Nine agents do the gathering. The biller still signs.
A Fortune Global 500 insurance brokerage with $9.7B in annual revenue recovered 96 analyst hours every working day, without migrating a single system.
Global scale. A billing operation still running on people.
One of the world’s largest insurance brokerages. Its North America billing team processes a continuous stream of B2C billing activities every working day.
AI had been discussed internally for years. It had never actually run the operation.
Where the 300 hours a day were going.
Every billing activity took a biller about 18 minutes of pure manual work. At a thousand activities a day, the loss showed up in four places.
No single source of truth
A core AMS for the policy, a repository for documents, separate tools for classification, routing and banking verification.
Nineteen fields, typed by hand
Policy numbers, premiums, dates, commissions, payable entities – re-keyed out of PDFs and emails on every activity.
Critical details fell through
Routing rules and banking references lived outside the main flow. A missed instruction became rework or a compliance question later.
No audit trail
Little visibility into where an activity stood or who changed what. The answer lived in a biller’s memory.
What they tried first, and where it stopped.
The first attempt was the one most enterprises make: put an assistant in front of the biller and let it summarise the document on screen.
Copilots and chat assistants were never going to move the number. To take 18 minutes off an activity, four things had to happen together, on every activity, every day.
Pull from every system, not one
Reach across all five the way a biller does, and bring the answer back to one place.
Read documents built for people
Nineteen fields out of PDFs, binders and attachments never designed to be parsed.
Validate before a human looks
Routing, banking and premium checks run first, so people only see what needs a decision.
Keep the biller in charge
Nothing reaches a customer without a named human on the record.
Nine agents. One pipeline. One screen for the biller.
Nine agents in four stages. The biller sees one confidence-scored screen at the end.
Slotted alongside the stack already in place.
No migration. No replatforming. The agents run against the systems already in place.
What happens when the agent is not sure.
When the pipeline is unsure, it is unsure loudly – in front of a person who can fix it.
The biller used to gather. Now the biller just signs.
Before: a biller opens five systems, hunts for nineteen fields across PDFs and email, and hopes nothing was missed. After: nine agents do that work in the background, the biller opens one screen, reviews a confidence score, and signs.
What changed, by the numbers.
Per activity, before and after.
The same pipeline, pointed at the next queue.
Billing was the first workflow. Gather, read, validate, then hand the decision to a named human. That pattern generalizes across the operation.
Got a use case in mind?
Platform, engineers and governance all in. We’ll map your workflow against the same four tests this one had to pass.
