Automation funnel metrics: which number tells you what

The first number an automation panel shows you is usually “this flow ran 1,284 times”. On its own it says nothing: 1,284 people entered, but how many reached the end, how many disappeared at the first step, how many were handed to a person? Funnel metrics answer exactly those questions.

The short answer: track five numbers together — entries, step-by-step drop-off, button click rate, handover rate and the count of messages skipped as outside the window. None of them means much alone; together they nearly always point at where the flow is jammed.

What a funnel is, and why step by step

An automation flow is a sequence of steps. At each one you lose some people: they don't answer, they close the chat, they ask about something else. That is normal. The problem is not knowing where the loss concentrates.

An example flow and a typical funnel:

One thing stands out: more than half drop at the menu step. Optimising the rest of the flow is pointless; the problem is there.

The five core metrics

1. Entries

How many people entered the flow. This measures reach, not success. If it is low the problem is the trigger, not the flow: the keyword may be too narrow, the campaign may have reached few people, or the flow may be wired to the wrong channel.

Also separate unique people from runs. If the same person entered three times, runs is 3 and people is 1. With a re-entry rule of “every time”, the two diverge fast.

2. Step-by-step drop-off

How many people moved from each step to the next. This is the most valuable metric because it points straight at a fix.

How to read it:

• Heavy drop at the first message: the message is too long, too advertising-like, or unexpected. People close it without reading.
• Heavy drop at the menu: the options are confusing, too many, or unrelated to what the person asked.
• Heavy drop at a wait-for-reply step: the question asks for too much effort. Try two buttons instead of “type your order number”.
• Drop at the final step: usually normal; the person got what they needed and ended the conversation.

3. Button click rate

How many people tapped a button on a message that offered them. You may want to count typing the answer as a “click” too — well-built systems treat typing a button's label as a tap.

A low click rate nearly always means one of two things: the button labels are vague (“Continue”, “Info”), or the options don't match the person's question. Turning labels into actions usually works: “Track my order” instead of “Continue”.

4. Handover rate

How many conversations left automation for an agent. This one is easy to misread: high is not bad.

• Very low handover (say under 5%): the flow may be solving everything — or people may be leaving without finding the answer. Read it against drop-off.
• Very high handover (say over 70%): the flow isn't working, it's just a waiting room. Look at which step precedes the handover; that step is missing something.
• A healthy range: depends on the business. High handover is normal on a support-heavy account and should be low on a catalogue campaign flow.

5. Messages skipped outside the window

The number most teams never look at and the one that causes the most trouble. On Instagram and WhatsApp you cannot send free text more than 24 hours after the user's last message. If your flow has a long wait, the next message quietly does not go.

If this number is above zero, revisit the flow's timing. The rule is covered in the WhatsApp 24-hour rule guide.

Reading the numbers together

Individually none of them is enough. Together they produce a diagnosis:

How often to look

Daily for the first week after publishing, then weekly. The only point of looking daily is catching errors: a broken connection, a message not going out, the wrong flow firing. Don't make optimisation decisions on daily data; at small volumes the daily noise is larger than the real change.

Run a monthly “dead flow” sweep too: switch off flows that get no entries. Old flows left on create keyword collisions with new ones and cause silent failures.

What not to measure

• Total messages sent. It looks like a good thing when it rises; usually it just means the flow got longer.
• Flow completion rate alone. A short flow always has a high completion rate. That does not make it better.
• The share of conversations automation “resolved”. Resolution usually means “not handed over”, which includes everyone who quietly left.

Frequently asked questions

How many conversations before I can trust these numbers?

Roughly, once you reach about 100 people per step the ratios begin to settle. Below that, a single day's noise moves the percentages noticeably. On low-volume accounts use monthly rather than weekly windows, and don't make decisions on small percentage differences.

What should the step drop-off be?

There is no universal target; it depends on the flow's length and purpose. The practical approach is to compare each step with its own history: if a menu that converted at 52% last month is at 38% now, something changed. Direction matters more than the absolute number.

If handover is low, is the flow successful?

You cannot tell from that alone. If handover is low and drop-off is high, the flow is not solving problems — people are leaving without an answer. Read the two together: in a healthy flow, low handover comes with low drop-off.

Can I run A/B tests?

You can, but be careful at low volumes. Rather than running two versions simultaneously, most teams find two consecutive weeks more practical and less error-prone. Change one thing per test: the wording or the button labels, not both.

How do I get out-of-window messages to zero?

Three steps: add up every wait in the flow, mark the points where the total passes 23 hours from the person's last message, and either bring those messages forward or convert them to templates. How template approval works on WhatsApp is covered in the template approval guide.

Where to start

Pick one flow this week and build the six-row funnel table above by hand. You will probably find one step with a bigger drop than you expected. Fix that, wait two weeks, measure again.

If you also want to measure response time, how to measure DM response time covers the method. For sourced market data, see Turkey messaging statistics.