PastBehavior / Product Architecture
When another person’s participation is part of the job
Most products treat sharing as something that happens after the useful part. You buy something, finish something, or create something, and then the product asks you to invite someone, refer a friend, post it, or send a link.
A different class of products works the other way around. Another person’s participation is part of completing the job. An event needs guests. A meeting needs participants. A group trip needs people to commit. A group purchase needs buyers.
That architecture guarantees something traditional sharing does not: exposure to another participant through normal product usage.
But exposure is not acquisition. Acquisition is not propagation. And the research found almost no public evidence showing that recipients reliably become future senders. That distinction is the story of this research.
18 classified cases · 8 negative controls · 3 matched pairs attempted, 0 surviving
Evidence labels: EVIDENCED (filing, documentation, or on-record statement), STRONG INFERENCE (conclusion drawn across cases, labeled as such), WEAK SIGNAL (thin or third-party evidence), UNDERDETERMINED (the public record cannot answer it).
Headline finding
Sendability is real. The growth advantage is unproven.
The strict architecture survives testing. Products can be separated cleanly based on whether another person’s action is necessary for the intended job to complete.
The commercial hierarchy does not. No public evidence base establishes that required coordination produces stronger recipient-to-sender propagation than optional or economically consequential participation.
Underdetermined on the headline empirical claim
Research universe
The brief called for 20–30 cases and 3–4 matched pairs. The research stopped at 18 cases and zero valid matched pairs rather than padding the evidence base or constructing favorable comparisons.
Executive summary
Real and definable. The strict test — can the sender's job complete without the recipient — cleanly separates Partiful, Zoom and Pinduoduo from Amazon Inspire, Duolingo and Splitwise.
Evidenced
No public evidence base can answer this. Four companies whose growth narratives depend on it declined to quantify it.
Underdetermined
Whether the second person's action gates completion or merely improves the outcome. The distinction holds under testing and reclassifies several products commonly cited as social.
Evidenced
Events, group travel, and multi-party meetings — where the real-world job is already multi-person and the coordination currently happens in messaging apps.
Strong inference
Invites and claimable inventory. Both carry a deadline, a named recipient, and an unambiguous next action.
Weak signal — rests on two cases
Zoom and Partiful both describe no-signup recipient action as central, and both grew fast. The one matched pair built to isolate this variable (Zoom against Webex) collapsed because the variable was not cleanly present.
Underdetermined
In financial services the pre-transaction version is the one that is legally buildable — a different and stronger claim than outperformance.
Strong inference on buildability · Underdetermined on performance
Pre-commitment collection against a defined event. Free-participant-to-paid-host conversion (Zoom, quantified). Group-threshold pricing (Pinduoduo, inseparable from subsidy).
Strong inference
Duolingo is the strongest: no sendable mechanic at all, and the best organic acquisition in the set. Organic growth does not require recipient necessity.
Evidenced
No company discloses this, including four that assert it.
Underdetermined from public sources, and structurally so
Where revenue occurs inside the coordinated action rather than beside it. BeReal is the counterexample: strong coordination behavior, no transaction to attach to, no viable model.
Strong inference
Pre-commitment collection against a defined event with an organizer. Ledger-only tracking. Not formalized informal debt.
Strong inference
Move the coordination that currently happens in the group chat inside the product, at the point money changes hands. Three of the eight brands examined return no viable mechanic.
Weak signal — hypotheses, not tests
Organic acquisition without any sendable mechanic (Duolingo). Required coordination that only worked because it was subsidized (Pinduoduo). The absence of any isolable natural experiment.
Evidenced
The entire causal core. See the open-questions register.
—
01 / The distinction
The claim under test: the strongest distribution loops do not ask people to share — they make another person’s participation necessary to complete the job. The test that makes this examinable is whether the mechanic sits beside the transaction or inside it. A share button, a referral code, and a comment feed sit beside; the customer can finish without them. A meeting invite, a group-purchase threshold, and an event RSVP sit inside; the customer cannot finish without them.
Tier 1
Load-bearing: YesWithout the second participant, the intended job cannot complete.
Zoom · Partiful · Eventbrite · SquadTrip · Braid · DoorDash group ordering (narrow job definition)
Tier 2
Load-bearing: NoThe job can complete without the recipient, but participation changes price, access, inventory, or economics.
Pinduoduo team purchase
Tier 3
Load-bearing: NoThe second person improves the result but is not required.
Airbnb wishlists · Splitwise · Tricount · Zelle (group-expense context)
Tier 4
Load-bearing: NoThe social behavior sits beside the job.
Amazon Spark · Amazon Posts · Amazon Live · Amazon Inspire · Venmo's global feed · BeReal · Duolingo
The tiers describe architecture, not proven commercial performance. Nothing below should be read as evidence that Tier 1 outperforms the others — that question is open, and Section 18 explains why.
02 / Methodology in one example
Whether a participant is “required” depends on what job the researcher says the customer is completing. DoorDash group ordering makes the problem concrete.
Define the job as
“Everyone gets the meal they chose.”
Recipient action is necessary — each person adds their own items.
→ Tier 1
Define the job as
“Dinner arrives.”
One person can order for everyone. Recipient participation is not load-bearing.
→ Not Tier 1
Both descriptions are available in the company’s own material. The tier follows the job definition, so the research recorded the functional job before assigning the tier and rejected definitions narrowed merely to manufacture Tier 1 cases.
Methodological note: Partiful survives this test more cleanly — no guests means no event, regardless of how narrowly or broadly the job is described.
03 / The core measurement problem
A sendable architecture guarantees the first step of a chain. The thesis requires the whole chain. These are not the same thing, and the distance between them is where the evidence runs out.
1
The product necessarily reaches another person.
2
They interact with the workflow.
3
They become an activated user or customer.
4
The missing metricThey later originate the same type of workflow.
5
The process repeats.
What companies disclose at stage 4
Four companies whose growth narratives depend on propagation describe the loop publicly. None publishes the recipient-to-sender conversion rate.
Companies publish the conversion that bills. They do not publish the social step that precedes it.
04 / Positive case
55%
of Zoom’s 344 customers contributing more than $100,000 of revenue in FY2019 started with at least one free host before subscribing.
Evidenced
This is the strongest quantified case in the research, and it still does not measure recipient-to-sender propagation. The metric sits one step downstream of the propagation event the thesis needs.
05 / Positive case
Reported usage
~500,000 MAU, Q1 2025 avg
+400% year over year
Weak signal — third-party estimate
The company has described usage in the millions while app trackers show substantially lower installed-user numbers. The gap is directionally consistent with something important: many people may touch a Partiful invite without becoming installed users.
Exposure can be much larger than acquisition.
Strong inference — not proof
06 / The confounded case
Small groups of buyers — typically two to five people, coordinating through WeChat — formed teams to unlock lower prices. This is not load-bearing under the strict definition, because a user could purchase alone. The second participant changes the economics.
In correspondence with the SEC, Pinduoduo itself tied coupons, credits and discounts to the team-purchase model that encouraged users to share and recruit friends and family.
Evidenced — primary regulatory correspondence
The coordination mechanic and the subsidy were one program.
The economics
The honest reading: required coordination worked here because it was subsidized — which makes Pinduoduo a poor template for any company without a comparable balance sheet.
07 / Negative control
2017 → 2019
Shoppable social feed.
Shut down two years after launch.
Launched 2019
Brand-led social feed.
Underutilized.
Launched 2019
Livestream commerce.
Struggled for adoption.
Dec 2022 → Feb 2025
TikTok-style shopping feed inside the app.
Discontinued after roughly 26 months.
Negative control — evidenced
Amazon had the traffic, catalog, transaction infrastructure, creators, and budget. Four successive attempts at social layers beside the shopping workflow failed to become durable distribution products. The pattern does not show that the shared architecture caused every failure — but it weakens the argument that adding social behavior around a single-player transaction is sufficient.
08 / Negative control
The job is learning a language. No second person is required. Leaderboards, streaks and social surfaces sit beside the learning job. Yet Duolingo’s S-1 explicitly describes organic growth driven primarily by word of mouth rather than paid acquisition. Evidenced
The strongest organic-acquisition story in the case set does not have a sendable architecture.
That severs the inference from organic growth to recipient necessity: organic distribution can emerge from a remarkable single-player product. This does not disprove sendability. It does show that sendability is not necessary for organic growth — and that organic share cannot be used as evidence for sendability anywhere in this report.
09 / Negative control
BeReal demonstrated that coordinated social behavior can be valuable to users without creating an obvious business model.
Coordination appears commercially strongest when revenue sits inside or immediately adjacent to the coordinated action.
Strong inference
10 / The pattern that holds
Revenue inside the action
Stronger commercial structure
Revenue outside the action
Weaker or less demonstrated
Coordination as pure software monetizes weakly. Coordination adjacent to money changing hands monetizes well.
Strong inference — not a causal law
11 / Financial services
76%
of Gen Z respondents who fronted money for a group trip or event said they were not fully repaid
55%
said it created tension or damaged a relationship
47%
went into debt covering it
1 in 5
canceled plans or muted a group chat to avoid the conversation
Evidenced — company research
After the obligation forms
Splitwise · Tricount
Before the obligation forms
SquadTrip
The buildable opportunity appears to be collection before an obligation forms — not better software for formalizing debt afterward.
Strong inference
Collection against a defined event works because the organizer functions like a merchant taking payment for a known deliverable. Money moves forward. Nobody extends credit and nobody collects a debt.
| Model | Example | Classification |
|---|---|---|
| Ledger without obligation | Splitwise, Tricount | Coordination problem. Buildable, weak monetization. |
| Pooled multi-user wallet | Braid | Unclear. One funded failure, cause unverified. |
| Pre-commitment collection | SquadTrip | Coordination problem. Buildable and monetizable. |
| Formalized informal debt | None surviving | Regulated obligation problem. |
A pooled, FDIC-insured multi-user wallet with a branded debit card. $10 million raised from Index Ventures, Accel and others. Shut down after four years.
Regulation is a candidate explanation, not a proven one. Weak demand, poor frequency, bad acquisition economics and adequate incumbent workarounds remain competing explanations — one funded failure cannot distinguish among them.
12 / A research result, not an embarrassment
The brief required pairs selected on similarity of job, mechanic, transaction structure and evidence availability — before commercial outcome. Recording the failures rather than substituting favorable pairs is the point of the rule.
Failed
Job: Buy consumer goods online.
Intended variable: Social beside the transaction versus participation inside it.
Retained as an architectural illustration only. Not evidence for the tier hierarchy.
Failed
Job: Hold a video meeting with specific people.
Intended variable: Recipient friction — can the invitee act before converting.
Not runnable
Job: Sell tickets to a self-serve event.
Required-coordination products tend to win or lose on many attributes simultaneously, and the cases that would isolate coordination as the single variable are private, historical, or nonexistent. The failure to find a clean matched pair is itself evidence about the limits of the public record — and this page does not manufacture a comparison simply because a comparison would look better.
13 / Reading the record carefully
Three dimensions run through this case set, and they do not move together.
How structurally necessary the second participant is.
Whether acquisition and propagation economics are visible at all.
Whether the company files documents anyone can read.
Tier 1 architectures may disproportionately exist in younger private companies. Private companies publish the least acquisition and propagation data. Public companies produce more audited evidence but may be more likely to have mature Tier 2–4 mechanics.
Better disclosure is not better performance.
The public record may systematically make optional mechanics look better evidenced than required ones. That is a warning about the data — not a claim that Tier 1 is secretly superior.
14 / The case matrix
Every case in the research, with the classification exactly as recorded. Negative controls, weak signals and experiments are shown, not hidden.
Positive · Evidenced
Remove signup from the recipient path
Positive · Weak signal
The invite is the product, not marketing for it
Positive · Claimed, unquantified
Naming a loop is not measuring one
Positive (confounded) · Evidenced (on confound)
Required coordination can be a subsidy delivery vehicle
Experiment · Weak signal
Collect before the obligation forms
Experiment · Weak signal
Job definition determines the tier
Experiment · Claimed, unquantified
Organic share is not a loop metric
Positive (weak monetization) · Evidenced
Ledger-only avoids regulation and monetizes weakly
Positive (weak monetization) · Evidenced
Competitor pressure caps ledger pricing
Experiment · Evidenced (on the problem)
The demand is documented; the product is not built
Negative control · Evidenced (outcome only)
Tier 1 is not sufficient
Negative control · Strong inference
Decorative social is removable
Negative control · Evidenced
Maximum resources do not rescue Tier 4
Negative control · Evidenced
—
Negative control · Evidenced
—
Negative control · Evidenced
Four attempts, same architecture, same result
Negative control · Evidenced
Coordination without commerce has no model
Negative control · Evidenced
Severs the organic-growth-implies-sendability inference
Tier distribution: Tier 1 = 6, Tier 2 = 1, Tier 3 = 4, Tier 4 = 7. Type distribution: Positive = 6, Experiment = 4, Negative control = 8.
15 / From evidence to action
If the job completes without them, do not assume a social mechanic creates distribution. Amazon spent four attempts learning this.
Zoom and Partiful — the two products with the cleanest architecture — both minimize recipient friction.
Strong design inference, not experimentally proven
The closer the money sits to the workflow, the stronger the commercial evidence appears. BeReal is the cost of getting this wrong.
Pre-commitment collection is buildable. Formalized informal debt is a lending and collections product.
This is the metric the public market does not provide. Nobody is benchmarked on it, so a company that measures it knows something its competitors do not.
A recipient seeing the product is only the first step of the funnel.
What the evidence does not support: any claim that required coordination outperforms economically consequential participation. Tier 1 is structurally more defensible because usage necessarily produces exposure. Whether it converts better than Tier 2 is unknown.
16 / Hypotheses, not findings
Eight brands were examined, each with a current-feature existence check performed first. Three returned no viable mechanic — which was the point of the kill criterion. Requiring a mechanic per brand manufactures one.
Ticketmaster hasn't built this. Imagine:
A temporary block of seats is held while each friend claims and pays for their own seat.
DoorDash already has group ordering. The hypothetical gap is payment.
Each participant commits and takes payment responsibility before the organizer fronts the bill.
Airbnb hasn't built this exact mechanic. Imagine:
The booking confirms only after required travelers approve dates and commit their portion.
Nothing to build
No viable sendable mechanic. Listening remains fundamentally single-player; Blend and Jam already provide collaboration.
No viable sendable mechanic. Profiles already handle the household layer.
No viable sendable mechanic under the strict test. A shared decision room could improve the buying process, but the second person's participation is not necessary to complete the purchase.
Collaboration is not coordination.
17 / The honest ledger
These do not kill the architecture. They kill the inflated version of the thesis.
18 / The underdetermined register
Core thesis — underdeterminedDetail +Hide −
Detail +Hide −
Detail +Hide −
Core thesis — underdeterminedDetail +Hide −
Detail +Hide −
Detail +Hide −
Detail +Hide −
Detail +Hide −
Zero of the eight questions were resolved. Two are the causal core of the thesis — and both appear structurally unobservable from public sources.
19 / A decision framework
No → Keep it single-player.
Yes ↓
No → Classify Tier 2–4 before proceeding.
Yes ↓
No → Weak candidate.
Yes ↓
No → Recipient-friction risk.
Yes ↓
No → Likely referral or social decoration.
Yes ↓
No → Monetization may be weak.
Yes ↓
Unknown → Instrument it.
Yes → Measured yes — a possible compounding loop.
Build into the workflow
Test and measure
Keep it single-player
Methodology / evidence quality
Each of these circulates widely. Each failed a source check. They are recorded here so the exclusion is visible.
DesignRush (attributed to Statista); Hootsuite (attributed to past studies)
Appears to be RadiumOne's 2016 mobile-only figure with the year, source and device restriction stripped.
Saffron Edge (attributed to HubSpot)
No traceable original.
Medium (attributed to “TekRevol 2025”)
Aggregator citing aggregator.
Medium (attributed to “Pew Research 2025”)
No corresponding Pew publication located. Likely false attribution.
Voodoo acquisition release, widely repeated
Acquirer figure in an earnout-structured deal; trackers show 23m (Jan 2024) and 16m (end 2024).
Widely repeated in marketing commentary
True but incomplete: the share moved from 77% partly because paid spend was cut by about $1.08bn.
Conclusion
Sendability is real. The strict test separates products meaningfully, produces useful product-design guidance, and explains why a meeting invite is structurally different from a share button.
But the evidence stops before the most commercially important claim. The metric that would prove a true distribution loop is whether recipients later become senders of the same workflow. No company in the public evidence base publishes it.
So the useful conclusion is narrower: another person’s participation is a defensible product architecture. It guarantees exposure. It may create acquisition. Under the right conditions it may compound into distribution. But the public evidence does not establish that it produces superior economics.
The product question is not “How do we make this social?” It is “Is there another person who already has to do something for this job to complete?”
Known weaknesses
PastBehavior studies how similar problems have played out, separates durable patterns from attractive stories, and turns that evidence into product, distribution, and execution decisions.
Discuss a product problem