Automation Mistakes Founders Make and How to Spot Them
Automation mistakes founders make: nine patterns that turn reliability into risk, each with the symptom you will notice and the fix that belongs to it.
The automation mistakes founders make come down to nine patterns: automating a guess, automating the relationship, starting with outward messages, forgetting the exit, allowing silent re-entry, branching on missing data, building one long chain, measuring that it ran and never switching anything off. All nine are avoidable.
These patterns turn automation from reliability into risk. What they share is that none is a failure of tooling. They are failures of selection, design and attention, and all three are decided by you, before and after the software does its part. That is good news, because it means a diagnosis works. This post is arranged by symptom, so you can start from what you are seeing.
Automation mistakes founders make: the nine at a glance
| Mistake | What it costs | Better habit |
|---|---|---|
| Automating a process that has never worked by hand | Customers find its flaws for you | Do it by hand until you know what good looks like |
| Automating the relationship instead of the follow-through | Indifference at the moment attention is highest | Keep complaints, bad news and negotiation with a person |
| Starting with outward messages | Trust, in public | Begin with internal tasks, assignments and notifications |
| Designing the entry and forgetting the exit | Chasing someone who already said yes | Write the goal that removes people |
| Allowing silent re-entry | Everything arrives twice | Decide re-entry on purpose |
| Branching on data you do not have | Half your contacts take the wrong path | Check how often a field is filled in |
| Building one long workflow | Hard to read, change and trust | Several short workflows that hand off |
| Measuring that it ran, not that it worked | Overhead with a progress bar | Measure the outcome before and after |
| Never switching anything off | Stale promises sent on your behalf | Retire what has outlived its purpose |
If customers are pointing out flaws
When the first people to notice a problem are the customers, the likely cause is that you automated a process that had never worked by hand. In that case you are not automating a process, you are automating a guess, and the people on the receiving end are doing your testing.
The cure is order. Do the process by hand enough times to know what a good outcome looks like, notice the exceptions and note which messages drew replies. Only when you can describe it in a few sentences, with no judgement calls inside it, is it ready. The full filter is in how to decide what to automate first.
If people say it feels cold
Complaints that your messages feel cold or processed usually mean you automated the relationship instead of the follow-through. Complaints, bad news, negotiation and the first serious conversation belong to a person, and a template in those moments reads as indifference.
Look at the moment, not the wording. Better phrasing will not rescue a message that should never have been a rule. The boundary is set out in what should you never automate.
If an error reached the public
An error that customers saw often means the workflow started with an outward message. Internal tasks, assignments and notifications are cheap to get wrong, because only your team sees the error. Customer messages are not. The guidance is to earn trust in your triggers before you point them at people outside.
If your earliest workflows speak to customers, consider whether an internal version could carry the same handoff for now. Many of the most valuable workflows never send a customer anything at all.
If people keep hearing from you after they said yes
A reminder to book a call arriving after someone booked, or a "just checking in" after a purchase, tells you that you designed the entry and forgot the exit. Without a goal that removes people, the automation keeps chasing someone who already said yes.
The same goal does a second job: it keeps messages from arriving after the situation changed. Every workflow that sends messages over time should be stopped by the goal it exists to achieve, whatever step the person is on.
If people receive everything twice
Duplicates point to silent re-entry. The same person was enrolled twice, because two events fired or because they filled in a form again, and so received every message twice. Duplicate messages tell the recipient, very precisely, how much attention is being paid.
The fix is a deliberate rule. A person already inside should not be enrolled again, and re-entry should be allowed only where it is clearly intended, such as a workflow that should run each time a customer renews. A deliberate test of this is part of how to test a workflow before switching it on.
If half the contacts go the wrong way
When a fork seems to treat people oddly, check the field it reads. A condition on a half-empty field quietly sends half your contacts down the wrong path, and nothing flags it. The check is simple: look at how often the field is filled in before building anything that depends on it, and define both paths of every fork.
If you cannot explain what a workflow does
A workflow nobody can explain is usually a long one. Long chains are hard to read, hard to change and easy to break, because a mistake early on carries through everything after it. When a process has several phases, several short workflows that hand off to each other are easier to trust than one that does everything.
If every report is green but nothing improved
This is the mistake of measuring that it ran, not whether it worked. A perfectly executing workflow that does not move the outcome is overhead with a progress bar. The cure is to measure the outcome you cared about before switching on, and again a few weeks later. The routine for this is in how to spot a silently failing workflow.
If you have lost count of what is running
Workflows outlive their processes, their owners and their purpose, and keep talking to people long after anyone remembers why. If you cannot list what is running and what each rule is for, you have the ninth mistake. Retiring a workflow is maintenance, and the signs are laid out in when to switch off a workflow.
How to run the diagnosis
Take one workflow at a time and walk it through the nine in the order they appear in the table. For each, ask the question in a single sentence and answer it from the workflow itself, not from memory of how you meant it to behave.
- Did this process work by hand before it was a rule?
- Does it stand in for a conversation that belongs to a person?
- Does it speak to customers when an internal version would do?
- What ends it, and does reaching that goal remove the person?
- Can the same person enter twice, and was that intended?
- Does every fork read a field that is reliably filled in?
- Is it short enough to read in one sitting?
- Do I know whether it changed the outcome?
- Does it still deserve to run?
Where you cannot answer a question, you have found the work. Write the gap down, give it an owner and handle it before building anything new. A workflow that passes all nine is not guaranteed to be right, but it has avoided the failures that cost the most.
Which to fix first
If several apply at once, order the repairs by how far the harm travels. A problem that reaches customers comes before one that only wastes your own time, because a customer cannot un-read a message.
| Priority | Kind of problem | Why |
|---|---|---|
| First | Anything that speaks to customers wrongly | The harm is public and repeats on every run |
| Second | Anything that leaves people stuck in a workflow | It keeps producing wrong messages |
| Third | Anything that hides its own failures | You cannot judge the rest until you can see |
| Fourth | Clutter and unowned workflows | It costs review time rather than trust |
A symptom guide
| What you notice | Look first at |
|---|---|
| Customers report errors before you do | Whether the process worked by hand, and whether it started outward |
| Replies ask whether a human is there | Whether it automates the relationship |
| Messages arrive after a purchase or booking | The exit and the goal |
| The same message arrives twice | Re-entry rules |
| One branch is full and the other is empty | The field behind the condition |
| Nobody can say what it does | Length, and whether it has an owner |
| The dashboards look fine and nothing changed | The outcome measure |
What the nine have in common
Read the list again and notice that none depends on which software you use. A machine will happily do any of these things for you, quickly and without complaint. The protection lies in your decisions: what to build, how to build it and when to look at what it does.
Seen that way, a review is not an audit of the tool. It is a conversation with yourself about selection, design and attention, held at intervals. The questions are short, and the answers are in your own records.
Frequently asked questions
Which mistake is the most common?
The guide does not rank them. In practice, start by asking whether the process worked by hand, since an unproven process feeds several of the others.
Can a good tool prevent these?
A good tool can make some of them harder, such as by making re-entry an explicit choice. It cannot choose what to automate, how to design it or when to retire it.
I have already made several of them. What now?
Fix one at a time, starting with whichever reaches customers. Test each repair as you would a new workflow, and read what happens afterwards.
How do I avoid repeating them?
Use a checklist before building and before switching on, and review what the workflows actually did on a regular schedule.
The full method, with the checklists, is in The Automation Playbook, a free 29-page playbook.