Reva. The AI astrologer that remembers you.

The problem

Astrology apps are everywhere.
None of them remember you.

• India has 400M+ astrology users. AstroSage, Kundli, dozens of WhatsApp groups. All free.
• The brief: build a paid astrology app people return to daily. From the first day we built with AI; even the product strategy pitch was made with it.
• The obvious move was personalising the horoscope and charging for it. Everyone could see it. Nobody had made it work.
• A horoscope is a one-time read. You finish it, you leave. Nothing brings you back.

The insight

Memory is the product.
The horoscope is just the door.

Replika retains 25% of free users as paying subscribers. Not through content. Through memory. Duolingo users with a streak are 3.6× more likely to stay. People pay for relationships, not content. No astrology app had built one. We made Reva the reason to return. The daily horoscope became the nudge that opens the app, not the whole point.

Around that core we keep launching features built for return visits: daily predictions, devotional music, chanting mantras, daily rituals, and affirmations. Each one gives a different kind of person a different reason to open Reva tomorrow.

The design

Never break the conversation. Not even for a form.

An astrology app needs a lot from you before it can say anything personal: name, birth date, birth time, birth place. Every one of those asks is a step where people quit. Onboarding churn is the tax every product pays for data. My call was to not pay it the usual way: no form at all. Every question happens inside the chat, so giving your details feels like talking to an astrologer, not filling a government form.

• Every onboarding question (birth date, birth time, name) happens inside the chat. No forms, no dropdowns.
• When users don't know their exact birth time, Reva adapts rather than blocking them.
• The paywall appears after the first meaningful exchange. You've felt the value before you're asked to pay.

The research

Trust is the metric no one tracks.

The retention chart looked healthy. The daily prediction had a strong habit around it: most people open it at a fixed moment, early morning before the day starts, or right at midnight when the next day's prediction unlocks. On the chart they all look the same, a user who came back. So we picked up the phone and interviewed them.

• One man opens Reva every morning. He rated his trust in it one out of ten. The predictions don't match his life, so he reads it and closes it.
• One woman rated her trust close to ten. She said she can tell Reva things she can't say out loud at home. On the chart, she and the first man are the same row.
• A third user trusted the predictions and deleted the app anyway. She wanted to know what was coming next year, not just today. Trust didn't keep her either.

What we learned

High engagement does not mean high trust.

The calls told us what moves trust. It goes up when a prediction feels written for one person, from their full birth chart, not their star sign. It goes down when the lucky colour is the same for everyone. It goes down when someone never knew their real birth date, entered a rough one, and then read the mismatch as the app being wrong.

The craft finding surprised us. Not one person complained about how Reva looks, and several said the look is why they picked Reva over other astrology apps. That's worth knowing, and it's also a warning. Craft is why they chose it. It isn't why they stay.

The general lesson goes past Reva. Retention is measured every day. Trust was measured once, by phone. A dashboard counts the one-out-of-ten user and the ten-out-of-ten user as the same win, and one of them is churn that hasn't happened yet. No dashboard we had could tell them apart. A ten-minute phone call could.

My role, and what's next

The call that made the numbers possible.

I pushed for conversation-first design before a single screen existed. That shaped the onboarding, the paywall position, and how the product asks for data. Onboarding is one retention loop, not the whole story. Chat memory is the bet that could become the core one, and the research chapter above is how we're testing that honestly instead of assuming it. Reva went from concept to ₹25L daily GMV with a small team and a short runway.

And the research set the next brief. The next thing to build isn't a feature. It's a question, inside the app, right after the prediction: was this accurate? Ask it every day, and trust stops being the metric no one tracks.