Consumer Validation

What one cohort proved in six months

Klar's market is anyone turning information into capability, at university, at work or in life. We chose to prove the consumer case before the institutional one, and students first inside it. The plan for the spring was to land the largest cohort we could reach at term start, then use the rest of the term to test it hard enough to know what a multi-campus push in the autumn is worth. Acquisition spend was $0 in every month of the period.

The six findings

• 01ARR reached $207,588 in month one of the launch term, February 2026, from a single campus cohort acquired with no media spend. It prices every live Pro subscription at list, trialling as well as paying, so it is an upper bound rather than a forecast. Section 02 splits it and follows both halves through the term.
• 02Customer acquisition cost is $0 and payback is immediate. All 675 paying customers to date, across 8,036 registered users, were acquired without a single dollar of media spend.
• 03Retention rises monotonically with tenure. 47% of everyone ever charged is still subscribed, 66% of those with at least two payments, 79% with three, 82% with four. Churn is a first-renewal problem, not a product problem.
• 04Engagement deepened far faster than the base grew. Subscriber WAU/MAU went 29% to 43% and DAU/MAU 8% to 17%. Habitual subscribers, active on eight or more days in a month, went from 6% of the active subscriber base in January to 39% in May.
• 05Distribution turned out to be free. 10 million organic views, 17,000 followers and inbound requests from campuses we never targeted, none of it planned or paid for. July, a month with no term, no exams, no campus presence and no campaign, still drew 1,107 new registrations, more than May and June combined.
• 06A deliberate summer retention campaign cut churn hazard to 2 to 3% in July for the four cohorts past their first renewal, against 23 to 57% in those same cohorts' spring months. The two youngest, still inside their first renewal window, fell to 10% and 15%.

What the term proved, and what comes next

Proved: there is a paying need at full list price, the product holds a daily habit, retention has a floor rather than a decay to zero, and distribution costs nothing. Next: the same playbook across multiple campuses at the autumn term start, priced on the retention numbers in this report.

Definitions for every metric in this document are in Appendix E. Methodology, known limitations and the open questions we are tracking are in Section 7. Nothing in this report is modelled or projected; every figure is a measurement of something that already happened.

Klar is a learning agent, not an edtech product. Learning is not confined to a degree: it is what anyone does when they turn information into capability, at university, at work or in life. The eventual market therefore includes companies and institutions as well as individuals. We chose to prove the individual first, and the order was deliberate. It has also turned out to be the way into the institutional and corporate sale rather than a detour around it.

Why consumer before enterprise

• —Consumers are the harshest testers there are. A private buyer either uses the product or does not, and either pays for it or does not. There is no procurement cycle, no internal champion, no committed budget and no company politics to carry a mediocre product through a renewal. Every retention and engagement figure in this report was earned against that bar, which is the bar we wanted to build the product against.
• —A model that stands on its own. We wanted a business that works without an institutional contract underneath it, so that a university or enterprise agreement is upside on something already working rather than the thing holding it up. All 675 paying customers to date are individuals paying the same list price by card, 199 SEK or $19 or €19 a month, with $399 of discounting across the entire period before the summer campaign.
• —Enterprise is a distribution channel, not a discovery channel. Selling into companies and institutions before the product retains means learning what a procurement process wants rather than what a user wants. With the consumer case proven, the same conversation becomes a channel for a product that already holds a habit.
• —Building bottom-up makes the sale at the top easier. Users carry Klar into their own universities and workplaces, so the conversations at that level have come to us rather than the other way round. Campuses outside our launch university have asked us to come to them, and companies have approached us wanting to work together, with no sales motion and no outbound behind either. Walking into that conversation with usage already inside the building, and with the retention curves in this report to hand, is a different sale from walking in cold.

Why students are the first ICP

Students are the sharpest available version of the same need. The stakes are high, the need recurs every week of term, the population is physically concentrated, and the verdict comes fast. They also arrive on a calendar, which makes a whole cohort reachable in a single move at term start and makes the experiment repeatable every term. That is a first ICP, not the market.

What this means for the figures that follow

Everything in this report is consumer revenue: individuals paying month to month by card, with no institutional contract, no pilot, no procurement and no annual commitment anywhere in the base. Nothing here depends on a B2B pipeline, and every subscription can be cancelled at any time, which is the standard the retention figures in Section 4 are measured against.

Students arrive on a calendar. Every new student starts at the beginning of term, which makes acquisition here a step function rather than a curve, and makes term start the moment worth concentrating everything on. The spring plan followed from that.

The plan

• —Land the cohort. One concentrated campus effort in January and February, to take as much of that term's intake as we could reach in a single move. It produced the February peak: 843 live Pro subscriptions and $207,588 of ARR, the largest demand capture of the period.
• —Test it. The rest of the term went to the questions that decide what a term start is worth: do they stay, do they build a daily habit, what does the retention curve do at 30, 60 and 90 days, and how far can the product carry them. Six months of that data is what the rest of this report contains.
• —Scale it. Run the same playbook at the autumn term start, across multiple campuses through partnerships, with the retention and engagement numbers in hand rather than assumed.

What the ARR line shows, and what sits underneath it

ARR here is every live Pro subscription at month end, paying and trialling, priced at list and annualised. It peaked at $207,588 in February, which is what a single campus push produced in a single term, and stood at $115,716 on 31 July. Because it counts trials it is an upper bound, not a forecast: the expected value sits between it and paying ARR below, at a trial-to-paid rate of 37 to 44% in the clean months.

The shape of the line follows the shape of the intake. This series counts trials alongside payers, and the term ran on a single intake rather than a rolling one, so as the February batch resolved into payers the trial half of the number came down with it. Live trials went from 744 at the peak to 72 in July. The paid half of the same base went the other way.

Paying ARR rose in five of the six month-over-month steps in the period, the one exception a 2% dip in June, to an all-time high of $93,493 in July. These are live subscriptions past their trial period; from 6 June most of them sit under a 100% off summer coupon, so the run-rate is real and the billing is paused until 1 September. Section 6 covers that in full. The honest read of the two series together: February shows how much demand one campus push captures, July shows how much of it converted and stayed.

8,036 registered users. 2,045 subscriptions. 675 paying customers. Media budget across seven months: nothing.

The channel that outgrew its brief

Alongside the campus effort we ran an organic content operation, and it compounded well past what we planned for it: 10 million organic views, 17,000 followers, and inbound invitations from campuses outside our launch university. There was no media budget behind any of it, and the term's focus stayed on testing the cohort we already had.

The clearest evidence that this is real distribution rather than vanity reach is July. No term, no exams, no campus presence, no campaign, and it still drew 1,107 new registrations, more than May and June combined and within 8% of April. That arrived with no campaign and no dollar behind it, which is why the autumn push starts from an audience rather than from a standing start.

What this changes for the next term

The batch playbook is repeatable at term start by construction, because the category resets every term. It now restarts from an audience of 17,000 rather than from zero, and from more than one campus, because the invitations came to us. The acquisition motion we validated at $0 is the motion we intend to scale, not a motion we intend to replace with paid.

Conversion through the funnel

View and follower counts are platform-reported and sit outside the database and Stripe scope of this report. Every other figure in this document is reconstructible from the two source systems. Acquisition-source tracking began 13 March 2026, so the January and February spike is unattributed at channel level. Cost, however, is not an unknown: spend was $0 throughout, so all growth to date is organic by arithmetic rather than by inference.

Month-by-month churn mixes two unlike populations: curious first-time payers and committed users. Split them and the picture inverts. The longer a customer stays, the less likely they are to leave.

• —63% of first-time payers make a second payment (300 of 473, clean months only).
• —59% of second payers make a third (177 of 298).
• —From month 3 of a cohort's life onward, monthly loss falls to 1 to 7% per month.

Cross the second payment and two thirds of customers are still with us. Cross the third and it is four out of five. That shape is the definition of a retention problem concentrated in the first renewal: a single identifiable transition rather than a diffuse decay.

Cohort curves flatten, they do not decay

The floor sits at roughly 33 to 40% of the original cohort, or 57 to 69% of everyone who made it past the first renewal. A curve that stabilises rather than decaying toward zero is the difference between a subscription business and a leaky funnel. February (33% at month 4) and March (40% at month 4) are the two cohorts with both the scale and the elapsed time to read it; April and May are tracking the same shape one and two months behind. The January cohort has a full six months but only 12 customers, and it sits below this range at 17 to 25%, which is why we do not read a floor off it.

Six-month evidence

Of the 96 customers who first paid in January or February, 36% are still subscribed five to six months later, and of those who made a second payment, 57% are. Nearly half of every customer we have ever charged is still a subscriber today.

The single highest-leverage number in the model

One transition, first payment to second, accounts for 244 of the 356 customers we have ever lost, or 69% of all churn. No dedicated intervention has been run against it yet, so every figure in this report reflects the unimproved rate.

Between January and May, at a constant price and an unchanged product category, subscriber stickiness roughly doubled. This is what the testing term was for, and it is the return on spending it on the product.

A 43% WAU/MAU means the average engaged subscriber returns in a little under half of the month's weeks, roughly twice the commonly cited band of 15 to 25% for consumer subscription apps. And it is a lower bound: activity is counted as chat messages only, so reading sources, revising flashcards, taking practice exams and listening to podcasts do not register as usage at all.

Habitual users went from 6% of subscribers to 39%

• —Habitual subscribers, active on eight or more days in a month, went from 14 to 119, an 8.5× increase.
• —Over the same window the active subscriber base grew 1.4× (223 to 306 paying MAU), so habit density went from 6% to 39% of active subscribers.
• —Messages per active subscriber went from 25 in January to a peak of 103 in April, four times as much work inside Klar per user.

June and July fall because Swedish universities finish exams in late May. January to May is the trend, June to July the summer floor, and the same shape should recur every year: the autumn term is a second January, not a recovery.

Swedish universities are out of session from early June, which is the highest churn-risk stretch of the academic year. From 6 June we applied a 100%-off coupon to loyalty-enrolled subscribers, running to 1 September. This section tests whether it worked.

Monthly churn hazard by cohort. For each calendar month, the share of a cohort's subscriptions that were live at the start of the month and cancelled during it. Because each line follows one cohort across its own lifetime, cohort age and size are held constant. This is a rate, not a count, so it is not flattered by fewer new signups over the summer.

What happened to churn

• —All four cohorts past their first renewal fell to a 2 to 3% monthly churn hazard in July, against 23 to 57% across their own March to May months. January 45% to 2%, February 46% to 3%, March 40% to 3%, April 57% to 2%.
• —The two youngest cohorts, still inside their first renewal window, fell to 10% (May) and 15% (June). For May that is its lowest recorded month; June has only one month of history, so it is a starting point rather than a trend.
• —June is the transition month, with the hazard already halved for the older cohorts (January 39% to 13%, February 34% to 15%).
• —Measured across all six cohorts simultaneously.

Age-matched survival

Comparing cohorts at the same age removes the advantage a younger cohort gets from simply having had less time to churn.

The May cohort has the best 60-day survival of any cohort, 45% against a 21 to 40% range, on 241 subscriptions, and its first two months fell squarely inside the campaign window. The June cohort leads at both day 15 and day 30, so its strength is not an artefact of one cut point; on 53 subscriptions its day-30 figure carries a 68 to 90% confidence interval, so we treat it as directional rather than conclusive.

The free period is constant across every cohort

Every cohort in the window had a free first month through the standard trial. The free period is the constant across all cohorts, not the variable. The June cohort still leads at day 15 (92% against a 68 to 78% range) and day 30 (79% against 44 to 66%), though on 53 subscriptions January's 66% at day 30 is the one figure it does not clearly beat.

Run-rate against collected cash

Paying MRR hit an all-time high of $7,791 in July on $812 collected. $12,477 of revenue was waived across June and July by design, against 324 coupon-carrying subscriptions in June and 344 in July. The subscriptions are intact; only the billing is paused. Set against $3,186 collected across those two months, we gave up roughly four fifths of the summer's cash for a base that walks into the autumn term whole.

Paying MRR at list price against cash actually collected. The gap in June and July is the campaign, not attrition.

Every number above can be checked against the source systems. The questions a diligent reader would raise are set out here, with our answer next to each.

The September test

Every summer retention figure in Section 6 measures customers holding a subscription they are not currently paying for. Retained is not the same as renewed. Some of the suppressed churn is deferred rather than eliminated: customers kept through a free summer who would have left in June or July may instead leave in September. The coupon expires on 1 September and the loyalty ladder price resumes. The size of that step-down is the number that settles whether the trade paid for itself. We will measure it and publish it against this report.

Base sizes and what is load-bearing

February and March paying retention sits on bases of 8 and 99 customers. Retention of 50% on a base of 8 is noise, not a trend, and we do not treat it as one. The June cohort is 53 subscriptions, so its day-30 survival is directional. The load-bearing evidence for the summer campaign is the cross-cohort churn hazard table, which follows every live cohort at once and is a rate rather than a count.

Seasonality is structural, not a decline

Swedish exam periods end in late May. June and July have no exams to study for, and the campaign sits on top of that. The honest signal is the January to May trend. The same seasonality should be expected every year, which also means the autumn term is the next real data point, not a recovery from a bad quarter.

Run-rate against cash

Both ARR series in this report are list-price run-rates, not collected cash, and cash collected is printed beside them in every table. The headline series counts every live Pro subscription including trials, so it is an upper bound: it is the run-rate the business would have if no trial ever cancelled. Read it against trial-to-paid of 37 to 44% in the clean months for expected value. Paying ARR counts only subscriptions past their trial end and is the floor. Figures are gross of Stripe fees, roughly 3 to 4%, and FX uses a single fixed rate rather than the daily rate Stripe settled at, so expect ±2 to 3% on the USD column.

We treated this term as the gate that has to be passed before spending money is rational. These are the four things it cleared.

• 01A paying need. 675 customers paid full list price in a segment famous for not paying for anything. Trial-to-paid ran 37 to 44% in the clean months.
• 02A habit. 44% peak subscriber WAU/MAU and 119 habitual subscribers, measured on chat activity alone, with four other product surfaces uncounted.
• 03A retention floor. Cohort curves flatten at 33 to 40% of the original cohort; 82% of the 50 customers with at least four payments are still subscribed; 47% of everyone charged is still here.
• 04Free distribution. 10 million organic views, 17,000 followers, $0 spend, and inbound demand from campuses we did not target.

What the autumn term adds

Scale, and only scale. The spring intake was one campus run by hand, which is the right size for a test and the wrong size for a business. The autumn applies the same playbook across multiple campuses through partnerships, with a product measurably stickier than the one the spring cohort met.

The lesson we take into the autumn

Two things stood out that we had not set out to test. Users we onboarded in person on campus converted and retained better than those who arrived on their own. And the product held its retention and stickiness with no onboarding flow at all, which was the bar we set for it: it had to work on its own before we propped it up. Every engagement figure in Section 05 is what a product with no onboarding produced, rising through the term on product work alone.

Both point the same way, so the autumn adds what was deliberately absent. In the product, an onboarding flow that gets a new user to the first moment of real value faster. On the ground, a stronger in-person onboarding motion at term start, which carries straight into partnerships and company customers, where arrival is always mediated by someone. Neither existed this term, so neither is priced into any number in this report.

The channel difference is an observation rather than a measurement: acquisition-source tracking only began on 13 March, so this dataset does not separate campus-onboarded users from the rest. It was consistent enough in what we saw to act on, and the autumn term will measure it properly.

The arithmetic we now know

We can acquire a term cohort at $0 CAC. Roughly 37 to 44% of trials convert to paid. Roughly a third to 40% of a paying cohort persists to a stable floor, and 82% of the 50 customers with four or more payments so far are still subscribed. Depth of use compounds through the term rather than decaying. Every incremental campus is that same arithmetic, and the term boundary is when it is run.

Sections 01 to 08 are a measurement of the January to July 2026 term: no forecast, no modelled cohort, no assumed improvement. Section 09 is the one forward-looking section in this document and is labelled as such throughout. The next measured data point is the 1 September renewal, and the one after that is the autumn term intake.

The spring term proved the unit on one campus, acquired by hand. Partnerships put the same playbook in front of populations we do not have to acquire at all. Two agreements are signed and the target is ten by year end.

Everything on this page is forward-looking. The figures below are modelled from the measured rates in Sections 03 to 05 applied to partner populations, not observed revenue. Every other figure in this report is a measurement.

Signed

• —Meitner. Sweden's leading and fastest-growing school platform. 150,000+ users secured for rollout under the agreement.
• —SERO. University partnership, signed, rolling out on the same model.

What the secured base is worth

The $6.4M baseline takes the trial-to-paid and retention rates measured in this report, at today's price, and applies them to users we already have contractual access to. It assumes no improvement in conversion, no improvement in retention and no change in price, all three of which the autumn term is set up to move. The $36M and $174M figures are ceilings rather than forecasts: full penetration of the secured base, and the same arithmetic across ten partnerships.

Klar · Cohort KPI Validation Report · January to July 2026