What yaadein is, right now

Three ways in, same engine behind them all.

What we saw, before we built anything

Both of Shefali's parents are in memory care with advanced dementia. She grew up on the story of how they met, heard at the dinner table so often she could recite the beats. She has her version of that story. She will never have theirs.

That looked like a problem about dementia, or about running out of time. It isn't. Shefali is here and can speak, and nobody has ever asked her what it was actually like to grow up as an Indian migrant in Sydney and start again in America. Not because anyone is careless. Because when someone does ask, they ask "so what was it like?", and that question has no way in.

Restating the problem that way changed three things. The people we could test with stopped being only the elderly and became everybody. The barrier stopped being time and became a skill. And that skill turned out to belong to people who ask for a living, who can do it but can't tell you how.

What we believed, and what would prove us wrong

Our hypothesis changed twice, and the change is part of the story (see what changed). Where it stands now:

What would prove it wrong

• The ladder's questions produce no more new disclosure than the Asker's own questions on the same topic.
• Answers don't lengthen as the rungs go deeper. If people talk just as long at the fact rung as the disclosure rung, the ladder isn't doing the work.
• People who ask genuinely good questions still get one-sentence answers. Then the barrier is the subject, not the question.

The baseline

In our first manual run, the fourth and hardest answer ran 1 minute 25 seconds in total across two unaided conversations. With framework-built questions on the same topics, it ran 4:16 (questions written by Claude) and 3:22 (an expert interviewer).

Wave 2 ran the app-guided conversation only. We did not capture an unaided comparison in Wave 2, so our baseline remains Wave 1's hand-timed result. When we tried running the unaided conversation a few times, participants got confused between the two conversations and recorded on their phone's voice memo app instead of through ours, so nothing comparable was captured. Running the unaided conversation through the app is our next step.

Our approach

We treated yaadein as an experiment with software attached, not software with a survey attached.

Eight steps, in order. The two marked in solid pink are where real people used it without us steering.

Six rules we held ourselves to

The ladder: four rungs, YAAD

Yaad is the root of yaadein, Hindi for memories. Each rung goes a little deeper, and costs the person answering a little more, than the one before. The first two are about the memory. The last two are about the person.

The app writes each next question from the last recorded answer. The Asker stays in charge: they can accept a question, reject it, or write their own. Off-limits topics can be set before the conversation starts.

What we ran

1. Sept 12

   Wave 1, by hand. Six recorded conversations across two threads, three rounds each: the Asker's own questions, Claude's, then an expert interviewer's. Nobody answering ever saw a screen.

2. Sept 13–20

   Redesign from the evidence. Expert round retired, two conversations on the same topic, nine measures defined with scales, surveys for both people after each conversation.

3. Sept 23

   Mentor working session. Live demo of the Mom Test app with our mentor as the respondent, and a steer to prioritize new disclosure, stay close to users, and not let the technology become the bottleneck.

4. Sept 23–Oct 4

   Wave 2, through the app. Askers and respondents recruited by the team and by referral, running the ladder through the Mom Test app, with every answer recorded, transcribed and timed.

How a session runs

Only two of the eight steps involve a model. Everything else is either a fixed rule or a person, and a person always decides what actually gets asked.

What we proved

"After the app closed, the conversations kept going."

Three waves, three months

How we got here, and where Wave 3 lands.

What we measure

Nine measures, each with a written definition, a scale, and a named way of scoring it.

Four come from short surveys both people answer after each conversation. Two of those, Novel Disclosure and Insight Shift, were then checked blind: two evaluators listened to every recording without seeing the survey answers, and their calls matched the participants' in all 27 conversations, the yeses and the nos. Five more are scored by an evaluator from the transcript or the recording, with a second evaluator scoring blind. Answer length, the share of answers over 60 seconds, supports them but isn't one of them: Wave 1 showed that a long answer and a meaningful one aren't the same thing.

Deferred until after this build: Self-Awareness, a validated dual-rater quiz we'll need for the future "asking yourself" memories.

One full ladder, start to finish

What the engine produces in practice. Four rungs, anonymised, consent cleared. The Asker is a daughter; the respondent is her father; the topic is their old house in Pune.

What we didn't expect: the ripple

We set out to help people ask better questions and keep the answers. Then people started telling us something we hadn't designed for and weren't measuring: the questions were changing their relationships.

Two things people told us afterward

Why this might be happening

Daniel Goleman describes emotional intelligence in four domains: self-awareness, self-management, social awareness and relationship management. Without planning to, the ladder seems to touch three of them. Answering a deep question makes you notice something about yourself. Asking one makes you notice something new about a person you thought you knew. And, sometimes, people act on it.

The signals were already in our data. 17 of 27 respondents wanted to ask a question back, and our focus group's sharpest complaint was that the first version "felt one-way." What people wanted wasn't a better archive. It was a two-way conversation.

How we'll find out if it's real

Two stories are not evidence of an effect, and we won't pretend they are. In Wave 3 we measure it with two questions. An hour after answering, both people are asked whether anything new came up for them. A week later, the Asker is asked: did you contact this person outside yaadein? If the ripple is real, that number will show it.

What we're building because of it

The ripple is why the next version adds Insights: a private view for the Asker of their year in voices (who they heard from, how many answers, how long they listened) and how long they've been talking with each person, with a no-penalty pause for illness or time away. An hour after answering, both people are asked privately whether anything new came up, so the Wave 3 measure is built into the product (see the screens). Of the 30-plus products we reviewed (see who else is in this space), none shows the people asking anything about the relationship itself. They show the archive. Insights is how we give the ripple somewhere to go.

Mapped onto the EQ framework above: Reflection and Keepsake build the self side (self-awareness and self-management); Together and Insights carry that into the social side. The output we're measuring is Goleman's bottom-right quadrant, Relationship Management, which for yaadein means increased connection with the friends and family you love.

Insights closes the loop. Tracking each user's own KPIs — people heard from, hours listened, memories made — encourages the next conversation, which deepens connection and adds another memory, which raises the numbers again. It becomes a flywheel in its own right (see how the main flywheel turns).

Why people don't ask

We assumed one barrier. The interviews found at least three, and better questions only fix the first.

What we measure next

Our hypothesis is about depth. The ripple is the discovery. Wave 3 measures both, and the ripple is the part we most want to know about.

What professionals said

Quotes from people who ask questions for a living — the kind who use yaadein's skill in their own workplace every day. Plus the two moments that most surprised us, and Maureen’s own voice.

Two moments we didn’t expect

In their own words

What was even more interesting was the experience didn't really end when we finished using the app. It went after that as well. Me and my wife ended up having a conversation about some of those memories for quite a time after that.Premal (Peter) Doshi, professional tester (see the evidence)

The demo

A short film the team shot to show yaadein in use.

Open the demo on Google Drive.

What Broke

Eleven things broke across both waves. Here they are in the order a session runs, with where each one stands.

Statuses as of September 29, 2026. *The point-of-view fix is in every prompt; Hansh is confirming the date it went live.

What changed, and why

Every change below came from evidence, not preference: what we believed going in, what we believe now, and what moved us.

How the app got here: the build log

From the first commit to the submitted version, Sept 21 to Oct 7

This traces yaadein from the first commit (Sept 21, 2026) to main as of Oct 7, 2026 (a5e5080). It covers the product logic, not the code: at each milestone it records what wasn't working, what was missing, or what we wanted to improve, and what changed as a result. The source is the commit history and the reasons written into each commit.

What the whole cycle shows

Milestone by milestone

Tap any milestone to open it.

How our app evolved

MVP 1 is a web-based app

Four versions in three weeks, each one shaped by what the last one got wrong.

Every version since September 23 is a web app: people open a link in their phone's browser, with nothing to download or install.

The interface

MVP 1 is a web-based app

For MVP 1, yaadein runs in the phone's browser. Every screen below is designed for a phone, and none needs a download.

For the next version of the interface — redesigned around the ripple, five pathways and 53 screens — see The Interface Beyond the Demo Day under the What's Next tab.

Golden rules and guardrails

A model writes the questions. These are the rules that keep it honest, kind, and on the right side of the conversation.

The guardrails run in three layers, from instant and cheap to careful and slower. Each one exists because we watched something go wrong.

Where each guardrail fires in the flow

One question moves through seven points. A guardrail is attached to the point where it can still change the outcome.

1. Session start. Language locked (English or Hindi, never auto-switched: G5–G8). Profile validated (V1–V5). Topic web-search runs only if explicitly enabled, with names masked before and after the query (K: W1–W17). Claude is the only provider; no silent fallback (M1–M4).
2. Model drafts the question. Prompt carries the shape rules (A: one question, one breath, plain language), direction (B: Asker speaks, no names in the question), depth (C: four rungs in order, L4 invites and never corners, no re-asking an earlier rung), and language script (G13 for Hindi: Devanagari only, never Urdu script).
3. Code post-checks the draft. Script check (G14), name check (D2), banned-phrase check (P1, 15 phrases), moderation screen by Claude Haiku (N1). A fail triggers a silent rewrite, up to 2 retries (P2). After the last retry: shown anyway and logged (P3), except moderation, which never ships flagged (N2 → 502).
4. Shown to the Asker. Only the question line is rendered; the rung and rationale lines are stripped (Q7). Owner-gated endpoints keep the drafts and raw replies off the respondent's device (R8).
5. Rejection loop. At most three options per round (R1). Asker can write their own only after three exist (R4), capped at 300 chars, moderation-screened (N3). Thread ends only when the respondent passes all three (R2). A replacement never drops a rung and never rephrases a declined draft (L6, L7).
6. Respondent answers. Audio must have bytes and speech (X11). Transcript is PII-masked in one pass before Redis, Claude, Supabase, or the Sheet see it: pattern layer for emails, IDs, 8+-digit runs, Aadhaar, spoken digit strings in English and Hindi (PII1); years, money, dates kept (PII2); birth days masked only in a born/जन्म window (PII3); addresses only when an address word is present, span-checked against the raw text (PII4–PII7).
7. Session rules across the whole thread. Junk answers (under 15 words or 8 seconds) pause the ladder. Consent beat before rung 4. Every question stamped with PROMPT_VERSION. Adults only. Sessions, profiles and audio expire after 24 hours; share tokens after 7 days (X6).

Example drafts the guardrails caught

Each pair shows what the model wrote first and what shipped after the rewrite loop. Taken from the rules in guardrails/guardrails.md.

Security

• Encrypted in storage and in transit. Answers are protected wherever they're kept and whenever they move.
• Not end-to-end sealed, and we say so. The model has to read each answer to write the next question, so answers can't be sealed the way a messaging app seals a chat.
• Personal details masked before they are stored or reach the model. Live since October 7.
• No participant data in our code. No keys, audio, transcripts or personal details anywhere in the repository.
• Consent before anything is shown. No quote, clip or voice leaves the app without the person's agreement.

Behind all of it sit the safety lines described below: adults only for now, off-limits topics, the right to decline, and stepping back from live-crisis topics.

Evaluators

Done · merged into the app on October 7

The guardrails check that a question follows the rules. They can't tell you whether it's a good question. So we wrote down what good looks like.

We built a golden set: worked examples that show, rung by rung, the question we want the engine to write and the kind of answer it should open up. The evaluators score the engine's real output against them. This is the step from "the question follows the rules" to "the question is good."

What each example holds

What gets scored

Every question the engine writes is scored on seven checks, and every simulated answer on its own shorter list. In plain terms, they ask five things:

• Right rung. Does the question sit at the depth it claims, no shallower and no deeper?
• Built on their words. Does it pick up something the person actually said? That is our golden rule, now measured.
• Sayable. Is it one question, short enough to read aloud to a parent?
• Safe. Does it stay clear of off-limits topics, stock phrases and anything that corners someone?
• Holds up if the answer is no. Does it still work when the person says "I don't remember," "no one" or "nothing"?

The examples cover light topics and heavy ones on purpose, so the set tests that a question about a college scooter stays light and one near a hard chapter stays gentle. The Hindi examples guard against a failure we saw in a real session, when a Hindi conversation slipped into another script.

One golden pair at each rung

A single arc through the four rungs, from the set. Priya is asking her father about their old house in Pune. Each rung reuses a handle from the answer before it.

The remaining examples, the angles we use at the deepest rung, and the full scoring recipe stay private.

Safety and privacy

These are deep conversations between people who love each other, which means they can also touch old hurt.

• Built: off-limits topics, set before the conversation begins, and the respondent's right to decline any question.
• Built: the Asker approves every question before it is asked.
• Built: personal details are masked in every answer before it is stored or reaches the model.
• Next: recognizing live-crisis topics and stepping back instead of going deeper.
• Adults only, for now. People asked about recording grandchildren. That is something we're considering down the line, and it will need its own safety design.
• Next: preventing use as a way to check up on someone who hasn't agreed.

The model has to read each answer to write the next question, so answers are protected in storage and in transit but are not end-to-end sealed. We say so plainly.

Head to head

Five rows that make the difference impossible to miss.

Where every other product automates the asking or the writing, yaadein keeps a person in both chairs and keeps the voice. That is the one row our flywheel is built on.

Does this need to be an app?

Our null test: the formal version of that question

The capstone asks every team to try the cheapest version of their idea first: a general assistant, with no special setup. If that's good enough, there's no product. So we took five real moments from Wave 2, with every name removed, gave each one to a free general assistant, asked what to ask next, and compared its question with the one yaadein wrote, against our own rules for a good question.

How we ran it: free ChatGPT in a temporary chat, a new chat for each moment, given the whole conversation so far with every name removed, and asked "What should I ask next? Give me just one question." yaadein's questions were scored against our own rules before ChatGPT's were collected.

What it tells us. ChatGPT scored 22 of 25 to yaadein’s 23, so a free assistant writes good single questions, and at moment 1 it did better than we did. Its clear slip was moment 2, where it ran long and skipped straight to the deepest rung. One limit of our own rules is worth saying: they don’t reward being specific, and ChatGPT’s first question was the broadest of all ten.

A general assistant can write a good single question. What it can't do is everything around that question, and that is why yaadein is an app. For MVP 1, that means a web app anyone can open from a link. Here is what it adds:

More than one case for why yaadein matters

Our own evidence comes later on this page. This section is the ground it stands on: what published research and public data already say. Every number links to its source.

At a glance. Each number is sourced in the cards below.

Nobody asks, and people want to be asked

Our starting claim was that stories disappear because nobody asks properly. A 2023 Ancestry survey of Americans found the same gap from both sides.

Industry survey. Source: Ancestry via Business Wire, April 2023

Surveys marked "industry survey" were run by companies that sell family-history products. We include them because they match the independent research, and we label them so you can weigh them.

Who else is in this space

Plenty of apps keep memories. Almost none of them change the conversation that makes one.

We sorted the products we found into five lanes. Each does one thing well, and in nearly all of them, either nobody asks anything or the app does the asking.

Where yaadein sits

Two questions separate everything on this page: who does the asking, and what you're left with afterward.

The seven closest, side by side

The closest product to yaadein is VoiceWeave: a weekly voice ritual, follow-ups that build on the last answer, and recordings you keep rather than a book. The one difference is the one that matters. Its AI asks on the family's behalf, so nobody in the family has the conversation. OverBiscuits is closest on who asks: a person sits with someone they love, but what they keep is a PDF.

The map shows where each product sits. The grid shows why. Who asks comes first because it's where trust starts: when the person answering already knows who is asking, they know who will hear them. That's the Discretion in our flywheel (see What's Next). It's our reasoning, not something we've measured yet. Every competitor cell was checked against the product's own site, app listing or published press in October 2026; "Not stated" means we couldn't find it, not that it's missing. Life Story AI's price comes from press coverage.

Why it lives on the phone

The grid shows yaadein gives up the printed book. That's a choice, and one of our professional testers put the reason better than we could.

As we age, even the child of the parent is tech savvy. That is a real plus. And their kids and their grandkids are going to be increasingly tech savvy.Mike Weaver, CEO of Dream Weaver Entertainment, professional tester

You won't be able to continuously, on the fly, in the moment, if you're at a coffee shop and you think of something, go back to a book, which might be sitting in your home or apartment. Being on the phone, it's something that's immediate … because the phone is where our lives live now.Mike Weaver

He isn't the only one. A participant asked for transcripts so he could read answers in a busy, noisy place where he couldn't listen. And about 9 in 10 Americans aged 50 to 64 already own a smartphone. One thing we kept on purpose: the person answering still downloads nothing. For them, yaadein is a link.

Everyone else we found

We kept looking, and found 24 more. Here is every one, sorted by who does the asking.

From each product's public pages, app listings and published reviews, October 2026. Some are very new or early versions, and we say so where we know.

Market signals

Care settings are attracting funding, and the biggest incumbent is moving toward voice: in June 2026 Storyworth added AI interviews and family calls. The corner built on AI avatars and afterlife voices is shifting rather than shrinking: HereAfter AI is reported to be winding down, while newer products such as Afterlife AI build talking personas of a living person.

Who it's for

We didn't pick that range by instinct. It's where well-studied forces in adult life meet: two tied to midlife, and one tied to the moment a parent starts to age.

Who we tested with

Two groups, two very different ages. The 20 people in our focus group, who saw the app and told us what they wanted from it, were every one of them 40 or older, and about half fell inside our 40-to-55 target. The people who actually used the app in Wave 2 did not: of the 30 who gave a birth year, 19 were in their twenties and none were 40 to 55, because we recruited through our own networks.

Two things follow. The findings held with a younger group than the one we're building for, and the next wave has to put the app itself in the hands of people aged 40 to 55.

The age split also showed up in conversation. People in their twenties told us they already have photos, stories and chat history. People in their thirties, forties and older were curious to try it.

Commitments and value

We haven't asked anyone for a letter of intent or a payment yet, and we'd rather say so than imply otherwise. Wave 2 was about one question: does the ladder work? Letters of intent are part of the next wave, once the version with vetted threads and one spoken memory is in people's hands.

What people would pay. We asked a small focus group, and the answers were too thin to price from, so we haven’t put a number here. Pricing comes after Wave 3, once more people have used the app. One option we’re exploring builds on our mentor’s advice to run sessions for people: yaadein as a service, where a trained guide asks the app’s questions with the respondent on the Asker’s behalf. An Asker who is short on time still gets the memory, and time stops being the reason it never happens.

In their words

If it comes with the voice of the person asking it, it would feel more personal.A respondent, on hearing questions as text

Good concept, but it felt one-way. If it were a two-way street, it would be better.A respondent, translated from Hindi

Ask a few follow-ups to understand what the Asker actually wants to ask. The app took it in a different direction.An Asker, paraphrased

What people asked for

Alongside the sessions, we held a focus group of 20 people, all aged 40 or older, at a home gathering organized for the purpose. Most had not used the app. We showed it, and asked what they would want from it.

Three ways in, and what the data says about each

yaadein has three ways in. Together LIVE is one person asking another, and it grows from one-on-one to a few people, then a group. Reflection LIVE is asking yourself. Keepsake DESIGNED starts from a photo or video and records why it matters. Here is where each one stands, judged only by evidence.

Proven means measured with real users and checked. Emerging means people are already doing it, unprompted or in its first days. Designed means planned from what we heard and saw, and not yet tested.

What's Next

A roadmap from what users told us

• The unaided conversation, inside the app. A step where the Asker records their own questions first, on the same topic, so every session carries its own baseline and nobody has to juggle two recording tools.
• The Asker's own voice reading each question, instead of text on a screen.
• Reciprocity built in: a turn for the respondent to ask back, on the same topic.
• Insights: a private year in voices, how long you've been talking with each person, and a one-tap check an hour after answering: did anything new come up? Added because of the ripple, and measured in Wave 3.
• Vetted threads to start from. An Asker who doesn't know what to ask picks a thread from a pack written to the framework. The first question comes straight from the pack. Each question after it starts from the pack's vetted question and is adapted to what the person just said, or written fresh when the answer goes somewhere the pack didn't expect. Logging which of the two happened tells us where the framework already has the right question, and where it runs out.
• A visible timer while recording, so people know how long they have.
• Better Hindi and mixed-language transcription.
• One spoken memory: each answer stitched into a single piece of audio, so it plays back as a story rather than an interview.
• Asking yourself, further: Reflection is live in a simple form. Next, it becomes a warm-up before you ask someone else, and a private place for journaling and leaving your own voice behind.
• The keepsake: upload a photo or video, and yaadein asks why it matters, so the reason survives with the memory.

The Interface Beyond the Demo Day

What we’re building after the capstone demo: a redesigned, research-driven interface across five pathways. Live, clickable, still a work in progress.

Where it's heading: the Pathways prototype

How the flywheel gets faster

Our mentor's advice was that software is no longer the moat. Data, distribution and design are, and each one feeds the next.

The flywheel is Jim Collins's idea from Good to Great: a heavy wheel that takes great effort to start, then builds momentum with every turn (Collins, Turning the Flywheel). Amazon made it famous after Collins visited in 2001 (Recode interview with Collins). Our mentor's version has three Ds: data, distribution and design. We added a fourth, and put one thing at the center.

Here is how we plan to turn it, three months at a time, from one conversation to a circle of them.

Conversation, then habit, then the ripple, then the circle. Connection is the destination, and each stage earns the data the next one is designed from.

Demand: the push that starts the wheel

Collins is plain that the first turns are the hardest: you push with great effort and the wheel barely moves (Collins). Marketing is that push. Distribution compounds what marketing seeds.

Each stage's push is in the Demand row of its column above.

What we measure: where each Asker came from, and what share of each channel completes a conversation.

Why Insights comes at months 3 to 6: the Hook Model

Nir Eyal's Hook Model describes how products become habits in four repeating steps: a trigger, an action, a reward and an investment (Eyal, Hooked, 2014). For yaadein:

• Trigger: a question arrives in a voice you know, or a date that matters comes round.
• Action: one tap to listen, and one to answer.
• Reward: you never know what someone will say. Insights shows what came back.
• Investment: every answer grows the playlist. In Eyal's model, investment is what raises the cost of leaving (Hook Model summary).

The model has critics, who say it can be used to build compulsion. We use it to build a habit of asking, with Discretion as the check: no streak pressure, and a pause that never counts against anyone.

Where the moat is: Helmer's 7 Powers

Hamilton Helmer's 7 Powers says an advantage only counts if it has both a benefit and a barrier that stops others copying it. He also finds that different powers become available at different stages: counter-positioning and cornered resources at the start, switching costs once a company takes off, and process power and branding only once it is stable (Helmer, 7 Powers, summarized). So here is where we stand, honestly.

What we're holding back, on purpose

Our designs already go further than this plan. These wait until the data says they're needed:

• Asking a group by QR code or link: months 9 to 12, once safety checks are in place.
• Bringing in a whole photo album: after month 6. One keepsake at a time first.
• Sharing a memory with a third person: months 6 to 9.
• Charging anyone: pricing focus groups in months 6 to 9, live pricing tests in months 9 to 12. Until then, yaadein is free (see Business Model).
• Your year in voices, and "a year ago today": months 9 to 12. They need a year of data.
• Hindi, all the way through: the interface has worked in Hindi since October 1. Next, Hindi and mixed-language transcription.

Some things don't wait. Stepping back kindly, choosing who can hear an answer, deleting and recovering, pausing, the 18+ check and downloading your memories are built from day one. Trust was one of the three barriers we found.

Business Model

Everyone who answers can ask. Answering and asking back are free forever. You pay only to start more conversations, and one payment covers the whole family.

In yaadein, Asker and respondent are roles, not customers. The person who answers today asks back tomorrow, then asks someone new. So we don't price a buyer and a recipient. We price the circle. Three things define the model: what's always free, what unlocks when you pay, and how one paying person carries the whole family.

How the market prices this today

The same eight products from The Market, compared on how they charge and what a family gets for it. Almost every one prices one storyteller: one person tells, someone else pays. Nobody prices two people who take turns asking.

What the grid tells us. The market sells two things: a book once a year (StoryWorth, Remento, Life Story AI) or an AI that asks every week (VoiceWeave, Lisova). Both charge per storyteller, because in both the storyteller only ever answers. Remento charges $99 for each extra storyteller, and StoryWorth's $199 plan is the only one that covers more than one. Every competitor cell was checked against the product's own pricing page, help center, app listing or published press in October 2026. "Not stated" means we couldn't find it, not that it's missing. Life Story AI's price comes from 2024 press coverage, and OverBiscuits' range comes from our Market check, because its own pages don't list a price.

Our pricing, and why

We price for the loop, not the product. The moment someone answers is the moment they could become an Asker, so nothing on that side ever costs money.

Answering needs nothing: no install, no account. Asking needs an account, because that's the moment you become a user in your own right.

Anyone in the circle can still ask people outside it. Answering is always free, and whoever answers can start a circle of their own.

Four choices we're making on purpose:

• Asking back is never behind a paywall. It's the moment a respondent becomes an Asker. Charging for it would cut the loop where it's strongest.
• We price the circle, not the seat. Every competitor charges per storyteller. In yaadein everyone takes a turn telling, so per-seat pricing would make a family decide who "deserves" to ask.
• No per-session cost. A flat rate removes the trade-off between "ask again" and "spend again."
• No ads, ever. The conversations are too intimate to monetize sideways. The only product is the subscription.

When we test it

Pricing is a hypothesis, not a decision. The free tier holds through Wave 3 so the flywheel has a chance to turn. Then we test in two steps.

• Variant A: $72 a year per circle (our baseline). The number we expect to land on.
• Variant B: $8 a month per circle, no yearly option. Tests a monthly commitment against a yearly anchor. We think yearly wins; this checks it.
• Variant C: $36 a year per person. Tests whether pricing the circle actually beats pricing the seat, or whether we're leaving money on the table.

If the ask-back rate or the ripple KPI drops more than 20% in paid groups compared with free, we pause the paid rollout and find out why.

What this makes

Because every person who answers can become an Asker, the people who might pay aren't only the Askers we find. They're everyone the loop creates.

• The loop: every Asker brings in about 3 people. In our test, 17 of 27 respondents wanted to ask back. If about 15% of the people brought in go on to ask someone new, 1,000 starting Askers become about 1,450 people who have asked by the second round, with no extra spend to find them.
• Revenue per circle: freemium consumer apps usually convert 2 to 5% of users to paid. At 4%, those 1,450 Askers start about 58 Family circles, or about $4,200 a year. At 8%, if the ripple does what we think it does, it's about 116 circles and about $8,350.
• Cost per circle: about $9 a year for each active Asker (Claude, Groq, Supabase, Vercel). A circle with 2 to 3 active Askers costs $18 to $27, a gross margin of roughly 63 to 75%. That's lower than per-person pricing would give, and it's the trade we're making for the loop. People who only answer cost a little too, for transcription.
• Cost to find a new Asker: close to zero when they arrive through the loop. Demand (our own marketing) only has to start the wheel.

Numbers are hypotheses. Wave 3 and the month 6 to 9 focus groups replace them with evidence.

Beyond Memories: where else the ladder earns its keep

We're building for families first. But strip the family away and what's left is a workflow that plenty of industries already pay for.

This is a vision, not part of our 12-month plan. For the next year, every turn of the flywheel goes to families. It answers one question an investor will ask: how big can this get?

Someone needs to ask well. Someone else knows something they have never put into words. Four parts of what we built carry over unchanged:

The same four rungs, in five other rooms

What changes is only the thread: who is asking, who is answering, and what the deepest rung is worth.

Which door first

An investor's question is where the most of today's build meets the most valuable conversation. Plotted, two doors stand out: capturing what retiring experts know, and customer research, the very method we used to test yaadein itself.

Four people, one question

Shefali Mody

Founder, vision and the YAAD framework

Executive coach and leadership development practitioner with 20+ years of asking people the questions they don't expect. Former management consultant.

Eddie Aditya Bhardwaj

Brand, UI and product design

Designed the yaadein brand and the full app experience in Figma.

Hansh Goyal

Backend, data architecture, market research

Built the data layer behind the Mom Test app and mapped the competitive landscape.

Munish Bhardwaj

Evidence, measures, evaluation and QA · Demo Video director

Designed the measurement system, built the Mom Test app, and leads blind scoring of the recordings. An amateur documentary director, he also directed our demo video.

What we're asking for

We built the instrument. Now we need Wave 3.

• Pilot families, 10 to 20, for Wave 3. Diaspora Indian households, ages 40 to 70, where one person is still asking and one person still has stories to tell.
• Advisors in dementia care, geriatric psychology, or oral history. People who have sat beside this problem for a long time, who can tell us what we are missing.
• Intros to end-of-life organizations. Hospices, memory-care centres, after-loss communities. We want to learn from them before we build for them.
• If you have lost someone and still have unasked questions, we want to talk. You are the reason this exists.