Funnels & automation
How to test an autofunnel and where it usually breaks

You test an autofunnel by walking through it, not by reviewing the diagram: go through it from someone else’s account as a customer, twice — with the script and against it. The second route finds most of the breaks, because that’s where the funnel meets actual people.
A flow chart in the editor always looks sensible. It breaks on what the chart doesn’t contain: someone asking «how much is it?» in the middle of a pre-written sequence.
How do you walk your own funnel properly?
From another account, on a phone, not from your work computer. And three times, behaving differently:
- 1
The perfect route
Everything on script. Confirms the plumbing is connected.
- 2
A wrong entry
The keyword misspelled, in a different case, inside a sentence. This is how real people behave.
- 3
Your own question
Instead of the expected reply, write «is it in stock?». Half of customers do this.
- 4
Silence
Don't reply for a day. See what the funnel does with silence — nothing, or a nudge.
The third scenario fails most funnels. The person asks their question, the bot repeats its own, and the conversation ends.
Five places funnels break most often
1. The keyword doesn’t fire. The customer wrote «price» instead of «PRICE», with a full stop, or inside a sentence. Fix: match the word in any form rather than exactly.
2. The next message never arrives. Usually because the transition condition demands an exact answer the customer didn’t give.
3. No fallback. If a reply isn’t recognised, the funnel should hand over to a human. Repeating the same question is a guaranteed exit.
4. An ending with no action. The funnel built interest and finished on «thanks». The last message must always contain a concrete step.
5. Duplication. Someone enters the funnel a second time and receives the whole chain again. It looks like a malfunction, and it is one.
What do you measure to find the real cause?
One number per step: how many people got through it. Not conversion to orders — the transition between adjacent steps.
That table answers the main question instantly. A sharp drop between two steps means a problem in one specific message, and you can see it in a second. An even decline across the whole funnel means it’s simply too long.
Pre-launch funnel checklist
- Walked from another account — intended route.
- Walked with a misspelled keyword.
- Walked with an off-script question mid-chain.
- Behaviour checked when the customer stays silent for a day.
- Repeat entry by the same person checked.
- Final message contains a concrete action.
- A «not recognised → hand to a human» fallback exists.
- Every link opens correctly on a phone.
Frequently asked questions
How do you test a funnel before launch?
Walk through it yourself from another account, as an ordinary customer, and take two routes: the intended one and the wrong one. Most breakages live on the wrong route — when someone types the wrong keyword or answers off-script.
Where do autofunnels break most often?
At the transition between steps: the person replied and the next message either didn't arrive or was the wrong one. The second most common spot is an ending with no action — the funnel builds interest and then offers nothing.
How many people need to go through before you can draw conclusions?
At least 30–50 entries. On a sample of ten every number is noise, and decisions based on it damage funnels more often than they improve them.
What if people answer in ways the script doesn't expect?
Don't add a branch for every variant — build a fallback: if the reply isn't recognised, the funnel hands the conversation to a human instead of repeating the same question.
How do you tell whether the problem is the message or the traffic?
Look at which step loses the most. Few people entering means a traffic question; plenty entering and vanishing at step two means a message question.



