The receipt is the review

Ask someone in Dhaka where to eat and watch what they do. They will not open an app. They will call a cousin, or ask in a group chat, or think of the place their colleague mentioned last week. The apps exist. Everyone has them installed. Nobody uses them for this.

It is easy to read that as a technology gap and go build a better interface. It is not a technology gap. It is that the ratings do not carry information, and everybody has quietly worked this out.

The arithmetic of a fake review

Consider what it costs to produce a five-star review on a mainstream platform. You need an email address. That is the entire cost.

Now consider what it is worth. A restaurant near Dhanmondi that opens next week and appears with a 4.8 from forty reviews will get walk-ins. The value of manufacturing those forty reviews is measured in customers per month. The cost is one afternoon and forty throwaway accounts.

When the value of faking a signal enormously exceeds the cost of faking it, the signal stops being a signal. That is not a moral observation about restaurant owners, it is arithmetic, and it produces the same result everywhere it holds. The reviews on those platforms are not untrustworthy because Bangladeshi users are dishonest. They are untrustworthy because the system asks nothing of anyone.

What Kemon asks for

To leave a review on Kemon you upload a receipt, a bill, or a service document, with the date of your visit.

That is it. That is the whole mechanism, and I have had a lot of conversations where someone waits for the clever part and there isn't one. The proof enters a moderation queue. Before a human sees it, the system runs an image hash against every proof already in the database, so a bill that has been screenshotted, re-uploaded, cropped slightly, or passed between accounts is caught automatically. Only once it clears does the review publish, carrying a Verified Visit or Verified Service badge.

Every account is also phone-OTP verified. There are no anonymous profiles, which removes the cheapest way to manufacture a reputation.

Recalculate the arithmetic. To fake forty reviews you now need forty phone numbers, forty distinct receipts that have never been seen by the system, and forty accounts old enough not to look like what they are. You have not made it impossible. You have made it cost more than it returns, which is the only kind of security that holds up.

The thing everyone objects to

Making reviewing harder reduces the number of reviews. Every growth instinct in product design says this is a catastrophe. Review volume is the metric; the funnel exists to raise it; a step that removes users is a bug.

I think that instinct is right for almost every product and wrong for this one. Kemon is not in the business of accumulating reviews. It is in the business of producing a number a stranger can act on. Ten verified reviews of a biryani place tell you more than four hundred unverified ones, because the four hundred tell you only that someone wanted you to see them.

There is a real cost and I am not going to pretend otherwise. Density will grow more slowly than a competitor's. The first year is harder. That is why the launch plan is narrow - Dhaka, food and restaurants, fully free - rather than national and monetised. You cannot build this kind of platform broad. You build one category dense enough to be genuinely useful and let it spread.

To make the cost worth paying, verification has to buy the reviewer something. Clearing the queue repeatedly builds a credibility score from account age and verification rate, with ranks, a leaderboard and rewards attached. The effort accrues to you rather than evaporating.

Businesses can answer. They cannot delete.

The third rule is the one that gets argued with in every conversation with a business owner.

An owner who claims their profile gets a lot: the WhatsApp number and exact map pin exposed, a verified badge, review management, analytics on views, calls, inquiries and direction requests, and the ability to request featured placement. They can reply publicly to any review. They can report one as suspicious and have moderators look at it.

They cannot remove it.

Not for a fee, not by escalation, not by claiming the profile. A review system where the reviewed party holds the delete button is not a review system, it is a brochure with a star rating on it. The moment that button exists, every rating on the platform means only that the business decided to leave it up - and the reader, correctly, discounts all of it back to zero.

This makes the product harder to sell. I have accepted that. The alternative is building the thing everyone already does not trust.

What I think this generalises to

Bangladesh is full of markets where the underlying transaction is fine and the information layer above it is broken. Tutors, contractors, mechanics, clinics, rentals. In each one people fall back to personal networks - asking a cousin - not because they prefer it but because it is the only channel where the speaker has something at stake.

Personal recommendation works precisely because reputation is on the line. The useful question for a platform is not how to collect more opinions. It is what we can make a stranger stake, cheaply enough that they will do it and expensively enough that it means something.

For a restaurant, it turns out the answer was the receipt they were going to throw away anyway.