Lesson objective
Choose one Monday behaviour for a session and trace it to the rule, standard or programme position that supports it, naming the jurisdiction, the body and whom it reaches.
Plain-language explanation
Why this matters now
Backwards design means deciding the last thing first. Before you write a slide, you settle what one person will do differently on Monday morning, at a named moment, with a named kind of file. Everything else in the session then has to earn its place by serving that.
The hard part in insurance is not the method. It is the sourcing. A behaviour you teach as required has to be traceable to something that actually requires it, and in this field the instruments that mention AI are mostly addressed to insurers, while the instruments that reach producers mostly do not mention AI. An instructor who blurs that teaches a licensed room something they may repeat to a client.
No source in this register validates backwards design, or any session structure. The method below is instructional practice, presented as best practice. It is not a finding about how adults learn, and nothing in this programme supports such a finding.
Here is the sourcing discipline. For every candidate behaviour, write three things: the instrument, the jurisdiction and body, and whom it reaches. Then mark whether applying it to an AI tool is an inference you are drawing.
| Candidate Monday behaviour | The instrument behind it | Who it reaches | Inference? |
|---|---|---|---|
| Route any coverage, limit or deductible question to a licensed colleague rather than letting a drafting tool answer it | New York OGC Opinion No. 04-12-03 (December 2004), interpreting N.Y. Ins. Law s.2101 as amended by Chapter 687 of the Laws of 2003 | New York licensees and their unlicensed employees; an interpretive opinion, not a regulation | Yes: the opinion predates AI. Applying its delegation reasoning to a tool is an inference |
| Write the basis of an annuity recommendation in the producer's own words before the application is submitted | State rules based on NAIC Model #275, section 6A(4), the documentation obligation | Producers and insurers, as adopted in each state | No for the record itself; yes for any statement about tools, since #275 does not mention AI |
| Classify the information in a document before it goes anywhere near a tool | State regulations based on NAIC Model #672, sections 18A and 19 for health information | Licensees of that state's insurance department | Yes: the model does not mention AI |
| Check every coverage statement in an AI-drafted client message against the policy and the carrier's own material | State unfair trade practices statutes based on NAIC Model #880, sections 4A and 4B | Insurers, and by the model's definition producers, as enacted | Yes: the model does not mention AI; applying 4A or 4B to an AI-drafted message is an inference |
| Confirm prior express consent before any call using an AI-generated or simulated voice | FCC Declaratory Ruling FCC 24-17, adopted February 2, 2024, released February 8, 2024 | Callers subject to the TCPA, federal | No: the ruling is expressly about AI-generated voices |
Read the instruments properly before you build on them.
Professional guidance: New York OGC Opinion No. 04-12-03, December 2004. It asks what unlicensed employees of an agent or broker may do without a licence. Authorised activities include answering phones and scheduling appointments without coverage discussion, maintaining files, data entry, referring prospects to licensed agents, and accepting payments on existing policies, provided, among other things, that the employee is not compensated based on sales and there is no discussion of coverage. Unauthorised activities include soliciting insurance, quoting premiums, discussing coverage, limits or deductibles, and binding or modifying coverage. Licensees must weigh "the nature and complexity of the task being delegated", the employee's qualifications, and "the type and extent of supervision and internal controls that will be in place", and they "will be held strictly accountable by the Department for the activities of such non-licensees". It is a 2004 interpretive opinion of one state's Office of General Counsel. Other states draw the line differently. Confirm current law before relying on it.
Professional rule as adopted: NAIC Model #275, section 6A. When recommending an annuity a producer must act in the consumer's best interest without placing the producer's or the insurer's financial interest ahead of the consumer's, satisfying care, disclosure, conflict of interest and documentation obligations. Section 6A(1)(a) requires the producer to know the consumer's financial situation, insurance needs and objectives, understand the available options, have a reasonable basis to believe the option effectively addresses them, and "Communicate the basis or bases of the recommendation". Section 6A(4) requires a written record of any recommendation and its basis. Section 6A(1)(d) says the care obligation does not create a fiduciary obligation or relationship. Adoptions vary by state, and #275 says nothing about AI. It permits an insurer's supervision review to apply a screening system and to be accomplished electronically; that is the insurer's supervision, not a producer delegating the best interest judgement to software.
Regulation: FCC Declaratory Ruling FCC 24-17. The Commission confirmed that the TCPA's restrictions on "artificial or prerecorded voice" encompass current AI technologies that generate human voices, so such calls require the prior express consent of the called party absent an emergency purpose or exemption, with prior express written consent for calls containing advertisements or telemarketing. Voice cloning falls within the prohibition because it artificially simulates a human voice, and the TCPA does not carve out technologies purporting to provide the equivalent of a live agent. It addresses outbound calls using artificial or prerecorded voices. Related consent rules have changed since, so tell learners to confirm current requirements.
A question to sit with
Write the behaviour you want on Monday in one sentence. Now delete every word that describes what the learner should understand, believe or be aware of. If nothing is left that someone could watch happen, you have not chosen a behaviour yet.
What goes wrong
1. An objective nobody can perform. "Understand the risks of AI in insurance" cannot be done at a desk and cannot be observed. It is a topic, not a behaviour.
2. A behaviour with no instrument behind it. The session teaches that producers must log every AI prompt. Nothing in this register requires that. If you want it, say it is the agency's own rule or this programme's recommendation, and say which.
3. The insurer's duty taught as the producer's. A session tells producers they must maintain a written AI programme. That expectation sits on insurers under the NAIC Model Bulletin. The producer behaviour that is actually supportable is different and smaller.
4. One state's line drawn nationally. The New York opinion's list of unlicensed activities is presented as the national rule. It is one state's interpretive opinion from 2004.
5. The inference left unlabelled. A statute that never mentions AI is quoted as if it regulated the tool. Say plainly: this is the rule, and here is the inference I am drawing about the tool.
6. Ten behaviours in ninety minutes. Each is named once, none is practised, and the room leaves with a list.
The practical answer
Best practice: write the behaviour in the verb-object-moment-evidence form. A usable behaviour names who does it, the verb, the object, the moment and the trace it leaves.
- Weak: "Be careful with client data in AI tools."
- Usable: "Before pasting any document into the approved tool, the service representative classifies it as containing nonpublic personal financial information, health information, or neither, and records the classification on the activity."
Best practice: run the three-column trace before you commit to a behaviour. Instrument, reach, inference. If the instrument column is empty, the behaviour is either the agency's own rule or this programme's recommendation, and the slide must say which. Neither is a weaker reason to do it. Both are honest.
Best practice: pick one behaviour per session, and at most one more for a second role. A ninety-minute session that teaches one behaviour to three roles is coherent. One that teaches nine is a list.
Best practice: check the behaviour against the licence line first. Any behaviour you teach unlicensed staff must be one they may perform in their state. The New York opinion is a model of the reasoning, not a national rule: the licensee weighs the task's nature and complexity, the person's qualifications, and the supervision and internal controls in place, and remains strictly accountable. Applying that reasoning to a tool is an inference, and the tool holds no licence at all.
Best practice: for annuity work, keep the behaviour on the producer's own reasoning. The documentation obligation in section 6A(4) asks for the basis of the recommendation. A behaviour such as "the producer writes the basis in their own words before any drafting tool is opened, then uses the tool only to tidy the wording, and re-reads the result against what they wrote" keeps the judgement where the rule puts it.
AI Coalition Network recommendation: write the behaviour on the first slide and leave it up. If a segment does not serve the sentence on that slide, cut the segment. Module 03 covers keeping the material accurate as the instruments themselves change.
Remember this
- Decide the Monday behaviour first; everything else earns its place by serving it.
- Trace every behaviour to an instrument, its jurisdiction and whom it reaches, or label it as the agency's rule or this programme's recommendation.
- Say out loud when applying a rule that does not mention AI is an inference you are drawing.
- The New York opinion is one state's 2004 interpretation; the FCC AI-voice ruling is federal and expressly about AI voices.
- One behaviour, named in verb-object-moment-evidence form, beats nine topics.
Insurance example
Hollis Bramanti, a fictional licensed producer and trainer at the fictional Quillfeather Insurance Partners, is asked to design a ninety-minute session called "AI and Compliance". She refuses the title and starts from Monday instead. She lists six candidate behaviours and runs each through the three-column trace. "Log every prompt" has an empty instrument column, so she marks it as an agency rule the agency has not written, and drops it. "Maintain a written AI governance programme" traces to the NAIC Model Bulletin, which is addressed to insurers, so she drops it as a producer behaviour and notes it for a separate conversation with the principal about carrier questionnaires. "Confirm prior express consent before any AI-voice call" traces to FCC 24-17, federal, expressly about AI-generated voices, no inference needed; she keeps it for the one fictional staff member who runs outbound calling. "Route coverage questions to a licensed colleague" traces to the New York OGC opinion, one state, 2004, with an inference about tools; the agency is licensed in that state, so she keeps it for the two unlicensed assistants and writes the label on the slide. "Write the annuity basis in your own words first" traces to section 6A(4) of her state's rule based on Model #275; she keeps it for the three life and annuity producers. That leaves four candidates for three roles, so she cuts the consent behaviour to a five-minute mention for the whole room and a separate fifteen-minute clinic for the one person who calls. Her first slide reads: "Monday: before any AI-drafted client message goes out, the licensed producer checks every coverage statement against the policy and the carrier material, and notes the check on the activity." She writes the label underneath: the statute is the state unfair trade practices law based on Model #880, it does not mention AI, and the application to a drafted message is her inference. She then cuts a fourteen-slide segment on how language models are built, because it does not serve the sentence.
Responsible-use warning
Never teach a behaviour as required without naming the instrument, the jurisdiction and whom it reaches. Never present the NAIC Model Bulletin expectation of a written AI programme as a producer duty; it is addressed to insurers. Never present the New York OGC opinion as a national rule: it is one state's 2004 interpretive opinion and other states draw the line differently. Never apply a rule that does not mention AI without saying out loud that you are drawing an inference.
Human-review requirement
The instructor owns the trace behind every behaviour taught. Before delivery, a licensed reviewer, such as the agency principal or the designated responsible licensed producer, checks each behaviour against the instrument cited, confirms the jurisdiction and reached population appear on the slide, and confirms every unlabelled application of a non-AI rule has been labelled as an inference. Where a behaviour rests on the agency's own rule rather than an instrument, the reviewer confirms the agency has actually written that rule. The licensed producer, not the instructor, remains responsible for what the agency does with any output.
Practical exercise
Choose and trace one Monday behaviour. Use a fictional agency, fictional roles and fictional files throughout.
- Describe a fictional agency in four sentences: lines written, licensed and unlicensed headcount, whether it sells annuities, and whether it runs outbound calling.
- Write six candidate Monday behaviours for that agency, each naming a role, a verb, an object and a moment in the working day.
- For each candidate, fill the three-column trace: the instrument, the jurisdiction and body with whom it reaches, and whether applying it to an AI tool is an inference.
- Mark any candidate with an empty instrument column as either the agency's own rule or an AI Coalition Network recommendation, and say which.
- Delete any candidate whose instrument is addressed to insurers rather than producers, and write one sentence on how that expectation could still reach the agency.
- Choose one behaviour for the whole room and at most one more for a second role. Write each in verb-object-moment-evidence form in a single sentence.
- Write the exact label sentence you will say when you apply a rule that does not mention AI.
- Name the licensed person who reviews the resulting output, and write the trace the behaviour leaves in the agency system.
- List three segments you would have included in a session called "AI and Compliance" and cut each one that does not serve your chosen sentence, writing one line each on why.
Knowledge check
-
What three things must be recorded for every behaviour taught as required?
Answer guidance: The instrument behind it, the jurisdiction and body with whom it reaches, and whether applying it to an AI tool is an inference the instructor is drawing.
-
What does section 6A(4) of Model #275 require, as adopted by a state?
Answer guidance: A written record of any annuity recommendation and the basis for that recommendation, alongside the care, disclosure and conflict of interest obligations in section 6A.
-
What did the FCC confirm in the February 2024 declaratory ruling?
Answer guidance: That the TCPA restrictions on artificial or prerecorded voice encompass current AI technologies generating human voices, so such calls need the called party's prior express consent absent an emergency purpose or exemption.
-
Why can the New York OGC opinion not be taught as the national rule?
Answer guidance: It is a 2004 interpretive opinion of one state's Office of General Counsel on New York law, not a regulation, and other states draw the unlicensed activity line differently.
Sources and further reading
- New York State Department of Financial Services (Office of General Counsel) — OGC Opinion No. 04-12-03: Chapter 687 of the Laws of 2003: Permissible Duties of Unlicensed Customer Service Representatives of Agency or Brokerage (2004-12-01). Accessed 2026-09-15.
- National Association of Insurance Commissioners (NAIC) — Suitability in Annuity Transactions Model Regulation (Model #275) and the best-interest standard. Accessed 2026-09-15.
- National Association of Insurance Commissioners (NAIC) — Unfair Trade Practices Act (Model #880). Accessed 2026-09-15.
- Federal Communications Commission — Declaratory Ruling, Implications of Artificial Intelligence Technologies on Protecting Consumers from Unwanted Robocalls and Robotexts, CG Docket No. 23-362, FCC 24-17 (2024-02-08). Accessed 2026-09-15.
- AI Coalition Network — Certification Educational Standard: Human Review of AI-Assisted Professional Work. Accessed 2026-09-15.