In-house product · insurance distribution

The question PolicyDesk answers every morning: what do I need to do today?

India has over 25 lakh licensed insurance agents and distributors — the people who actually sell and service policies for millions of families. Almost none of them have software built for how they actually work. PolicyDesk is.

Insurance distribution Document AI In-house product
policydesk · intake trace In-house
  • 01RECEIVEpolicy document · PDF · scan · WhatsApp photo
  • 02READinsurer-aware extraction · the layout each insurer actually uses
  • 03LINKrenewal chain · matched by the identifiers on the page
  • 04MAPfamily structure · covered-members table, read once
  • 05FLAGhonest gap · “this figure isn’t available yet”
  • 06WRITEclient record updated · no manual entry
Built to fail loudly, not quietly. A gap shown honestly beats a guess shown confidently.
The system, in three runs

Watch what PolicyDesk actually does

Three traces from the same system, played in the order a distributor’s day produces them: the book is scanned for gaps, a WhatsApp enquiry is recognised, and a renewal is chained — each landing on today’s list with its reasoning attached.

policydesk · gap scan run 1/3
  • 01READentire book · every client · every category · every family
  • 02MATCHclient holds life + motor · no health policy
  • 03GATEchildren uncovered under the floater · a gap, not a mystery
  • 04SURFACEa conversation waiting to happen · reasoning attached
Cover that is missing. The hardest part of a book is noticing what a client does not have.
policydesk · enquiry match run 2/3
  • 01RECEIVEWhatsApp message · “can you also cover my son”
  • 02CLASSIFYgenuine business enquiry · recognised in the conversation
  • 03LINKmatched to a known client · their existing record
  • 04WRITEadded to today’s list · reasoning alongside
Where the business already happens. The enquiry is read where it arrives, not retyped into a CRM.
policydesk · renewal chain run 3/3
  • 01READpolicy renewed five times · one relationship, not five rows
  • 02LINKidentifiers on the documents · chain continued
  • 03GATEmotor cover lapsing in three weeks · due, for whom, why
  • 04SURFACErenewal on today’s list · money at risk shown
Renewals due, clients to call, money at risk. The morning list, with the reasoning shown beside each item.
The problem

Four tools, none of them built for this work

A family holding four policies across three insurers looks like four unrelated clients. A motor cover lapsing in three weeks and a health floater running out of family coverage sit in two separate places, and nothing connects them.

Insurer portals

A different login, layout and export format for every company represented.

Fragmented

WhatsApp

Where client conversations, policy documents and renewal reminders actually happen, with no structure and no memory.

Unstructured

Excel

The closest thing most distributors have to a client database, built and maintained by hand.

Manual

Paper and memory

For everything the first three do not cover.

Everything else
Every missed renewal is lost income for the distributor and lost protection for the family. Tools built for global insurance markets do not fit here — Indian insurers do not standardise on data formats, most policy documents still arrive as scans or phone photos, and the agents who need help most have the least time to learn new software.
What it does

AI does the data entry, so distributors don’t have to

A distributor uploads a policy document — a clean PDF, a scanned copy, or a photo taken on WhatsApp — and the details are extracted automatically, without a single field typed by hand.

Insurer-aware, not one-size-fits-all. Extraction is built to handle the variation between Indian insurers’ documents, not assume it away.
Finding the business that’s already there

Noticing what a client does not have

The hardest part of running an insurance book is not tracking what a client has — it is noticing what they don’t. PolicyDesk reads across a distributor’s entire book and surfaces exactly these moments.

Cover that is missing

A client with life and motor cover but no health policy is a conversation waiting to happen, not a mystery.

Benefits about to expire unused

Cover a family is already paying for, about to lapse without ever being used.

Renewal windows as sales conversations

A due date is also the moment to ask what else the family needs covered.

WhatsApp enquiries, connected

PolicyDesk reads the conversation where the business actually happens, recognises a genuine enquiry, and connects it back to what the distributor already knows about that client — then it joins today’s list, with the reasoning shown alongside it.

Why this is harder than it looks

Ambiguity that breaks naive automation

A lot of what PolicyDesk does sounds simple — read a PDF, remind someone of a date. In practice, Indian insurance data is full of exactly the kind of ambiguity that breaks automation built on hope.

Policy numbers are not unique across insurers

Identifying “the same client” needs real judgement, not a simple lookup.

Motor cover often splits across insurers

Own-damage and third-party cover can sit with different insurers and different renewal dates for the same vehicle — easy to model wrong, expensive to get wrong.

Contact details are rarely printed

Mobile numbers and email addresses almost never appear on Indian policy documents, because they were never meant to be extracted this way.

Same category, different structure

A certificate of insurance and a full policy schedule need to be read completely differently, even within one category.

None of this is solved by pointing a generic AI model at a PDF and hoping. It is solved by building extraction, matching and reasoning that actually understand Indian insurance documents — and by being honest about where the data genuinely is not there, rather than guessing and presenting a guess as fact.
PolicyDesk is built to fail loudly, not quietly.
The principle behind the product

A missing detail on screen — “this figure isn’t available yet,” “this distributor hasn’t confirmed this date” — is always preferable to inventing an answer and letting a distributor act on something that is not real. In an industry built on trust and long-term relationships, a wrong number shown confidently is worse than an honest gap. That discipline shapes every part of the product: how data is extracted, how gaps are surfaced, and how AI-generated leads are shown to a distributor — as information to act on, with reasoning attached, never as a decision already made for them.

Who it is for

Multi-product insurance distributors and small agencies in India — people managing real, growing books across life, health, motor and general insurance, often for the same families over many years. A distributor whose book has outgrown Excel, and whose time is worth more than manual data entry.

No charge · 60 minutes · the notes are yours either way