The word "funnel" has been so thoroughly adopted into product vocabulary that we've stopped questioning whether it's an accurate description of what actually happens when a new user tries your product. It isn't.
A funnel implies sequential stages, gravity pulling things downward, and irreversible movement. Liquid enters the top and exits through a narrow bottom. Users don't work that way. They skip steps, they revisit steps, they complete step 4 before step 2, they bounce and come back three days later and pick up somewhere in the middle. Modeling this as a funnel produces systematically misleading data and systematically wrong interventions.
What the funnel model gets wrong
The core problem with funnel thinking is the assumption of sequentiality. A standard activation funnel report shows you: 100% started step 1, 72% reached step 2, 48% reached step 3, 31% reached step 4. This makes it look like 28% of users "dropped off between step 1 and step 2."
But what actually happened is more complicated. Some of those 28% came back the next day and completed step 2. Some completed step 3 without completing step 2. Some completed step 2 but then went to step 4 before doing step 3. The funnel collapses all of this into a single linear sequence and shows you a misleading drop-off number.
When you optimize for a metric derived from a wrong model, the optimization is wrong even if the metric improves. You might add a nudge that drives step 2 completion, not realizing that 40% of the apparent "step 2 drop-off" was actually users who come back later and self-complete. You've added friction to the experience of the 40% who didn't need help, to recover the 60% who did.
What the journey actually looks like
When we look at session-level event data for PLG products, we typically see several journey patterns occurring simultaneously:
Linear completers: these users move through activation steps roughly in order, usually in one or two sessions. They're the users your funnel model is built to serve, and they're often 20-35% of all signups depending on product complexity.
Returners: users who start, hit a step that requires data or a decision they don't have available, leave, and come back. Collaboration tools see this frequently because "invite a teammate" requires the user to have a teammate in mind. The user who leaves at step 3 isn't lost; they're thinking about it. Your funnel says "dropped off at step 3." The truth is they converted three days later.
Explorers: users who skip the prescribed onboarding sequence entirely and navigate directly to the feature they're interested in. These users often have prior experience with similar tools and don't need to be walked through basics. If your activation model requires sequential step completion, explorers will never register as "activated" even if they're getting genuine value and have no intention of churning.
Partial adopters: users who complete the steps relevant to their job-to-be-done and ignore the rest. A solo user on a team tool might complete setup and core feature use but never complete the "invite teammates" step. If that step is required for activation, partial adopters are permanently misclassified.
A better mental model: activation as a set of conditions
We find it more useful to think of activation as a set of conditions that must all be true, rather than as a sequence that must be traversed in order. The conditions might be:
- User has completed setup (necessary prerequisite)
- User has used the core feature at least once
- User has had at least two separate sessions in the product
- User has seen a result from using the core feature
All four conditions must be met, but the order doesn't necessarily matter as long as setup comes first. A user who has all four conditions met within 14 days of signup is activated, regardless of the path they took.
This model has several advantages over a linear funnel. It correctly classifies returners (who met the conditions over multiple sessions). It correctly classifies explorers (who may have reached conditions 3 and 4 before technically completing condition 2 in the canonical order). It gives you a binary activated/not-activated signal that you can use to build behavioral cohorts and churn prediction models.
How this changes your optimization priorities
Once you stop thinking in terms of sequential stages, the questions you ask about your activation data change.
Instead of "why are users dropping off between step 2 and step 3," you ask "what percentage of users who completed condition 1 and condition 2 go on to complete conditions 3 and 4 within 14 days?" The cohort comparison becomes: users who met conditions 1+2 within their first session vs. users who took longer. You're isolating where the activation timeline is slipping, not where a prescribed sequence is breaking.
The nudge interventions you design are also different. Rather than a nudge that fires when a user hasn't reached "step 3," you design a nudge that fires when a user has met conditions 1 and 2 but hasn't met condition 3 within 5 days. That's a more precise target with a more useful message.
The nuance: some sequences do matter
We're not saying that all ordering in activation is irrelevant. Some steps are genuine prerequisites. You can't use a product's core feature without completing an initial setup. And for products where the onboarding sequence introduces important concepts, out-of-order traversal can lead to confusion and premature drop-off.
The point isn't that sequences are bad. It's that treating your activation model as a strictly linear funnel by default causes you to misclassify non-linear but successful journeys as drop-offs. The appropriate response is to audit your activation model and ask: which of these ordering requirements are genuine prerequisites, and which are just the sequence we happened to design?
For prerequisites, keep the sequence. For arbitrary ordering, relax it. Let users move through activation in a way that matches their natural behavior, and measure activation against conditions met rather than steps traversed.
Practical implications for Onboardvue customers
In Onboardvue, activation milestones support two ordering modes: "strict sequence required" and "conditions-based (any order)." For most products, we recommend using strict sequence only for the steps that are genuine technical prerequisites, and conditions-based for everything else.
When you switch from strict-sequence to conditions-based, the most common result is that your activation rate goes up, not because users are performing better, but because you were previously classifying returning users and explorers as inactive. You were solving the wrong problem.
That recalibration is uncomfortable for a week and then genuinely useful for everything that follows. You're measuring what you meant to measure, not a proxy shaped by the vocabulary of a model that was never quite right for activation in the first place.