Product · 10 min read · 2026-10-28
Activation Depth: Core-Action Count Beats a Binary Flag
Most teams instrument activation as a single switch: the user either crossed the bar or they did not. That switch throws away the most useful thing your onboarding data knows — that conversion scales with how many core actions a user completes, not with whether they tripped one threshold. The sharper instrument is activation depth: count the value-delivering actions each signup completes in their first days, then look at how trial-to-paid conversion moves as that count climbs. The relationship is almost always a dose-response curve, and the shape of that curve tells you where to put your activation bar and which onboarding nudge is worth building next.
The stakes are concrete. Spectacle documents a B2B SaaS case where users who completed three specific actions converted trial-to-paid at roughly 62%, against about 3% for users who completed none — a twenty-fold gap hiding inside the same signup population. A binary activated flag set anywhere in that range would blur the gradient into a single average and tell you almost nothing about where the leverage is. This post is the method for measuring the gradient instead of averaging it away.
Binary activation hides the gradient
A yes/no activation flag is a lossy summary of a richer signal. When you label every user above a threshold "activated," two users who behaved completely differently — one who did the bare minimum to clear the bar and one who completed every core workflow twice — land in the same bucket and get counted as the same outcome. Your activation rate then reports the share of users who cleared an arbitrary line, not the distribution of how deeply users engaged, and the distribution is where the conversion signal lives.
The binary framing also quietly fixes a decision you have never actually examined: the threshold itself. If you only ever look at conversion above versus below one cut point, you cannot tell whether a higher or lower bar would separate converters from non-converters more cleanly. This is the same failure mode as defining the aha moment by copying a competitor's number: the threshold is asserted, not derived from your own data. Activation depth replaces the asserted line with a curve you can read.
Instrument a core-action count, not a flag
Start from a shortlist of actions that plausibly deliver core value — the things that map to the job your product does, like connecting a data source, inviting a teammate, publishing a first report, or completing the setup flow. Keep the list mechanistic and short, five to fifteen events, and avoid vanity actions such as "logged in three times" that measure presence without naming value. This is why a clean event taxonomy is a prerequisite: you can only count core actions reliably if each one is a well-defined event your tracking already captures.
Then, for every signup, count how many distinct core actions they completed within a fixed early window — commonly the first 7 or 14 days. Store that count as a first-class property on the user. Keep activation depth separate from adoption, which is habitual repeated use; counting the same action five times is an adoption signal, whereas counting five different core actions once each is a depth signal. Holding the activation-versus-adoption boundary clear keeps the count interpretable as breadth of early value, which is what predicts conversion.
Chart the dose-response curve
Bucket your historical signups by core-action count — 0, 1, 2, 3, and so on — and compute the trial-to-paid conversion rate for each bucket against a fixed downstream outcome. Plot conversion on the vertical axis against action count on the horizontal, and you have the dose-response curve. The typical shape rises steeply through the first few actions and then flattens: the first two or three core actions buy most of the conversion lift, and additional actions add progressively less. The steep region is where onboarding effort pays off; the flat region is where pushing users to do more stops moving the number.
Draw both cohorts honestly. Build the buckets from signups old enough that the conversion outcome is actually known, and include the accounts that churned — analysing only currently active users is the survivorship bias that makes every activation definition look better than it is. The denominator for each bucket's conversion rate must be every signup who reached that depth, not just the ones who are still around. A curve built on survivors slopes upward for the wrong reason.
Set the threshold at the knee
Your activation threshold belongs at the knee of the curve — the action count past which the marginal conversion lift per additional action drops sharply — as long as a meaningful share of signups can realistically reach it. If three actions take you from 20% to 55% conversion but the fourth only moves it to 58%, three is your bar: it captures almost all the available lift while remaining a reachable onboarding target. The knee turns a judgement call into a reading off the data, and it gives product and lifecycle teams one unambiguous definition to aim at, the same way a derived activation funnel from signup to first value does.
Watch two failure shapes. If the curve keeps climbing with no knee, your action list is probably too coarse — you are measuring something that behaves like general engagement rather than a small set of discrete value milestones, so split the actions finer. If the curve is flat, the actions you chose simply do not predict conversion, and no threshold will help; go back and enumerate better candidates. A good core-action set produces a curve with a clear steep section and a clear flattening, which is exactly what makes the threshold decision obvious.
Use the curve to prioritise onboarding
The dose-response curve is not just a threshold-setting tool; it is a priority list. The actions in the steep region are the ones where moving a user up one step buys the most conversion, so they are the ones your onboarding checklist, empty states, and lifecycle emails should drive hardest. Instead of nudging every action equally, you concentrate effort where the marginal lift is largest — the second core action if that is where the curve climbs fastest, or the specific integration step that separates the 1-action bucket from the 3-action bucket. A common heuristic for picking which actions belong on the list at all is to find the ones completed by most converters and few non-converters; Tandem frames it as looking for actions appearing in 80%+ of converted users but less than 20% of churned users.
Segment the curve before you act on it. A blended curve can hide that self-serve signups and sales-assisted signups have different shapes, or that one acquisition channel delivers users who complete more actions for reasons unrelated to your onboarding. Comparing the curve across acquisition and behavioural cohorts tells you whether the dose-response is a product property or a mix effect, and it stops you from optimising onboarding for a pattern that only exists in one segment's blend.
Depth is correlation until you test it
A rising dose-response curve does not prove that making users complete more actions will make them convert. Motivated users complete more core actions and also convert more, so action count is confounded with the underlying intent to buy — the same correlation-versus-causation trap that undoes most activation analyses. If you push a reluctant user through a third action with a modal, you may move the action count without moving the intent that actually drives conversion, and the curve will have lied to you.
The defences are the usual three. Demand a plausible mechanism for why each action delivers value rather than being an incidental side effect; confirm the curve reproduces across cohorts and on a hold-out window the analysis never saw; and, decisively, run an onboarding experiment that nudges a random subset toward the next action and measures whether their conversion rises relative to a control. Depth analysis is the fastest way to generate a ranked list of hypotheses about what to test. It earns you the threshold and the priority order — but only a controlled test turns "users who did this convert more" into "driving this makes users convert." Wire the validated threshold into your north-star metric hierarchy as the earliest checkpoint on the path to revenue.
FAQ
What is activation depth?
Activation depth is the number of distinct core actions a user completes in their early window, treated as a graded 0-to-N score rather than a single yes/no activated flag. Instead of asking 'did this user activate?', you ask 'how many of the value-delivering actions did they complete?' — then measure how trial-to-paid conversion changes at each level. The point is that activation is rarely a cliff; it is usually a slope, and the slope carries information a binary flag discards.
Why is a binary activated flag not enough?
A binary flag collapses everyone above a threshold into one bucket and everyone below into another, hiding the dose-response relationship between effort and conversion. Two users who both 'activated' can convert at wildly different rates if one did the minimum and the other completed five core actions. The binary flag also makes the threshold itself arbitrary: you cannot see whether moving the bar up or down would separate converters from non-converters better, because you only ever look at one cut point.
How do you build an activation dose-response curve?
Define a short list of core value actions, count how many each signup completed within a fixed window (say 7 or 14 days), and bucket users by that count: 0, 1, 2, 3, and so on. For each bucket, compute the trial-to-paid conversion rate against a fixed downstream outcome. Plotting conversion against action count gives the dose-response curve. The steepest part of the curve is where additional actions buy the most conversion; the point where the curve flattens is a natural activation threshold.
Where should you set the activated threshold?
Set it at the knee of the dose-response curve — the action count past which the marginal conversion lift per additional action drops sharply — provided a meaningful share of signups can realistically reach it. A threshold only a few percent of users ever hit describes your power users, not a reachable onboarding target. If the curve keeps rising with no knee, your action list may be too coarse; if it is flat, the actions you chose do not predict conversion and you need better candidates.
Does a dose-response curve prove the actions cause conversion?
No. A rising curve is correlational: motivated users complete more actions and also convert more, so action count is confounded with underlying intent. To move toward causation you look for a plausible value mechanism, check the curve holds across acquisition cohorts and segments, and ideally run an onboarding experiment that nudges a random subset toward the next action and measures whether their conversion actually rises. The curve earns a hypothesis and a threshold; only a controlled test earns a causal claim.