Collection rate on the deployed bucket, five months in.
+15% recovered. Nothing added to the cost.
Collections is won on timing and tone, and both collapse under human fatigue. The AI calls at the right moment, in the borrower's own language, hears hardship for what it is, and routes waivers and settlements to a person with authority. On the right, one morning of a real book being worked.
"'Salary late aayi hai' is hardship, not refusal. A platform that can't hear the difference collects less and burns more."The recovery doctrine · tone is a collections strategy
Collections is not a deflection problem. It is a resolution problem.
Most voice AI is sold on how many calls it can take off your team. That is the wrong number for a recovery book. The only figures that matter here are money recovered, promises kept, and roll rate held down.
So we build the motion around resolution, not containment. The AI works the account until it reaches an outcome, and the moment judgement is needed, a specialist closes it. Every conversation ends somewhere; nothing is deflected into silence.
One platform, every kind of debt.
Secured or unsecured, early bucket or written down, the motion adapts to how the money is actually recovered in that category. The same brain runs all of it, so what it learns on one book helps the next.
Minimum due, then the full arc
Cycle-timed nudges before the due date, minimum-due secured first, then a plan for the balance; disputes and chargebacks flagged and routed, never argued by a bot.
Holds: roll to next bucketSelf-cure in the early bucket
High-volume 0 to 30 DPD worked entirely by AI with payment links in-call; partial-today, balance-on-payday offers where the rules allow.
Moves: early-bucket cure rateCure before the asset is touched
Renewal and interest reminders that keep a pledged asset out of auction; escalation to a person the moment the conversation turns to release or repossession.
Moves: pre-auction cureThe reminder that beats the repo queue
Field-visit and repo actions are expensive and late. Timed voice outreach clears most arrears before a field agent is ever dispatched, and books the visit cleanly when it is warranted.
Moves: field-cost per recoveryThe long book, still covered
Low-frequency, high-value accounts stay in contact through the tenure; early signs of stress caught and handed to a relationship manager while there is still room to restructure.
Moves: early stress captureLow ticket, mass scale
The accounts that were never economic to call by hand, worked every cycle at a cost that makes the segment viable; arrears cleared alongside the next bill or renewal.
Moves: cost to serve the tailDay zero to the day it needs a person.
The book descends this ladder. The AI owns the top of it, in the accent colour. Where a call needs authority or a human ear, the rung shifts to the second colour and a specialist steps in, already briefed.
The nudge that never becomes a case
A warm reminder in the borrower's language with a one-tap payment link. Most accounts never go a day late, so they never enter the expensive part of the book.
The conversation, not the script
The AI hears hardship for what it is and changes the play: partial today, balance on payday, a date shift where the rules allow. A flow chart cannot do this; a conversation can.
The follow-up that lands on the day
Broken promises re-approached with the commitment on record: not accusatory, not amnesiac. The promise history travels with the account, so the next call starts where the last one ended.
A person, with authority
Waivers, settlements and disputes route to a recovery specialist mid-call, full history on screen, the offer pre-authorised. This is the handoff, and it is the product, not the fallback.
The accounts everyone stopped calling
Aged books worked at economics no agency can match; every account covered every cycle, and the true unrecoverables flagged with an evidence trail instead of quietly abandoned.
From 18–21% to 31%, in five months.
A large Indian lender's early-bucket book, run on the platform. These are the numbers the programme reported; they go named on this page the day the register and approvals land.
Borrowers worked every month, no added collector headcount.
Promises, objections and disputes captured correctly, so follow-ups land right.
Conversations completed, most running past the first objection.
Every rule that governs an Indian recovery call, enforced in the dialler.
The architecture is the point. The rule lives in the platform, not the training deck, so a non-compliant call cannot be placed and the trail your regulator wants already exists.
Conduct standards for recovery outreach and outsourced collections enforced in script and routing, with the full trail retrievable.
Disclosure requirements and conduct boundaries for digital-lending recovery built into every call flow.
Calling-hour windows, frequency caps and DND registry checks gate the dialler before a number is placed.
Consent status honoured per contact, data resident in India, recordings encrypted and access-controlled.
Every recovery conversation scored, AI and human, where the industry samples under 5%. Your audit answer is the whole book.
Payment links and card capture stay inside the standard; card numbers never linger in a transcript.
How fast does a pilot go live, and on what terms?
A production pilot on your real portfolio, not a sandbox. First live results by Day 7, a prove-it window agreed up front, and the baseline written down before the first call is dialled. If the number does not move, you walk.
What happens to hardship and dispute cases?
They always reach a person. The AI is trained to hear hardship as hardship, not refusal; disputes and distress route to your specialists mid-call with the full context attached. That routing is a platform rule, not an agent's judgement call.
Does this replace our dialler and CRM?
It can, but it does not have to start that way. The platform connects to Indian CBS, CRM and telephony stacks with pre-built connectors; dispositions and payments sync back to your system of record from day one.
Is our portfolio data used to train shared models?
No. Your data stays yours, resident in India, and what the platform learns on your book improves your book. Tenancy detail ships in the security pack.
How do you price this?
Consumption on handled volume, plus a success fee on uplift above your baseline. No seats, no shelfware; the full construct is on the outcome pricing page.
What proof exists on books like ours?
Ask for the reference call. The programme above reported a move from 18–21% to 31% in five months, and a separate financial-services book saw a 15% recovery uplift with the cost line held flat. Portfolio specifics beat averages; bring yours.
The numbers this motion is judged on.
Financial-services portfolio, with no added cost to collect.
India benchmark across collections and retention.
No account left unworked, and 100% of calls scored.