DirectoryWorkflowsCase studiesPodcastsWho's Noah Take the stack assessment
← All case studies
Insurance · 8 agencies

$318k a year back across eight insurance agencies

Quotes, policies, forms → read, check, file → straight into the AMS

Scope 8 independent US insurance agencies
Volume ~3,000 documents / month
Built Document processing · renewal tracking
Value ~$318,000 / year
Agencies worked with
Clifford & Bradford Insurance AgencyBailey Family InsurancePolicyWatchSterling InsuranceMarathon InsuranceNau InsuranceHoward InsuranceWasatch-Truck
$318k/yr value across 8 agencies
6,000 hrs staff time back per year
~270 policies saved per year

Eight agencies ran on four different systems. Every quote, policy, and form was read and typed in by hand. The typing took ten minutes each. The waiting took far longer. In a busy week a document could sit for over a week, and by then the prospect had bought elsewhere. Now around 3,000 documents a month are read, checked, and filed automatically. That is 6,000 staff hours a year back, plus about 270 policies that used to go cold.

What it's worth

$318,000 per year across the eight agencies. About $40,000 each.

Staff time back

Documents per month
3,000
Hand-keying time per document
10 min
Hours back per month
500 hrs
Hours back per year
6,000 hrs
Producer / CSR cost
$35 / hr
per year
$210,000

Commission saved

New-business quotes per month
750
Share lost to slow replies
10%
Quotes saved per month
75
How many of those bind
30%
Extra policies per year
~270
Commission per policy
$400 / yr
per year
$108,000
Total per year $318,000
How this was calculated — every assumption
  • Volumes come from the client: about 3,000 documents a month across the eight agencies, at roughly 10 minutes of hand-keying each.
  • Producer and CSR time is costed at $35/hour. That is salary plus tax, benefits, and overhead.
  • About a quarter of documents are new-business quotes. The rest only count toward staff time.
  • The 10% covers quotes lost because the reply came too late, not quotes lost on price. Only 30% of those saved quotes are counted as binding.
  • Renewal flagging is not in this total. It clearly saves policies, but we have no measured figure for it yet, so it is left out rather than guessed at.
  • These are models built from reported volumes, not audited accounts.

Results at a glance

The problem

The client runs eight independent agencies across four systems: AMS360, Applied Epic, EZLynx, and Dynamics. Producers read every incoming quote, endorsement, policy, and form, then typed the data into whichever system that agency used. Documents arrived as scanned PDFs, phone photos, and hand-filled forms. No two carriers used the same layout. The ten minutes of typing was the visible cost. The wait in front of it was the expensive one. How long a document sat depended on how busy the week was: minutes when things were quiet, over a week at peak. A prospect who waits a week has already bought somewhere else.

What was built

01

Read any format

Emailed PDFs, scans, phone photos. The pipeline pulls the fields out with AI, plus OCR for photos and handwriting. Nothing waits for a person to open it.

02

Check every field

Each result is checked against a strict schema. If it fails, it goes to a human review queue. Nothing half-formed reaches a policy system.

03

File it in the right system

Clean data is mapped into the shape each agency's system expects. APIs where they exist. Browser automation where they don't.

04

Flag renewals early

The same pipeline watches policy dates and flags renewals before they come due. Nobody finds out a policy lapsed after the fact.

05

Handle the next step

Routing, notifications, moving a document forward. All the follow-on work producers used to do by hand.

06

Log everything

Every run is logged. A test set of real documents runs against every change, so accuracy never slips quietly.

What keeps it from breaking

Nothing unchecked reaches your AMS

Validation is a hard stop, not a warning. Failed records go to a person, not into your policy system.

Accuracy is tested, not assumed

Real documents with known-correct answers run against every change. Quiet accuracy drift is the failure that would poison a book of business.

APIs first, browser automation second

Browser automation is more fragile. It only covers systems with no API, and it is monitored on its own.

How the approach worked

Keep it simple. Good prompting and clean API calls before anything custom. The hard part was never one model call. It was staying reliable across eight agencies, four systems, and documents that never look the same twice. And the value was never just the keying time. A ten-minute job that waits three days is a three-day job to the person waiting.

The results

About 3,000 documents a month now go through the pipelines instead of through a producer's hands. That is roughly 6,000 staff hours a year, worth about $210,000. Faster turnaround is worth more than it looks: around 270 policies a year that used to go cold now bind, worth about $108,000 in commission. Renewals get flagged before they lapse, and each agency keeps using the systems it already paid for.

Before

Ten minutes of typing per document, behind a queue that ran from minutes to over a week. Eight agencies, four systems, all by hand.

After

Documents handled on arrival, checked, and filed automatically. Same turnaround whether the week is quiet or busy.

Tools & technologies

n8nPlaywrightLLM extractionOCRZod validationAMS360Applied EpicEZLynxDynamics CRM

Have a process like this slowing your business down? Tell me your stack and goals and I'll send back a build plan.

Take the stack assessment →