Salary lands after the mandate presents
Agentic collections
Agents that work your book
and know more every cycle.
Seven agents read, rank, write, call and learn — on your rules, inside the hours the RBI allows. Every contact teaches the next one, so the book gets easier to collect, not harder.
Your model choice. Your data, in your region. Every decision on a record you can read.
The compounding part
What the agents work out
Every contact is measured, so the plan that runs tomorrow is not the plan that ran yesterday. These are the four findings a book gives up first, and what each one changes.
One channel beats another in one band, at one hour
Promises are kept at a rate nobody assumed
Calls connect far better after a certain hour
The roster
Seven agents. One job each.
Each does a thing a person on a collections floor does and hands the result to the next. None can send anything: the governor decides that, and it is a kernel rule rather than a model, so no prompt can argue with it.
Reads a reply or a call and writes what it means. A stop, a dispute, a promise, a hardship — each one changes what happens next, in seconds, not on Monday.
Narrows the brief for one borrower: their language, their hour, their channel, what they said last time.
Turns a target into a plan per segment, and re-plans when the measurements move under it.
Makes the call, on the brief, inside the window. Quotes only what is on the account and records a promise only after the judge has checked it.
Writes what the call taught, each learning with the quote that supports it, into memory the next brief starts from.
Checks a promise against what is actually owed before it is accepted. An agent proposes; it never approves.
Allows or refuses every send, and writes the reason. A kernel rule, never a model, so it cannot be talked into a different answer.
Every settlement, waiver and notice is approved by someone you named, up to a limit you set.
Start in ten minutes →Onboarding
Four things, then it runs
Read-only against your loan system. It stamps at day-end and the agents rebuild the queue before anyone is awake.
Your carrier posts replies and receipts straight in, in its own format, signed its own way.
One real message to your own phone, and the provider's actual answer on screen. Wrong credentials fail in the room.
“Hold GNPA at 3.5%.” It tells you what cure rate that implies, and refuses a target your book has never produced.
On your floor
What changes on Monday
An account at 88 days with ₹40,000 outranks one at 12 days with ₹3 lakh, because only one is about to cost you a provision.
Outside the window, over the cap, a borrower who asked us to stop. A refusal costs a ledger row exactly like a send.
Read on the call, written to the account, every channel closed before the next message is queued.
A branch manager runs their branch. An officer proposes a settlement and cannot approve their own.
Pricing
Priced per account, not per seat
Every user you have, on every branch and product, for the same price. What you pay scales with the book you hand over, and nothing else.
Pilot
₹2,50,000once, for 60 days
One product line, up to 5,000 accounts.
- Measured against a holdout you choose
- Every agent, every channel
- Your book stays in your region
- Ends on the day it says it ends
Growth
₹35per account under management, a month
Minimum ₹1,50,000 a month. Carrier costs are yours, at cost.
- The whole book, every product
- Unlimited users, roles and branches
- Everything the agents learn stays yours
- Support inside four working hours
Enterprise
From ₹12,00,000a year
Several entities, co-lending, or your own cloud.
- Runs in your VPC if you need it
- Your model choice, or ours
- SSO, your retention, your DPA
- A named engineer
In exchange for one thing: when the numbers are yours, you let us quote them. Not a free pilot. A discount against a price, so both sides know what the thing is worth.
What the first month turns up
The findings are the product
Three things a book gives up almost immediately, once something is reading every reply and every refusal instead of a weekly export.
A borrower whose salary lands on the 30th and whose mandate presents on the 5th is not refusing to pay. The desk separates timing from intent on the first call, and the two need completely different work.
Every contact that did not happen carries the reason it did not: outside the window, over the cap, a borrower who asked to stop. That is how you show a rule held instead of asserting it.
Sixty to ninety days is where effort changes the reported number. Above it nothing short of the entire arrears upgrades an account; below it most of them cure on their own. The queue re-ranks on that every night.
First five lenders
We are looking for five design partners.
You bring a book and the people who work it. You get the whole platform at forty percent off the first year, a named engineer, and every measurement we take on your book — the roll rates, the cure rates, what each channel is actually worth — whether or not you stay.
Talk to us- One product line is enough to start. It does not have to be the whole book.
- Measured against a holdout you pick, so the lift is yours to verify and not ours to claim.
- In exchange: when the numbers are good, you let us quote them with your name on them.
Before you hand over a book
What your compliance team will ask
- Which model does it run on?
- Yours, or ours. The agents are model-agnostic and the governor is not a model at all — it is a kernel rule, so no prompt can talk it into allowing a contact.
- Can an agent contact outside 08:00 to 19:00?
- No. You can narrow the window; the kernel will not widen it. Every attempt outside it is refused and recorded as such.
- Can an agent settle, waive or issue a notice?
- Never. It proposes, and a person you named, up to a limit you set, decides.
- Where does what it learns live?
- In your tenant, in your region, with the quote behind each learning. It is yours, and it leaves with you.
Ten minutes from now, your book is ranked.
Sign up, connect the book, paste two URLs, send yourself a test. Email works today, which is why nothing waits on a regulator.