Marketing automation has a reputation problem. A decade of badly configured drip sequences taught a generation of buyers to ignore anything that arrives on a schedule, and taught a generation of marketers that automation means sending more email to more people more often.

That is not what it is for, and the gap between what automation can do and what most teams use it for is the largest untapped margin we see in client accounts. Here is how we think about it.

The problem automation actually solves

Every business has a set of moments where a customer’s behaviour tells you something specific and useful, and where a timely, relevant response would materially improve the outcome. Someone starts a trial and never logs in. Someone views pricing three times in a week. Someone’s usage drops for a month. A customer hits the limit of their plan.

In a small business, a human notices these and reacts. At any scale beyond that, nobody notices, and the moment passes. Automation is the mechanism for noticing at scale. Everything else — the templates, the visual builders, the sequences — is plumbing.

Framed that way, the question is not “should we do marketing automation”. It is “which moments in our customer’s life are we currently failing to respond to”, and that question has an answer you can write on one page.

Start with the moments, not the tool

The most common implementation failure is buying the platform first. A team spends four months on migration, integrations and template design, launches a welcome series, and then stalls — because nobody ever wrote down what the automation was supposed to do.

We run it in the other order. Before any tooling conversation, we map the customer journey and mark every point where the customer takes an action that changes what they need. Then we rank those moments by how much revenue sits behind them. Most businesses find that three or four moments account for the majority of the opportunity, and that they can be built in a fortnight in whatever tool they already own.

Automation is not a channel. It is the difference between a company that notices what its customers are doing and one that finds out at renewal.

— Daniel Okafor, Lifecycle Lead at Marrky

The five flows that earn their keep

  • Onboarding by behaviour, not by day. A sequence that sends day one, day three and day seven regardless of what the customer did is a newsletter with extra steps. A sequence that branches on whether they completed setup is a product.
  • Abandonment, defined broadly. Cart abandonment is the famous one, but the same logic applies to an unfinished application, a viewed-then-left pricing page, or a demo booked and not attended.
  • Usage-triggered expansion. The best time to talk about a bigger plan is the week somebody starts pushing against the limits of the current one. This is almost always the highest-return flow we build, and almost always the one nobody has.
  • Dormancy recovery. Cheaper than acquisition by an order of magnitude, and the trigger is trivially available in any product analytics tool.
  • Post-purchase, treated seriously. The weeks after a purchase determine the review, the referral and the renewal, and they are usually left to a receipt.

What good looks like in practice

Two constraints keep automation from degrading into spam. The first is a global frequency cap that sits above every individual flow — no customer receives more than a set number of automated messages in a week, regardless of how many triggers they hit. Without it, your most engaged customers get the most email, which is precisely backwards.

The second is an exit condition on every flow. If the customer does the thing the flow was designed to prompt, the flow stops immediately. This sounds obvious and is violated constantly; the classic failure is the abandonment sequence that keeps arriving after the purchase.

Measuring it honestly

Automation reporting flatters itself more than any other channel, because the people it reaches were already engaged. Open rates on a triggered flow will always look extraordinary next to a broadcast campaign, and that comparison means nothing.

The measurement that means something is a holdout: withhold the flow from a random ten percent of eligible customers and compare outcomes. It is easy to set up, it costs almost nothing, and it is the only way to distinguish a flow that causes revenue from one that merely accompanies it. We have retired flows on the basis of a holdout that looked like our best performers on every vanity metric.

The maintenance nobody budgets for

An automation programme is not a project with an end date. Products change, offers change, and a flow written eighteen months ago is quietly telling new customers about a feature that no longer exists. We schedule a quarterly review of every live flow: read the messages end to end as a customer would, check the triggers still fire, and delete anything that no longer earns its place.

Teams that skip this end up with forty flows, no idea which ones are running, and an inbox experience they would be embarrassed to receive.

Where to begin

Pick the single moment in your customer journey where you know an opportunity is being missed — most teams can name it without looking at data. Build one flow, with a holdout, and measure it for a quarter. That one build will teach you more about your data quality, your customer’s actual behaviour and your team’s capacity than any platform evaluation.

Then do the next one. A programme of six well-maintained flows outperforms a library of sixty every time.

Choosing a platform, briefly

Once the moments are mapped, the tooling question becomes much smaller than vendors would like it to be. Three things matter and the rest is preference.

Can it see your product data? A platform that only knows about email opens can only build email-open logic. If your most valuable triggers live in the product — usage, limits, feature adoption — then the integration between product and platform is the entire decision, and everything else is interface taste.

Can a marketer change a flow without an engineer? The programmes that keep improving are the ones where the person who notices a problem can fix it that afternoon. Every hand-off adds a week and, more corrosively, adds a reason not to bother.

Can you get your data back out? Automation platforms accumulate history that becomes valuable exactly when you want to leave. Check the export before you sign, not after.

Notably absent from that list: the number of channels supported, the template gallery, and the presence of artificial intelligence in the marketing copy. We have yet to see a programme succeed or fail on any of them.

The organisational part

The reason automation stalls is rarely technical. It is that nobody owns it. Email belongs to marketing, product triggers belong to engineering, and lifecycle messaging sits in the gap where neither team’s roadmap reaches. The flows that exist were built by whoever cared most at the time, and when that person changes role the flows keep running unattended.

The fix is unglamorous: name one owner, give them a quarterly review slot, and put the holdout results in front of the same leadership meeting that sees paid media performance. Automation gets treated as infrastructure precisely because it is invisible when it works. Making the results visible is what keeps it funded.

A reasonable first ninety days

Weeks one and two: map the moments, rank them by revenue, pick one. Weeks three and four: build it, with an exit condition and a ten percent holdout. Weeks five through twelve: leave it alone and let it gather data, while you map the second flow. At the end of the quarter you will have one measured, working flow and a queue of ranked candidates.

That is slower than any implementation plan a vendor will show you, and it is the pace at which programmes survive their second year.

One last caution. Resist the temptation to launch all five flows in the first month because the tool makes it easy. Flows interact — a customer can qualify for three triggers in the same week — and the only way to understand those interactions is to introduce them one at a time. Teams that launch everything at once cannot tell which flow caused the lift, cannot tell which one caused the unsubscribes, and end up switching the whole programme off when the first complaint arrives.