- Data residency options
- Role-based access & PII masking
- Immutable conversation archive
- Maker-checker on outbound
- Consent & opt-out capture
- The institution owns the number
Acquisition & Onboarding
From enquiry to first EMI without a dropped lead
An AI agent picks up every loan, card and policy enquiry the moment it lands — qualifies it conversationally, collects the paperwork, and hands your RM a warm, complete file instead of a phone number.
-
Qualified before a human picks up
Income band, employment type, city, existing obligations and product fit captured in the thread. Ineligible applicants get a clear answer in a minute rather than a place in a callback queue.
-
Documents collected in-thread
PAN, address proof, selfie and income proof requested one at a time. Blurry, cropped or expired uploads are detected and re-asked automatically, so the file that reaches credit ops is already complete.
-
Pre-approved offers that actually land
Broadcast personal-loan top-ups, limit enhancements and card upgrades to the exact eligible cohort, with an accept-now link. Voice campaigns pick up the numbers that never open a message.
-
Nobody drops off in silence
Insurance quote comparison, SIP mandates and demat account opening all resume where the customer stopped. Abandoned journeys get a timed nudge sequence instead of a cold call three weeks later.
-
Branch visit or RM callback, booked
The agent offers real slots, confirms the branch, and writes the appointment to the RM’s calendar — with the qualification summary already attached.
Are you salaried or self-employed?10:02 am
Shall I collect your documents now? Takes about 3 minutes.10:04 am
- ProductPersonal loan · ₹5,00,000
- QualificationSalaried · ₹1.1L/mo · Pune
- Documents4 of 5 received
- ConsentCaptured 10:02 am
Servicing & Deflection
The requests that fill your call centre, answered in seconds
Customers authenticate once inside the thread, then self-serve the high-volume, low-complexity requests that make up most of your inbound queue — with clean escalation to a human the moment a case actually needs one.
Accounts & cards
Authenticated balance, mini-statement and last-ten transactions, plus the card controls that drive the most repeat calls.
Loan servicing
Every document a borrower calls in for, generated and delivered in the thread without a ticket.
Policy servicing
Renewal and revival windows are where insurance books leak. The agent works them conversationally, in the customer’s language.
Claims & disputes
Intimation, evidence upload and status tracking in one thread, so customers stop calling to ask where their claim is.
A replacement can reach you in 3–5 working days at your registered address.08:42 pm
Deflect the call before it is answered
- IVR offers “continue on WhatsApp” before the queue
- Live calls hand off to chat with the context carried across
- AI voice agents answer overflow in Hindi, Tamil, Marathi and more
- Anything sensitive routes to a human with the full thread attached
Collections
A recovery cadence that scales past your calling capacity
Pre-due nudges, due-date reminders and overdue escalation run across WhatsApp and AI voice at portfolio scale. Your tele-callers stop dialling numbers that would have paid anyway, and spend their day on the accounts where a conversation changes the outcome.
- D−3 · Pre-due
Amount, due date and a pay-now link, delivered where the customer already reads. The cheapest recovery you will ever make.
- D0 · Due date
WhatsApp first. Numbers that stay unread roll into an AI voice call in the borrower’s own language the same evening.
- D+1 to D+30 · Overdue
A distinct cadence per DPD bucket. The voice agent captures promise-to-pay date and amount and writes it straight back to your LMS — no call-disposition form to fill in.
- Human threshold
Settlement, restructure, hardship, dispute or any sign of distress routes to a trained agent with the whole history and an AI summary already loaded.
Every message and call is templated, approved and recorded — so recovery activity stays inside the conduct rules your compliance team has signed off.
Collections — September cycle
RunningRetention, Cross-sell & Wealth
Keep the relationship on the institution’s number, not the RM’s phone
The renewals, maturities and portfolio moments that decide whether a customer stays are the ones most likely to be missed. Put them on a channel your customers read, and keep the whole history where the next RM can pick it up.
Renewals, maturities and revivals
Policy renewal windows, FD maturities and lapse-revival deadlines each get their own sequence — reminder, options, and a link that completes the action without a branch visit.
Portfolio and market nudges
NAV moves, failed SIP mandates, rebalancing prompts and market alerts for wealth and HNI books — personalised per holding, delivered as chat rather than an unopened email.
One branded number, every RM behind it
Customers message the institution, not an individual. Every RM works the same Unified Inbox, so conversations, commitments and documents stay with the bank when people move on.
Handovers the customer never feels
When the relationship manager changes, the new one opens the same thread with an AI summary of every prior conversation and call — no re-introduction, no repeated questions.
Governance, Compliance & Risk
The part that decides whether risk signs off
Customer conversations in BFSI are regulated records. Everything below exists so that scaling them does not scale your exposure — who can see what, who approved what went out, and what was actually said on every single conversation.
The institution owns the conversation
Numbers, threads, documents and customer data live in the platform — never on a relationship manager’s personal handset. When an employee resigns, the book does not walk out with their phone.
Role-based access and PII masking
Agents see only what their role permits. Account numbers, PAN and card digits can be masked in the console, in search and in exports, with every unmasking event recorded against a named user.
Maker-checker on everything outbound
Templates, campaigns and agent instructions pass a named approver before a single message is sent. Versions and approvals are retained, so you can always show what was live on a given date.
An archive built for examination
Every message, call recording, transcript, agent action and consent event is retained and exportable — the evidence pack a regulatory examination or a complaint adjudication actually asks for.
Consent, opt-out and data-principal rights
Consent is captured and timestamped at the point of collection. Opt-outs propagate across every channel and campaign immediately, and erasure and access requests are servable from one place.
Themes surfaced before they escalate
Conversational analytics clusters what customers are actually complaining about across the whole book — a mis-priced fee, a broken app flow, a confusing renewal notice — while it is still a trend and not an escalation.
Every conversation scored, not a 2% sample
Manual QA listens to a fraction of a percent of calls and almost none of the chats. CX Governance reads all of them — flagging mis-selling language, guaranteed-return claims, unapproved product comparisons and missing disclosures on the day they happen, with the exact line quoted.
Conduct surveillance All chat and voice conversations, last 7 days
Scoring 100%What changes
Where the operating leverage shows up
- 60–70%of tier-one service requests answered without reaching a call-centre agent
- 3–5×more leads qualified per relationship manager per working day
- 100%of chats and calls scored for conduct, against a typical 2% manual QA sample
- 24–48 hrstaken out of document-to-disbursal turnaround by in-thread collection
Illustrative ranges modelled on common BFSI deployment patterns for this class of workflow. They are not measured results from a named customer, and your own numbers will depend on product mix, channel mix and existing automation.
Questions from risk, IT and compliance
The objections we get asked first
Short answers here; your security and compliance teams get the full documentation pack during evaluation.
Where does our data physically sit?
Data residency options are available, including in-region hosting, so customer data and conversation records stay within the jurisdiction your regulator expects. The specific region and retention configuration is set per deployment and documented for your audit file.
Can we deploy inside our own VPC or a dedicated environment?
Dedicated and customer-VPC deployments are supported for regulated workloads, alongside the standard multi-tenant cloud. Which model fits depends on your data-classification policy and integration surface — that is a scoping conversation, and we will tell you plainly if a requirement is out of scope rather than promise around it.
What exactly is in the audit trail?
Every inbound and outbound message with delivery and read state; every call recording with its transcript; agent actions including assignment, escalation, note and unmasking events; template and campaign approvals with the named approver and version; and consent, opt-out and preference changes. All timestamped, immutable once written, and exportable for an examination or a complaint adjudication.
Will WhatsApp approve templates for financial content?
Outbound templates go through Meta’s approval process, and the platform manages submission, versioning and rejection reasons in one place. Financial content has real constraints — no guaranteed-return claims, careful handling of account data, clear opt-out language — so the maker-checker step runs before submission. In practice that means fewer rejections and a documented approval chain your compliance team can point at.
How does this integrate with core banking, LOS, LMS and our CRM?
Three ways, usually combined. The Embedded SDK drops the unified inbox directly into your CRM so RMs never leave it. APIs and webhooks move events both directions with your LOS and LMS — application stage, disbursal, DPD bucket, promise-to-pay. And AI agents call your systems as tools at conversation time, so an eligibility answer, an EMI schedule or a claim status is read live from the system of record rather than from a stale copy.
What happens to our existing contact centre?
It gets smaller queues and harder problems. IVR deflection moves routine requests to WhatsApp before the caller waits; AI voice agents absorb overflow and after-hours volume; and the agents who remain work inside the Unified Inbox with AI Copilot drafting replies, summarising history and flagging chats that are still waiting on a response.
Which languages do the agents handle?
Chat and voice agents operate across major Indian languages, and a single agent definition serves all of them — you write the policy and product knowledge once rather than maintaining a script per language. Conduct scoring runs on the transcript in-language, so a mis-selling flag in Tamil surfaces the same way one in English does.