Voice match with clone screening, sold to banks for phone and IVR payments
TL;DR Callerproof gives banks a voice factor for customers who struggle with PINs and OTPs, like blind and low-vision users, older people and anyone who can't read the screen. It checks that the caller's voice matches their enrolment and isn't synthetic. Banks pay per verified customer, and voice never works alone.
Digital payments in India assume you can see a screen and type quickly. Plenty of people can't. A 2026 review of 23 studies on visually impaired users found authentication was the most reported barrier, showing up in 18 of them, with OTP flows that force app-switching cited 13 times[1]. India supplied 7 of those studies. Blind users often hand their phone to a sighted helper to log in or pay, which gives away the PIN, and some go back to cash[1].
Feature-phone channels only help so much. Banks such as PNB let customers pay by calling an IVR number in their own language[2], but the transaction still ends with a PIN keyed on the handset, which is the exact step that trips people up.
What it does, in regulatory terms: an authentication technology service supplied to RBI-regulated issuers. Callerproof does not move money, hold funds or decide whether a payment goes through. The bank stays the issuer under RBI's 2025 authentication directions and chooses how to combine factors[3].
Here's how it works. A customer enrols by speaking short phrases at a branch or on a recorded bank call, with consent captured for voice data. On later IVR or phone-banking payments, Callerproof returns two scores to the bank: how closely the voice matches, and how likely it is to be cloned or replayed. The bank pairs that with something the customer has, such as the registered SIM, and can step up to a callback when scores are weak.
The seed pitched voice as the way to pay, and I've changed that. A survey of audio deepfakes notes that a commercial tool claims to copy a voice from about a minute of audio, and that detectors trained on one kind of fake miss others[4]. Voice is useful as a second factor with live clone screening. On its own it is a liability.
The end user is a bank customer who struggles with screens or keypads. The payer is the bank, usually through its digital-channels or accessibility budget. Pricing would be per enrolled customer per year plus a per-verification fee, both my assumptions, because I couldn't find a public price for bank voice biometrics in India. Payments banks and small finance banks with large rural books are the likeliest first buyers.
RBI's authentication directions, issued in September 2025 with compliance due by 1 April 2026, explicitly encourage new factors and let issuers add risk-based checks[3]. IVR payment rails already run through several banks[2], so the channel exists. Cloned-voice fraud has also made banks more careful about any voice check that lacks spoof screening[4].
There's no public figure for voice authentication spending by Indian banks. The firmest signal sits on the problem side - India accounts for 7 of 23 studies on payment barriers for blind users[1].
Adjacent, one US voice-security vendor says its platform is used by 14 of the top 20 banks it counts[5]. That's a US company's claim though, not your market.
US voice-security vendors sell voice authentication and synthetic-speech detection to contact centres, mostly in the US[5]. Global speech vendors offer speaker verification inside call-centre suites. In India I found no bank using voice as a payment factor, and no vendor tuning speaker match and clone screening for Indian languages, feature-phone audio and noisy rural lines. That's the gap.
| Difficulty | Hard (57/100) |
| Starting capital | ₹2.2–4 Cr |
| Time to launch | 12–18 months |
The models are hard to get right on 8 kHz feature-phone audio across many languages and accents, and clone screening needs constant retraining as synthesis tools change. Selling to banks means security audits, data held in India and a long procurement cycle before a single paid pilot.
You don't need your own payments licence, because the bank owns the payment. You do inherit the bank's rulebook.
Building this outside India? The categories above usually have a local equivalent, so check with your own regulator.
Never let the model approve a payment by itself. A new spoofing method can beat any detector, so the bank must keep a second factor every time.
Want to work out the AI layer for Callerproof? Talk to Yash →
Here's what can go south -
I think this is worth a shot for a team with real speech-model depth and someone who has sold into Indian bank technology teams. If you're a consumer-app founder then skip it, because you can't own the payment. Prove one thing first - in a pilot with one bank and an organisation of blind users, that the false rejection rate on real IVR calls stays under 5% while every cloned sample in a test set is caught.
Plan for roughly ₹2.2–4 Cr, mostly on speech-model work, security audits and a long first sale to a bank.
Not your own payments licence. The bank stays the issuer, but you must pass its IT outsourcing review, keep data in India and comply with the DPDP Act.
Yes, banks everywhere face cloned-voice fraud. Check your local banking authentication rules and biometric data law before a pilot.
The free Callerproof validation kit has the line-by-line capital breakdown, the effort scoring, the RBI authentication and outsourcing checklist, a 14-day validation sprint with pass/fail targets, a 90-day launch plan, 10 interview questions for banks and users, and 5 assumptions to test. Download the kit →
Thinking about building Callerproof or something close to it? The make-or-break work is getting one bank to run a pilot with blind and low-literacy customers on real phone lines, and that's the kind of thing I help founders with. If you need a hand, you know where to find me. Talk to Yash →
For illustration only. Names are fictional and ideas hypothetical; sourced facts are real but may be out of date. Nothing here is legal, financial or professional advice, and Yash Khatri is not personally liable for any use of it. Read the full disclaimer.