The expensive automation mistakes are invisible from inside the platform, because the platform reports what it sent rather than how it read.
A sequence with a 99% delivery rate can still be sending a purchase prompt to somebody who bought yesterday, addressing a customer as {{first_name}}, or continuing a five-email onboarding flow to a person who cancelled on day two. The dashboard shows green.
Nine failures, ordered roughly by how much damage they do per instance.
1. Selling to someone who already bought
The most damaging and the most common.
A cart recovery flow that keeps running after the purchase completed. A "still thinking about it?" email to someone who has had the product for a week. A discount offered to a customer who paid full price two days ago — which converts a satisfied buyer into someone who feels overcharged, and often generates a refund request that would never have existed.
The cause is almost always the same: exit conditions were set on the wrong event, or on an event the platform receives late. A flow that exits on "order completed" while the payment gateway confirms asynchronously will send before the exit fires.
The fix: exit on the earliest reliable signal, not the tidiest one. And test it by buying something yourself, mid-sequence, and watching what arrives.
2. Sequences that ignore what the person did next
Automation that does not listen is worse than no automation, because it demonstrates that nothing is being noticed.
The pattern: somebody replies to an automated email with a question, and the next scheduled message arrives on time, ignoring the reply entirely.
Two things to configure on every sequence:
- Exit on reply, where the platform supports it
- Exit on the goal being met — booked, purchased, subscribed, whatever the sequence is for
A reply is the most valuable event in a sequence and the one most often discarded.
3. Broken personalisation
Hi {{first_name}}, is the visible version. The invisible versions are worse.
Hi ,— the token resolved to empty and the greeting still renderedHi there,as a fallback, applied to a customer whose name you have- A company name from a form field somebody typed sarcastically
- A location from an IP lookup that puts a customer in the wrong country
Two rules cover most of it. Set a fallback for every token, and check that the fallback reads naturally in the whole sentence rather than in isolation. "Hi there" is fine. "Thanks for your order, there" is not.
And use less personalisation, not better personalisation. A message that works without any tokens cannot break.
4. Nobody has entered their own automations
A flow that has not been tested from the inside is a flow you are guessing about.
Do this quarterly, with a real address: subscribe, abandon a cart, trigger the win-back condition, reply to a message, unsubscribe, resubscribe. Twenty minutes.
It catches things no dashboard reports: timing that feels wrong, two flows firing on the same day, a broken link, an unsubscribe that errors, a message whose tone is fine in isolation and cold in sequence.
5. Sending to people who stopped engaging
Every flow after the welcome sequence targets contacts defined by not engaging, and that is where the reputation cost sits.
Re-engagement flows are the clearest case. They deliberately mail people who have not opened for months — a segment containing dead addresses, people who forgot subscribing, and recycled spam traps, which are former real addresses that providers repurposed precisely because nothing legitimate should still be mailing them. What traps do.
The consequences reach beyond the flow. Complaints and bounces degrade the domain's reputation, and domain reputation is shared with your order confirmations and password resets. A re-engagement campaign can make your receipts stop arriving. Why streams should be separated.
Run re-engagement small, infrequently, and with a hard exit: one or two messages, then suppress. A contact who ignores a re-engagement attempt has answered.
6. Frequency nobody is counting
Each automation knows what it sent. Nothing knows what the contact received.
The compound failure: a person triggers the welcome sequence, browses a product, abandons a cart and hits a segment refresh in the same week — and receives eleven emails from a business they encountered on Monday.
Set a global frequency cap if your platform supports one. If it does not, keep a manual map of which flows can overlap, and add a minimum gap between sends to the same contact.
The number that matters is per-contact, not per-flow, and it is the number almost no report shows by default.
7. Automations that broke silently
A stopped automation looks exactly like a segment nobody entered. No error, no alert, zero sends — which is also what a quiet week looks like.
How they break:
- An integration's API changed and the trigger stopped receiving events
- A field was renamed and the condition no longer matches
- An upstream tag was retired, making a condition unreachable
- A platform migration carried the sequence but not the trigger
Keep a register. One page listing every automation, its trigger, its owner, and when it was last confirmed working. Sites accumulate flows built by people who left, and nobody turns them off because nobody knows what they do.
8. Automating the apology
Anything a complaint could become should not be automated.
Refunds, cancellations, billing disputes, service failures. An automated response to a person who is already annoyed converts a recoverable situation into a public one, and it is the single fastest way to turn a customer into a reviewer.
The same applies to the first message to anyone who matters commercially. A prospect who can tell it was generated has learned something about how you will treat them after they buy.
9. Measuring the automation by what it is credited with
Attributed revenue is not caused revenue. A cart recovery email sent to someone who was returning anyway gets credited with the sale, and a longer attribution window credits it with more.
Three better questions:
- What is the incremental effect? Hold back 5% and compare. Almost nobody does this, and it is the highest-value unused measurement available
- What did it cost in complaints and unsubscribes, for that flow specifically rather than the account average?
- What did it cost in attention? A flow that earns a little and exhausts the audience before a launch has a cost that never appears in its own report
The pattern underneath
Eight of the nine are the same failure: the automation is not receiving information it needs, and nothing reports the gap.
Purchases that arrive late. Replies that are discarded. Fields that stopped resolving. Triggers that stopped firing. Sends the other flows do not know about.
Which suggests the useful maintenance habit is not reviewing performance but confirming inputs — quarterly, from the inside, with a real address. The wider case for automation restraint.
Frequently asked questions
Why is my automated email still selling to customers who bought?
The exit condition is set on an event the platform receives late — commonly an order confirmation that arrives asynchronously from the payment gateway. Exit on the earliest reliable signal instead, and test it by making a purchase yourself mid-sequence.
How often should I test my marketing automations?
Quarterly, from the inside, with a real address: subscribe, abandon a cart, reply, trigger the win-back, unsubscribe and resubscribe. It takes twenty minutes and catches failures no dashboard reports, because a broken automation looks identical to an empty segment.
Do re-engagement campaigns hurt deliverability?
They can, because they deliberately mail the least engaged segment — which contains dead addresses and recycled spam traps. Complaints and bounces from that send degrade the whole domain's reputation, including transactional mail. Keep them short, infrequent, and suppress non-responders after one or two attempts.
How many emails is too many from automation?
The number that matters is per contact, not per flow, and most reports do not show it. Someone can enter four flows in a week and receive eleven messages while every individual flow looks reasonable. Set a global frequency cap, or map which flows can overlap.
Should personalisation tokens have fallbacks?
Yes, and the fallback must read naturally in the full sentence rather than on its own. "Hi there" works; "Thanks for your order, there" does not. Using fewer tokens is more reliable than configuring them better.
What should never be automated?
Anything a complaint could become — refunds, cancellations, billing disputes, service failures — and the first message to anyone who matters commercially. An automated reply to an annoyed customer converts a recoverable situation into a public one.
Free: The 60-Minute Email Authentication Fix
A no-fluff checklist to set up SPF, DKIM & DMARC correctly and pass Gmail & Yahoo's sender requirements.
Muhammad Basim has worked in digital marketing since 2013, focused on email deliverability and AI-assisted content production. He is the author of The Email Deliverability Playbook and The Email Copywriting Playbook.
Related Articles
Does Send Time Optimisation Work?
Send time optimisation models are trained on when recipients opened messages, and open timestamps now include machine activity that has nothing to do with when anyone was reading. Privacy protection pre-fetches message content on its own schedule, regardless of whether the recipient opened it. Security gateways fetch content during pre-delivery scanning, which happens before the […]
Prompting for Marketers: The Part That Actually Matters
Output quality is set mostly by the material you supply, not the phrasing you use. That is the finding people take longest to accept, because it is less interesting than the alternative. A carefully worded request with no context produces a competent generic answer. A plainly worded request with your actual customer language, your positioning […]
What You Grant When You Connect a Tool
Your security perimeter includes every vendor you have ever connected, including the ones you stopped using and never disconnected. A connection is a standing grant. It does not expire because you stopped logging in, it does not lapse because the trial ended, and it does not disappear when you delete the app from your phone. […]