Case study · Title insurance AI reads the deed.
AI reads the deed.
You verify the data.
A regional title company eliminated its manual transcription step — every email, PDF, and spreadsheet now pre-parsed into 25 structured fields before an employee opens the order.
38% prior title found rate
<2min email to review queue
350 avg daily volume
100% human-reviewed
The problem Manual transcription before
Manual transcription before
any value-producing work.
Every title search request arrived in a different format — informal emails, scanned photocopies, digital PDFs, spreadsheets. Before any search work could begin, an employee had to read the full document, identify the correct property address, find the current owner in dense deed language, decipher recording references, and manually enter 25 fields.
This was pure transcription overhead — 10 to 15 minutes per order, every order, and a source of downstream errors when the wrong address or owner was entered.
Pain points
Mailing vs. property address
Deeds list multiple addresses. Entering the wrong one caused errors throughout the order.
Grantee vs. grantor
Dense conveyance language requires reading comprehension, not pattern matching.
Non-standard recording refs
"DB 1565/168", "Book 1565 Page 168", and "O.R. 1565/168" all mean the same thing.
Scanned documents
Old photocopies produce unreliable OCR output. Staff printed deeds to read them manually.
Multi-property requests
One email listing five properties meant creating five forms by hand.
The solution An intelligent reader
An intelligent reader
for every intake format.
Cardinal built an AI parser that accepts the raw request — email, PDFs, spreadsheets — and routes pre-populated records to a human review queue. The employee verifies. They no longer transcribe.
01
Ingest
Requests arrive via Postmark inbound webhook or web upload. Email body and attachments are processed together — PDFs, scanned images, and spreadsheets handled in a single pass.
02
Extract
Claude reads the full request as an intelligent reader: disambiguating property vs. mailing addresses, identifying the current owner from deed conveyance language, normalizing recording references, and reading scanned pages via vision — no OCR required.
03
Review
Pre-populated records enter a queue showing a completeness score, source badges per field (email body vs. attachment), and inline editing. Reviewers claim records, verify fields, and mark complete.
38%
prior title found rate post-deployment — up from 32% before launch
Prior title found rate · %
BeforeAfter
Every order starts with verified data.
The employee's job shifted from transcription to verification. Intake data quality improved measurably — prior title found rate rose from 32% to 38% post-deployment, driven by more consistent property address and owner extraction.
- 2.4 minutes of time saved per request, on average
- Multi-property emails auto-generate one record per property — no manual duplication
- Scanned deed images read directly via AI vision — no printing, no OCR failures
- Every field tagged with its source document for instant reviewer verification
- Zero records reach the order system without human review and sign-off
Design principle Human-in-the-loop
Human-in-the-loop
by design.
This system pre-populates, it does not automate. Every record passes through a human verification step before any data reaches the order management system.
AI pre-populates, humans approve
No record is acted on until a staff member has reviewed and confirmed every field. The blank form is gone — the verify step remains.
Field provenance
Every parsed value is tagged with its source — email body, attachment, or both — so reviewers can check Claude's work against the original in one click.
Completeness score
Each record shows the share of fields auto-populated vs. needing manual entry, so reviewers can triage at a glance and focus attention where it matters.
Full audit trail
Original email text and attachment filenames are preserved alongside every record. The input that produced any extraction is always one click away.
Work with us A workflow like this.
A workflow like this.
Your industry.
A 90-minute discovery call with a senior partner. We trace one of your workflows, identify the model-tractable steps, and write a one-page baseline you can take to procurement. No charge.
Book a discovery call