What to Automate First (And What to Leave Alone)
Automation should remove repetition, not judgement. A practical order of operations for small marketing teams.
The short answer
Automate the things that are repetitive, rule-based and already working. Never automate the first version of anything, and never automate judgement, relationships or the founder's voice.
Automation is sold as a way to do more. In practice it is a way to stop redoing the same thing, which is a smaller claim and a more useful one.
The failure mode is automating too early: building an elaborate workflow for a process that was never validated, then maintaining it for two years.
The rule
Automate what is repetitive, rule-based, and already working.
All three conditions matter:
- Repetitive — you do it often enough that the setup pays back
- Rule-based — the decision can be written as a rule without losing anything
- Already working — you have done it manually and it produced results
The third is the one people skip, and it is the expensive one. Automating an unvalidated process just makes a mistake happen faster and more consistently.
Automate these first
Lifecycle email flows. The clearest case: repetitive, rule-based, and — if you built them after doing the manual version — validated. Welcome, abandoned checkout, post-purchase, win-back. These are Architect the System work and they compound.
Review and UGC requests. Asking every customer for a review at the right moment is pure repetition. Automate the ask; keep the response human.
Reporting. Assembling the same numbers weekly is exactly what software is for. Automate the assembly, never the interpretation.
Back-in-stock and price-drop alerts. High intent, entirely rule-based, and they capture demand you would otherwise lose silently.
Content repurposing mechanics. Not the writing — the mechanical parts. Clipping, resizing, scheduling, cross-posting. The judgement about what is worth repurposing stays human.
Internal handoffs. When a form is submitted, someone should be notified and a task created. Small, unglamorous, and it stops things falling through cracks.
Automate these later
Segmentation logic, once you have enough data for the segments to be real rather than theoretical.
Personalisation, once you know which variables actually change behaviour. Personalising on a variable that does not matter adds complexity for nothing.
Ad creative rotation, once you have enough winning creative that rotation is a real decision.
Do not automate these
Judgement. Which campaign to run, what the strategy is, whether something is on-brand. Tools can inform this; they cannot make it.
Relationships. Automated DMs, connection-request sequences, comment bots. These are visible, and being caught faking a relationship costs more than the relationship was worth. Trust is the asset — automating the appearance of it is the fastest way to spend it.
The founder’s voice. You can use tools to draft and edit. But content that has never passed through the founder’s actual thinking reads like it, and the entire advantage of founder-led content is that it could not have come from anyone else.
Customer service, past the first layer. Deflecting simple queries is fine. Automating the hard ones is how you turn an annoyed customer into a public one.
The first version of anything. Do it manually until you know it works. Then automate it.
The maintenance cost people forget
Every automation is a small ongoing liability. It breaks when a platform changes, it needs updating when your process changes, and — worst — it can fail silently while appearing to work.
Two habits worth adopting:
- Document what each automation does and why, so the person who inherits it can maintain it.
- Audit them quarterly. Turn off anything nobody can justify. The number of dormant, half-broken workflows in a typical account is genuinely surprising.
Human-first, then extended
The framing we use internally: strategy, judgement and relationships are people work and always will be. The operational layer underneath — the repetitive, rule-based, validated parts — should run on tools.
That is what lets a small team work like a larger one. It is not a replacement for the human parts; it is what buys the time to do them properly.
Where this sits in the framework
Automation is Nurture & Scale — stage five, deliberately near the end.
You cannot automate your way to a strategy, and automating an unvalidated process just scales a mistake. Get it working manually first. Then make it repeat without heroics.
Related reading
Black Friday Planning Starts in September
The brands that have a good Q4 do the work in the quarter before it. A practical timeline and the traps to avoid.
One Asset, Five Channels: A Repurposing System That Holds
Most brands under-produce because they treat every channel as a separate job. It isn't — it's one idea, expressed five ways.
Build a UGC System, Not a UGC Campaign
Customer content is the highest-trust creative you can get. The problem is never quality — it's that most brands ask once and then stop.