Skip to content
AppDues
Log in
Contact salesSign up

Growing paid tiers without gutting the free tier

A strong free tier and a healthy subscription business can pull in opposite directions. Life360's experience, as described by Giordano Contestabile in a Sub Club conversation with David Barnard, is a useful case for teams facing that tension: the company treats certain free features as untouchable, runs monetization as a portfolio rather than a string of one-off tests, and got its clearest win from a model that changed which tier was shown — not what was given away. This guide restates the operating logic and turns it into checks you can apply to your own paywall work.

The rule that comes before the test

Six months into his tenure as VP of Product, Contestabile ran an experiment whose numbers pointed to a large revenue gain. He brought it to the executive team expecting praise and instead met anger, because the test had crossed a line the company had already drawn.

That line was what then-CEO Chris Hulls called the "freemium bill of rights": a set of core principles holding that specific features — especially anything tied to family safety — stay available to free users and are never withdrawn to push a conversion. Contestabile's own summary of the user research is blunt: the most common reason people give for not subscribing is that the free version already does enough. The company's answer is to add value and widen the paid offering rather than subtract from the free one.

With close to 100 million monthly active users, the free tier is also the growth engine. The product depends on network effects — it does little in solo mode — and its main discovery path is a parent recommending it to another parent. Putting core utility behind a paywall could lift short-term revenue while damaging the loop that brought people in.

Before you design a monetization test, write down which capabilities are off limits and why. A one-line charter is enough: the features that make the product work for a non-paying user, the safety or trust functions you will not gate, and the growth mechanism you refuse to break. Tests that violate it should not reach the roadmap, no matter how good the projected number looks.

Run experiments as a portfolio, not a queue

Scaling experimentation at Life360 meant changing how results were judged. Contestabile describes the program as a portfolio whose output depends on three levers: velocity, or how many experiments ship; win rate, or the share that succeed; and average win size. Notably, a low win rate is not treated as a problem.

The outcome that draws real frustration is an inconclusive test — one that produces no usable answer, usually because the setup was wrong or the hypothesis was poorly formed. A loss is different: it settles a question. And a losing test often carries a second finding, showing that a feature failed across the board but landed strongly with a narrow group, such as suburban users who had been on the platform about a month and own a dog. Those segmented learnings become input to the next round of tests.

Practical implication: judge the program by how many questions it answers, not by how many tests it wins. Track inconclusive rate as a quality metric alongside velocity and win size, and require every test to state the decision it will inform before it launches.

The Platinum test: change the offer, not the product

Life360 sells three tiers — Silver, Gold and Platinum — and most users historically picked Gold. Presenting all three clearly inside a paywall flow is hard without overwhelming the buyer, so the team usually defaulted to showing Gold.

To test that default, the team built a machine learning model using roughly 900 distinct data points to score users on their likelihood of buying Platinum. When a user scored high, the flow presented the Platinum offer in place of Gold. The reported result: the share of new users subscribing to Platinum doubled, and no Gold subscribers were lost — an outcome Contestabile calls surprising.

The interpretation offered is that the model surfaced users who would not have converted on Gold anyway but fit Platinum's specific benefits. Treat that reading as the hypothesis to verify in your own data, not a guaranteed mechanism. The important structural point is that the experiment changed which offer a user saw, while the free experience and the existing tiers stayed intact.

A worked example of the flow, using only the captured facts:

  • Signal set: about 900 distinct data points per user feed the propensity model.
  • Decision point: the paywall flow, where the default is Gold.
  • Action: high-propensity users see Platinum instead of Gold; everyone else sees the default.
  • Primary metric: share of new users subscribing to Platinum.
  • Guardrail metric: count of Gold subscribers lost — reported as zero in this test.
  • Reading: a lift counts only if the guardrail holds; a Platinum gain paired with Gold erosion is a mix shift, not growth.

When the audience shares offline, digital referrals stall

Not every data-driven bet worked. Because 40% of users said they found the app through word of mouth — typically parents talking at school pickup — the team tried to move that behavior into the product. They added in-app referral mechanics, offered free Silver subscriptions to users who invited friends, and tested several sharing-button variants.

Contestabile's account of the outcome is that all of it failed. His explanation: the demographic in question — families, older users — does not behave virally on mobile, while teenagers, who do share readily, are not the ones choosing and paying for the app. The lesson drawn is that a digital referral loop cannot simply be imposed on an audience whose sharing happens in person.

Use this as a design caution rather than a law. If your acquisition depends on offline conversation, test whether the in-app step removes friction from that conversation — a shareable summary, a join code, a one-tap invite a parent can hand over — instead of assuming the invite itself is the missing piece.

A pre-launch checklist for monetization tests

Use these checks before a tier or paywall experiment goes live:

  1. Charter check: does the test leave every protected free feature in place?
  2. Hypothesis check: is the expected outcome stated in a form that can be confirmed or refuted?
  3. Decision check: which roadmap or pricing decision will this result change?
  4. Guardrail check: which existing tier or cohort must not shrink for the win to count?
  5. Segmentation check: if the overall result is flat or negative, will you still learn which cohort responded?
  6. Inconclusive check: what sample, duration or instrumentation would make the result unreadable, and how will you avoid that?
  7. Channel check: if the test depends on user sharing, does the audience actually share in the channel you are building for?
Propensity-based tier selection: inputs, controls and guardrails
StageWhat happensEvidence status
Signal collectionAbout 900 distinct data points per user feed the propensity model.Reported in the episode
ScoringModel estimates likelihood of buying Platinum.Reported in the episode
Offer decisionHigh-propensity users are shown Platinum instead of the default Gold offer.Reported in the episode
Primary metricPercentage of new users subscribing to Platinum.Reported as doubled
Guardrail metricGold subscribers lost during the test.Reported as zero
InterpretationGain attributed to users who would not have converted on Gold but matched Platinum benefits.Stated as the team's reading, not independently verified

Figures reflect the single test as described in the podcast conversation and have not been independently verified; treat them as a dated case reference and re-measure against your own baseline.

Reading experiment outcomes: loss versus inconclusive
OutcomeWhat it tells youWhat to fix next
WinHypothesis supported at the measured level.Check guardrails, then look for the segment driving the lift.
LossHypothesis refuted; the feature did not work broadly.Mine cohort-level results for a narrow group that did respond.
InconclusiveNothing settled.Fix setup or hypothesis formation before rerunning; this is the outcome to minimize.

Framing follows Contestabile's stated view of experiment portfolios; thresholds for what counts as conclusive depend on your own statistical standards.

Frequently asked questions

Why didn't Life360 just weaken the free tier to boost conversions?

Because the most cited reason users give for not subscribing is that the free version already meets their needs, and the company's stated approach is to respond by adding value and broadening the paid offering rather than removing free features. Core family-safety functionality is treated as permanently free, and the free tier also drives the network effects and parent-to-parent discovery the product depends on.

What counts as a failed experiment in this portfolio model?

An inconclusive result — one that answers nothing, typically because the test was set up incorrectly or the hypothesis was poorly framed. A losing test is still useful because it confirms or refutes the hypothesis, and it often reveals a narrow cohort that responded well even when the overall result was negative.

How did the machine learning model double Platinum subscriptions?

The model scored users on roughly 900 distinct data points to estimate who was likely to buy Platinum. For high-propensity users, the paywall flow presented Platinum instead of the usual Gold default. The reported result was a doubling of the share of new users subscribing to Platinum with no Gold subscribers lost.

Why did in-app referral features fail?

The team built referral mechanics, offered free Silver subscriptions for invitations, and tested several sharing-button variants, and none of it worked. The explanation given is that the paying audience — families and older users — does not share virally on mobile, while teenagers, who do, are not the ones installing and paying for the app.

Does a Platinum lift always mean growth?

Only if the guardrail holds. A rise in top-tier signups accompanied by a drop in mid-tier subscribers is a mix shift rather than added revenue. In the reported test, no Gold subscribers were lost, which is what made the result meaningful — so pair any tier experiment with an explicit metric for the tier you must not cannibalize.