The product did not become remarkable because it produced more text. Users treated it as a thinking surface: their notes shaped the AI output, and the finished document still felt like theirs. The long beta gave the team enough repeated meetings to see that distinction, which a launch-week activation chart would have hidden.
COMPANY PLAYBOOK 12 / 15
THE GRANOLA SYSTEM
Granola
150-person closed beta for a year before a 500-install launch day
Earn word of mouth by understanding the moment a habit becomes indispensable.
THE QUICK READ
Granola launched into a crowded meeting-notes category without paid acquisition or a forced sharing loop. The team spent a year with roughly 150 beta users, missed its strongest use case for months, and used close observation to find the behavior users would recommend on their own.
ProductAudience
01 / WHAT THE RECEIPTS SAY
02 / BACKLOG-READY PLAYS
3 plays worth stealing.
Open a play for the mechanism, the exact receipt, the failure mode, and a deployment brief Elena can put into your backlog.
30 Keep the beta small enough to observe a habit
THE MOVE
Granola stayed in closed beta for about a year with roughly 150 people instead of optimizing for a large launch list.
WHY IT COMPOUNDS
A meeting product reveals value over repeated use. A small cohort lets the team inspect sequences across many meetings, compare what users say with what they keep doing, and notice when a personal workflow becomes habitual.
THE RECEIPT
The team says those 150 users taught it more than a broad launch would have. Granola recorded about 500 installs on launch day without paid marketing.
STEAL THIS
Choose a cohort small enough that one owner can review every retained user each week. Track repeated jobs, manual workarounds, and the first unprompted recommendation.
DO NOT COPY BLINDLY
A long beta is useful only when observation density is high. Time behind a gate is not learning by itself.
RECEIPT STATUS
Operator-reported. Christopher Pedregal reports the one-year beta, 150 users, and launch-day installs.
Open the exact source ↗ ELENA DEPLOYMENT BRIEF
Turn this observation into a real experiment.
- Run it when
- Value appears only across a repeated workflow and session-level feedback is misleading.
- Owner
- Product lead and user researcher
- First sprint
- Recruit 50 matched users, inspect every retained user weekly, and log repeated jobs, workarounds, and recommendations for eight weeks.
- Leading signal
- Frequency by user, habit depth, unprompted recommendation, and qualitative saturation.
- Stop rule
- End the closed beta when new cohorts repeat known patterns or when observation no longer changes decisions.
31 Refuse the growth loop your product has not earned
THE MOVE
Granola grew without forced sharing, automated invitations, or a built-in referral mechanic. Users recommended the private tool in conversations where the problem was already salient.
WHY IT COMPOUNDS
Voluntary recommendation carries more information than a branded export. The sender is staking personal credibility on the product, and the recipient receives it at the moment of need.
THE RECEIPT
The founders describe growth as viral despite having no engineered viral loop. Investors told TechCrunch they repeatedly heard about the product from other investors who used it.
STEAL THIS
Interview referred users about the exact sentence, situation, and relationship that produced the recommendation. Improve that moment before adding incentives.
DO NOT COPY BLINDLY
Do not romanticize word of mouth. If referred-user volume and activation cannot be measured, the team cannot tell a strong loop from a small social bubble.
RECEIPT STATUS
Operator-reported. Pedregal explicitly describes viral growth without forced sharing or automated loops.
Open the exact source ↗ ELENA DEPLOYMENT BRIEF
Turn this observation into a real experiment.
- Run it when
- Users already recommend the product in high-context conversations.
- Owner
- Product growth and research
- First sprint
- Interview 20 referred users and their senders. Capture the exact trigger, sentence, relationship, and time to value.
- Leading signal
- Referral share of activated users, sender-to-recipient conversion, and referred retention versus baseline.
- Stop rule
- Do not add incentives if referred users are low fit or the recommendation depends on a narrow social cluster.
32 Measure retention as a field of users, not one average
THE MOVE
The Granola team uses user-level views of behavior and feedback to see who is deepening a habit, who is flattening, and which product changes move each group.
WHY IT COMPOUNDS
An aggregate retention line can improve while important cohorts deteriorate. A user-level dot plot keeps the team close to the distribution and makes qualitative follow-up targetable.
THE RECEIPT
Pedregal describes short explore-and-exploit cycles and a retention dot-plot mindset as central to finding and extending the product’s strongest behavior.
STEAL THIS
Plot every active account by frequency and depth. Annotate releases and interviews on the same view. Pick the next sprint from a visible cluster, not a blended average.
DO NOT COPY BLINDLY
User-level analysis breaks at scale unless the team defines which segment and behavior it is diagnosing before opening the chart.
RECEIPT STATUS
Operator-reported. The user-level retention view and explore-exploit cadence come from Pedregal’s operating account.
Open the exact source ↗ ELENA DEPLOYMENT BRIEF
Turn this observation into a real experiment.
- Run it when
- A blended retention curve hides very different user trajectories.
- Owner
- Product analyst and product manager
- First sprint
- Plot active accounts by frequency and depth, annotate releases, choose one cluster, and pair the chart with five interviews.
- Leading signal
- Movement of the target cluster, release-level behavior change, and retained depth after four weeks.
- Stop rule
- Change the segmentation if the chosen cluster cannot be explained or influenced by a specific product decision.
03 / MAKE IT YOURS
Do not copy Granola. Adapt the system to your constraint.
Elena learns your product, customer, funnel, and current bets. Then she chooses the relevant pattern, scopes the first sprint, and watches the leading signal.