Healthcare excess — GL and umbrella, with professional liability placed separately — turns on facilities and workforce: how many sites, what services, and a loss picture of premises and patient-handling claims. We ran a real 28-facility health system, Cedarcrest Health Network, through InsightXtract.

Insured: Cedarcrest Health NetworkRevenue $1.24B GL payroll $61.0MWC payroll $70.8M Facilities 28Named insureds 30 5-yr claims 16 / $4.2M incurredLead umbrella $25M

Why healthcare is an excess casualty risk

A health system's GL/umbrella tower is distinct from its med-mal program. The excess sits over premises liability across hospitals, clinics and surgery centers, plus a very large patient-handling workers-comp book. Facility count and mix, clinical payroll, and the premises loss picture drive it — and the coded record keeps the excess GL view cleanly separate from professional liability.

The submission packet

A real healthcare placement is rarely one clean file. This one is five, in four different shapes — the everyday reality an underwriting team has to re-key by hand today:

  • broker_email.pdf — the renewal narrative: account description, requested limits, exposure snapshot, and the excess-layer options to quote.
  • rfq.pdf — the umbrella/excess application: general information, coverage requested, and the exposure bases.
  • exposure_workbook.xlsx — the heart of the file: a 28-facility SOV (hospitals, clinics, surgery centers), GL exposure by facility class, clinical WC payroll by class, and an affiliated-entity schedule.
  • loss_run.pdf — five years of currently-valued losses, premises general liability plus a large clinical workers-comp book.
  • schedule_of_underlying.pdf — the primary casualty and lead umbrella the excess attaches over.

Step 1 — Classification

Every file is routed to a document type first — evidence-bearing, with a confidence, not a black-box label:

Document
Classified as
Conf.
broker_email.pdf
excess_casualty_submission_email
0.98
rfq.pdf
excess_casualty_application
0.97
exposure_workbook.xlsx
excess_casualty_exposure_workbook
0.99
loss_run.pdf
excess_casualty_loss_run
0.98
schedule_of_underlying.pdf
excess_casualty_schedule_of_underlying
0.96

Step 2 — Coverage structure, with citations

The requested tower is read from the email and application, every value grounded to its page:

CoverageRequestedSource
General Liability — each occurrence$1,000,000rfq p.2
General Aggregate$3,000,000rfq p.2
Products / Completed-Ops Aggregate$3,000,000rfq p.2
Commercial Auto — CSL$1,000,000rfq p.3
Employers Liability$1,000,000rfq p.3
Lead Umbrella$25Memail p.1

Step 3 — Exposure schedules, typed & normalized

The workbook is where template OCR falls over — many sheets, hundreds of rows, broker-specific wording. InsightXtract reads every sheet, maps each to a typed schedule, and normalizes the codes against governed glossaries. A slice of the GL exposure schedule:

StateWC CodeDescriptionRevenuePayroll
OH us_state_codes68706 ncciHospital - general & surgical$26,800,000$4,824,000
IN us_state_codes68707 ncciOutpatient clinic operation$12,400,000$2,232,000
KY us_state_codes62003 ncciMedical office building$8,100,000$1,458,000
… 44 GL rows · workers_compensation 58 · named_insured_mix 30 · location_list 28

Column bindings: state → us_state_codes, class → ncci_wc_class_codes / iso_gl_class_codes, occupancy → occupancy_types. Out-of-vocabulary values are flagged by validation, not silently kept — and fuzzy column resolution means a rule written for payroll still binds a broker’s “Total Payroll” column.

Step 4 — Derived exposures (deterministic, not guessed)

The account is rated on totals. Rather than ask a model to eyeball-sum hundreds of rows, the document type declares deterministic sum / count / group_by rules that reduce the extracted schedules exactly — the same numbers every run, each traceable to its source table:

$61.0M
GL payroll
∑ general_liability.payroll
$70.8M
WC payroll
∑ workers_compensation.payroll
28
Facilities
count(location_list)
30
Named insureds
count(named_insured_mix)

Step 5 — The loss run

For a health system's GL/umbrella, the loss run is premises injuries, visitor claims, security incidents and a heavy patient-handling WC book. Every claim is extracted — coverage line, status and cause normalized — and rolled up, with a by-line split for the loss pick:

Claim #CoverageCauseStatusIncurred
CED-3110GL coverage_linesPremises slip/fallClosed claim_statuses$96,000
CED-3088GL coverage_linesVisitor injuryOpen claim_statuses$142,000
CED-3051WC coverage_linesPatient-handling injuryOpen claim_statuses$188,000
… 16 claims across 5 policy years, currently valued
16
Claims (5 yr)
count(claims)
$4.2M
Total incurred
∑ claims.incurred
by line
Incurred split
group_by coverage_line

What the underwriter reads first

The coded record surfaces exactly the drivers that move this class of business:

  • Facility count & mix from the SOV — acute vs. outpatient drives the premises exposure.
  • Clinical WC payroll — the patient-handling book that dominates severity.
  • Excess GL kept separate from the professional-liability tower placed elsewhere.
  • Premises & security loss frequency by facility, from the coded loss run.

Why it holds up in production

  • Every value is cited — page/region provenance on fields and schedules; a file review can click any number to its source.
  • Codes are normalized against governed glossaries; out-of-vocabulary values are validated, not hidden.
  • Totals are computed, not guessed — deterministic derived fields over the extracted rows, exact and reproducible.
  • Configuration is versioned — document types, glossaries and rules pinned to a published version, so an output made today reproduces tomorrow.
  • Nothing is dropped — the all-sheets extractor surfaces every schedule, even ones the base schema didn’t anticipate.