How to Define Funnel Steps That Describe Real Decisions
How to define funnel steps as ordered states a person reaches, built around the customer's decision and not your org chart, so the numbers mean something.
To define funnel steps, write each one as a state a person reaches, put the states in the order they genuinely happen, and model the customer's decision from end to end rather than your company's departments. A funnel defined badly cannot be rescued by good instrumentation, so this is the part to get right first.
Most bad funnels are bad for one reason. They model the company's internal structure instead of the decision the customer is making. The fix is not technical. It is a single session of careful thinking, done before anything is measured.
How to define funnel steps as states, not pages
A step is a state the person reaches. It is not a page they saw and it is not a department they touched. Here are states written the way they should be written:
- Arrived.
- Understood what this is.
- Gave us their email.
- Completed setup.
- Used it once.
- Paid.
Read that list as a story about one person. Each line is something that has become true of them. None of the lines names a button, a template or a team.
The reason to insist on this is durability. Steps written as states survive changes to your website. Steps written as pages break the moment someone reorganises the navigation, renames a section or merges two screens. If your funnel only makes sense for the current layout, you will rebuild it every time the layout changes, and your history will break each time.
| Written as a page or team | Written as a state | What goes wrong with the first |
|---|---|---|
| Visited the pricing page | Understood what this is | Breaks when the page is renamed or moved |
| Reached the sales team | Gave us their email | Describes a hand-off, not what the person did |
| Saw the welcome screen | Completed setup | Counts a view, not the thing that mattered |
| Passed to the product team | Used it once | Names a department and says nothing of the person |
The right column survives a redesign. The left column does not, and it also hides what you actually want to know, which is whether the person got somewhere.
The steps must be genuinely sequential
The second test is about order. Each step must follow the one before it, so that nobody can reach step four without having passed through step two.
If a person can reach step four without step two, then step two is not part of this funnel. It is a side path. Including it will make your drop-off numbers meaningless, because some people never needed to pass through it and will look like they were lost there. A quick way to check is to take each pair of neighbouring steps and ask a plain question: is there any way to be at the later one without having been at the earlier one? If there is, one of them does not belong.
When you find a side path, do not delete it from your thinking. Give it its own short funnel later if it matters. What you must not do is let it sit inside the main one and distort every percentage that comes after it.
Model the customer's decision, not your org chart
The most common structural error looks reasonable on a slide. Marketing has a funnel that ends at the lead. Sales has a funnel that starts at the lead. Product has a funnel that starts at signup. Each team can see and improve its own part.
The result is that nobody owns the whole path, and the losses at the seams are invisible to everyone. The customer does not experience three funnels. They walk one path. So draw the one they walk, end to end, and accept that it crosses internal boundaries.
| Approach | Who can see the loss at the seam? | Typical result |
|---|---|---|
| Three team funnels, handed off | Nobody | Every team looks fine while customers disappear between them |
| One funnel modelling the customer | Whoever looks at the whole path | The seams show up as steps with a number and an owner |
It is worth saying plainly that the steps at the boundaries are frequently the worst performing ones, precisely because they have never had an owner. That is not a coincidence and it is not bad luck. It is what you should expect from any step that belongs to nobody. Drawing the whole path is how you discover them.
Reading the six states as one decision
Take the six states above and ask what each one tells you about the person. Arrived says they found you. Understood what this is says the message landed. Gave us their email says they were willing to be known. Completed setup says they did the work of starting. Used it once says the product delivered something. Paid says they were convinced.
Notice that this reads as the customer's decision, one commitment after another, and that no team is mentioned anywhere. That is the quality to aim for. If you can read your own list aloud as a story about one person, it is probably modelling the right thing. If you stumble because a line names a department or a screen, rewrite that line before you go further.
A short test for each step
Before you accept a step into the funnel, put it through these questions. A step that fails any of them needs rewriting.
- Is it a state the person reaches, rather than a page or a department?
- Does it follow the previous step, with no way around it?
- Does it describe the customer's decision, not how we are organised?
- Would it still make sense if the website were redesigned tomorrow?
This is a few minutes per step. It is cheap compared with discovering, after a month of data, that half your steps were pages that moved.
Keep the number of steps small
Five or six steps are enough. More steps feel more precise, and in practice they are worse: each extra step adds noise, splits your volume into groups too small to read and adds maintenance. Five or six well chosen steps describe almost any purchase decision at the level of detail a founder can act on.
If one step is genuinely complex inside, you can break that one down later, once you know it is the problem. That is a better order of work than starting with twelve steps and hoping the structure reveals itself. For how to choose the single funnel worth defining first, see how to choose your first funnel.
Write down the rules alongside the steps
A funnel is more than a list of steps. It also needs a few rules, and they belong in the same document as the steps so that nobody has to remember them.
- A time window. How long a person has to complete the next step before they are counted as having dropped. Set it from how long the move actually takes for those who complete it.
- A counting rule. Each step counts only people who completed every step before it. The reasoning is in funnel cohort analysis.
- The outcome. What the funnel is ultimately measuring, stated in one sentence.
- The first segments. Which groups you will look at first. Traffic source is usually one of them, as described in how to segment a funnel.
Say the window out loud whenever you quote a conversion rate. Two people quoting different windows will disagree forever without discovering why.
Errors that show up in a first draft
- A step named after a screen. Rewrite it as what became true of the person.
- A step that can be skipped. Move it to a separate side path.
- A step owned by one team only. Check whether the funnel has quietly become a team funnel and not a customer one.
- Too many steps. Merge neighbours until there are five or six.
- No outcome stated. If you cannot say what the funnel measures, you cannot say whether it is working.
How the definition feeds the rest of the work
Definition comes before instrumentation for a reason. Once the steps are written as states, you can decide which of them are pages and which are actions, which is the question covered in how to track funnel events. If you started with the tools, you would be tempted to define the funnel as whatever the tools happen to record, and that is how a funnel ends up describing the software and not the person.
A well defined funnel also changes the conversation. Instead of asking whose numbers are right, people ask where in one shared sequence the loss is. That is the point at which the work becomes productive, and it is also the point at which finding where a funnel loses people becomes straightforward.
Frequently asked questions
Should a funnel step ever be a page?
Only when the state you care about really is reaching that page. Even then, think of it as the state the page represents. If the page is reorganised, the state should still hold.
What if my customers take different routes?
Pick the main route to the outcome that matters most and define that. Routes that can skip steps are side paths, and mixing them in makes every percentage meaningless.
Can marketing and sales share one funnel?
They should, because the customer walks one path. The funnel crosses internal boundaries, and the steps at those boundaries need an owner.
How many steps is too many?
More than five or six is usually too many at the level a founder can act on. Break one step down later, once it is the problem.
The full method, with the checklists, is in The Funnel Playbook, a free 30-page playbook.