If you want to talk it through, book a free 20-minute intro call. If not, take the shortlist above and start with status report drafts this week.
Frequently asked questions
What should a small marketing agency automate first?
Start with weekly client status report drafts, then routing new leads and requests, then handoffs between project stages. All three repeat often, follow the same steps each time, and make mistakes easy to spot. Build them one at a time so you can confirm each one saves time.
Should we let AI send client reports automatically?
No. Let the automation draft the report and have the account manager review and send it. A wrong number in a client report costs more trust than the automation saves in time.
Is AI good at strategy and analysis for clients?
It can help you think, but I would not automate it. In the Harvard Business School and BCG study, people using AI on a task outside what AI handles well were less likely to get the right answer. Keep strategy, attribution calls and recommendations with a person who knows the account.
Should a small agency use Make or Zapier?
I default to Make, but both can work for the shortlist in this post. What matters more is that error handling is turned on, each automation has a named owner, and every account is in the company's name. Pick the one your team will maintain.
How do I know if an automation has quietly stopped working?
Add a simple check for the result, not only for errors, such as an alert if zero leads were routed today or a report came out empty. Some failures throw no error at all because the run technically succeeded. Give every automation an owner who looks at it on a set schedule.
How many automations should we build at once?
One. Time the task before you build, run the automation for two weeks, and confirm it saves time before you start the next. Three automations that work beat ten that half work.
Bryce Walls, Founder, Keeplight
Bryce runs marketing for a global consumer brand and builds the AI systems he uses every day. He has run paid media, analysis and marketing ops in-house for more than seven years, and Keeplight sets up the same kind of AI and automation for small marketing teams.