Level 1 taught how AI systems work and fail with insurance information, and which duties do not move when a tool arrives. Level 2 is where that becomes practice. It is written for licensed producers, account managers, customer service representatives, Medicare and benefits enrolment staff and agency claims support staff. Every module takes one task that fills an agency day and builds a supervised, repeatable, defensible way to do it, with the licensed reviewer's step shown explicitly and the licensing line marked.
The first three modules build the instrument. Module one sets the operating model: triaging a task on consequence, data class, verifiability and who is licensed to own it before any tool is opened; a seven-stage workflow from carrier documents to a deliverable a named producer has reviewed; questioning a vendor accuracy claim and designing a pilot that measures errors that matter; and keeping the account AI use record while the work happens. Module two builds the reusable agency prompt: a context header naming line, state, carrier, form edition and policy period, supplied documents instead of recall, constraints a reviewer can check, an explicit cannot-be-determined answer, and a design that treats client emails and attachments as data rather than instructions. Module three turns shared prompts into an agency library with licensed owners, versions, release tests, retest triggers and deliberate retirement.
Modules four to six are production work. Module four assembles the verified account file a draft may rely on, writes closed-file drafting instructions, runs a four-pass check for invented coverage, transposed figures and smoothed-over gaps, and sets where a first draft stops. Module five builds the field-level extraction checklist, works out what a small per-field error rate does to an application and a book of business, runs the three-way comparison between what was requested, quoted and issued, and handles off-checklist items and instruction-like text in documents. Module six covers what the account file must show about preparation, the retention duties that already apply however a document was produced, version discipline and the reviewer sign-off record.
Modules seven to nine are research and judgement. Module seven explains what grounding and linked citations do and do not guarantee, how a grounded answer still misstates a form or a rule, why a retrieval corpus is both an attack surface and a drift surface, and how to frame research questions so the limits show. Module eight is the five-step Form, Guideline and Rule Verification Protocol: existence, the quoted text, scope by state, line, entity and date, state action and variation from the model, and whether the provision is still in force. Module nine works recommendations: the consumer profile and the care obligation, the disclosure, conflict and documentation artefacts, what an insurer's electronic screening does not do for the producer, and which standard applies when several instruments appear to speak at once.
Modules ten to twelve turn to the client and the money. Module ten separates what the account file supports from what only the insurer can say, sorts the narrow cases where disclosing AI use is required from the many where it is a judgement call, tests public statements about the agency's own technology, and handles a request for the reasons behind an adverse underwriting decision. Module eleven works the licensing line at first contact: what unlicensed staff and intake tools may do, how to scope and supervise an assistant that answers strangers, lead handling and Medicare marketing, and what consent an AI voice call needs. Module twelve covers a fee charged alongside insurer compensation, compensation disclosure in annuity recommendations, the rebating analysis a free AI service must survive, and measuring savings honestly.
Modules thirteen and fourteen close the loop. Module thirteen classifies account material at the moment of use against every regime that can attach to it, works the health-information authorization and its listed insurance functions, tests an AI vendor against the service-provider conditions, and keeps the Medicare and Marketplace consent records. Module fourteen tests the use record against the questions a supervising producer, a carrier auditor or an examiner would ask, then works the first hour after an AI incident, the notice deadlines a producer can owe, and the practice review that follows.
Statements of authority are labelled throughout as law or rule, professional standard, professional guidance, best practice, emerging practice, or an AI Coalition Network recommendation, with whom each binds. Insurance is regulated by the states, so the course never presents an NAIC model, one state's enactment, a membership code or a voluntary framework as binding on everyone, and it never describes an instrument addressed to insurers as imposing duties on an agency. The level ships with a printable workbook and ten templates, from the prompt pattern library and the verification protocol to the data classification quick card, the incident first-hour card and the capstone portfolio template. The final examination draws forty scenario questions at random from a reviewed bank, and the capstone is an Account Workflow Portfolio of three AI-assisted workflows run on fictional accounts.
Everything here is professional education. It is not legal, regulatory or coverage advice, it does not replace producer licensing, carrier appointments or the insurance laws of any state, and learners must check the rules that apply to their own licence, states and carrier agency agreements. Completing the level earns an independent educational certificate issued by AI Coalition Network with a public verification page. It is not a licence, an appointment or a line of authority, it carries no continuing-education hours, and it satisfies no state, CMS or carrier training requirement.