Before you automate a marketing process, classify it on two axes: volume (how often it happens) and complexity (how much judgement, exceptions and systems it involves). Where they cross tells you what to automate now, what to simplify first and what to leave with people. It’s the exercise that saves me the most misdirected work.
With tools like Make or Claude, building an automation keeps getting easier. That’s exactly why the risk has changed: it’s no longer «I can’t automate this», but «I automated the wrong thing».
Why classify before you automate
Automating a broken process just breaks it faster. If the lead-to-sales handoff depends on three spreadsheets and someone remembering to do it, an automated flow will simply repeat that confusion at higher speed.
There’s a broader signal too: Gartner predicts that over 40% of agentic AI projects will be cancelled by the end of 2027, due to rising costs, unclear value or inadequate risk controls. Not all of that is the technology’s fault. Very often teams start with the tool instead of the process.
The two axes: volume and complexity
Volume: how many times a month the process runs. Weekly reports, lead creation, campaign tagging, social publishing… The more it repeats, the more it pays to remove the manual work.
Complexity: how much human judgement it needs. I score it with three questions: does it have lots of exceptions? Does it touch more than two systems? Is a mistake expensive (for the brand, the budget or a customer relationship)? Two yeses out of three means it’s complex.
The matrix and what to do in each quadrant
| Quadrant | What it means | What to do | Examples |
|---|---|---|---|
| High volume, low complexity | Repetitive and predictable | Automate now | Creating leads in the CRM, UTM naming, form alerts, standard metrics reports |
| High volume, high complexity | Repetitive but full of exceptions | Simplify first, then automate in parts | Lead routing and qualification, multichannel reporting on messy data |
| Low volume, low complexity | Simple but infrequent | Document it with a checklist | Onboarding a new tool, monthly budget close |
| Low volume, high complexity | Infrequent and sensitive | Keep it with people, with AI as support | Positioning strategy, crisis management, vendor negotiations |
The interesting one is the second quadrant. It’s where the most time is lost and where it’s most tempting to launch an AI agent that «solves everything». First, reduce the exceptions: set clear rules, unify fields and decide who owns each step. Then automate the usual path and leave a manual exit for the odd cases.
How to do it in an afternoon, step by step
- Take inventory. List everything your team does in a normal month. No filtering: include the small stuff and the things that have «always been done this way».
- Estimate volume. Times per month and minutes each time. You don’t need precision, orders of magnitude are enough.
- Score complexity with the three questions above (exceptions, systems, cost of error).
- Place each process in its quadrant and pick two or three from the first one. Those are your first candidates.
- Decide how you’ll measure the result before you build: hours saved, errors avoided or response time. Without a baseline you won’t know if it was worth it.
Where AI fits in each quadrant
In the high-volume, low-complexity quadrant you often don’t need AI at all: a classic rules-based automation is cheaper, more predictable and easier to debug. AI earns its place when there’s unstructured text or data, like classifying replies, summarising conversations or drafting something a person then reviews.
For complex, low-volume processes, AI works best as a copilot: it prepares context, suggests options and flags risks, but the decision stays human.
Common prioritisation mistakes
- Picking what looks impressive. An agent that posts to social on its own looks better than automating UTM tagging, but the second usually delivers more value.
- Ignoring maintenance. Every automation needs an owner, reviews and someone who knows what to do when it fails.
- Skipping data clean-up. If your CRM fields are a mess, the automation will inherit it.
- Never revisiting the matrix. Processes change quadrant as volume grows or they get simpler. Review it every quarter.
FAQ
Which marketing process should I automate first?
The one that repeats a lot and has few exceptions: creating leads in the CRM, campaign naming, internal alerts or standard reports. Quick wins with little risk.
Do I need AI to automate?
Not always. If the rules are fixed, a traditional automation is cheaper and more reliable. AI pays off when the information is unstructured, like free text or conversations.
How often should I review the classification?
Once a quarter is usually enough, or whenever volume, tools or the team change noticeably.
If you classify your team’s processes differently, tell me in the comments or get in touch. I’d like to hear how you do it.
