Written for Claims, operations and technology leaders, and the sponsors who have to fund a claims-document system around a process that is already running.
What's in this playbook
This playbook is the operational document for claims document processing. It is complementary to the Insurance AI Playbook already in this library, and is not a replacement for it. That playbook remains the industry map: six families, with document processing as one of them, and a first case that puts two existing claims pilots on an inventory before a live claims use. This playbook will not reprint that map, and will not give a different reading of the same firm. The Enterprise Workflow Automation Playbook remains the companion for the method; this playbook will not republish that method, and will not republish a worked NPV as a forecast.
What it is, instead, is a service playbook for the document case: six families of packet work, fourteen design tests, and a first-case path consistent with that industry playbook. Insurance document packets — notices, standard forms, estimates, photos — are a distinct buying problem; this playbook takes a named packet in depth. Fields that will later support a pay or a deny are proposals, and are re-checked. A named adjuster decides whether the file is complete. The model does not post a payment or a denial. A third claims pilot, an unsupervised pay or deny from a packet, an invented processing rate used as a baseline, or a twelve-week claims framework, is not a first case this playbook will recommend. This playbook will not invent a processing rate, a “cycle time,” or an “STP” sector figure, and will not close a named insurance rule.
How the tests and the method are meant to be used
The financial method is The Complete Guide to AI Automation ROI. This playbook will not invent a claims-document-sector return, and it will not promise a twelve-week go-live. The 3.8x average first-year ROI across client engagements is an average, not a forecast or a commitment for any individual project. ROI is measured across client engagements against a baseline agreed before work starts. It is an average, not a forecast for any document-processing project, and is not used as a sector benchmark. Built to SOC 2 standards is the exact wording we use; it is not a certification of the reader’s system.
The illustrative firm is the same hypothetical mid-size property and casualty insurer as in the Insurance AI Playbook, the Enterprise AI Governance Framework, and the 2026 diagnostic: overall 15, Emerging, Governance the constraint at 1, with a next-step target of 24, Developing. Two claims pilots have been run; neither is in production with an operator. The honest first work is to put those two existing pilots on an inventory, with an owner, a sampling plan and a stop, and only then, if the tests pass, fund a bounded document assist: extraction and completeness on a named packet, with a person on anything that opens, pays, or denies. An unsupervised pay or deny from a packet is not a first case. The target is a next level, not a timeline, and not a promise. A companion workbook, asymilink-claims-document-processing-playbook.xlsx, holds a family register, the fourteen tests, questions for counsel, a baseline sheet, and the same illustrative insurer. Do not wait to be Leading.
Six families of packet work
each with an honest first case, a case that is too wide, and a gate. Each item below is the shape of the map; the full wording, honest first cases and “too wide” warnings are in the PDF.
F1
Inventory of running claims systems (first)
F2
A bounded document assist — extraction and completeness on a named packet
F3
Photos, estimates and forms as evidence, not as a decision
F4
An exception queue, without an invented straight-through rate (later)
F5
A write into the claim file (later)
F6
Hosting, decided before the case is funded
The PDF gives, for each family, an honest first case, a case that is too wide, and a defined gate. The summaries above are the shape of the map, not a substitute for that wording.
Fourteen design tests (excerpt)
Fourteen design tests are applied to every case. The answers will differ. The standard does not. The full wording is in the PDF and the companion workbook.
T1–early
a named inventory; a named adjuster who re-checks any field that will be used to pay, deny, or reserve; counsel asked without a date taken from this document.
Mid-tests
a refusal to invent a processing rate or to recommend an unsupervised pay or deny from a packet as a first case; a trail and a stop; hosting decided before the case is funded.
Later tests
a signed baseline on the ROI method; a write into the claim file as a later design, not a first case.
A completed test sheet is not a certificate, not a compliance score, and not a finding that a system is safe or aligned with a named regime.
Get the guide and workbook
Enter your name, work email and company. The files appear on this page immediately. A backup download link is also emailed when email delivery is available.