Teach Insurance Teams to Explain AI-Assisted Work to Clients

This Partner Perspective is part of a series brought to you by Academy Journal.View series A customer emails your team to ask whether storm damage to a building is covered. An employee uses AI to draft a reassuring email. It says the claim is approved, the deductible is $1,000, and payment will arrive next week. The email reads beautifully. Unfortunately, nobody has made a coverage determination. The deductible in the policy summary is $5,000. The settlement amount and payment date were invented. Insurance managers can turn this kind of mistake into a useful teaching opportunity in 20 minutes. The exercise below gives agents, account managers, customer service representatives, and claims-support staff a repeatable way to check an AI-assisted explanation before it reaches a client. The skill they practice is simple to describe and surprisingly demanding: separate verified facts from open questions, explain both in everyday language, and know when a decision belongs to someone else. There is a good reason to practice. The National Institute of Standards and Technology’s generative AI guidance identifies confidently stated false information as a characteristic risk of these systems. And the National Association of Insurance Commissioners’ overview of AI in insurance emphasizes reviewing AI-generated material and maintaining human judgment in consumer-facing work. Give the Team a Case They Can Actually Check Use a fictional commercial property claim. Hand participants three short source cards and one draft client email. Keep the exercise fictional, so nobody has to put an actual customer’s policy, claim record, or personal information into an unapproved AI tool. Source card one, the sample policy summary: The insured operates a small retail store. The summary lists a $5,000 windstorm deductible. It describes potential coverage for certain direct wind damage, subject to the actual policy’s terms, conditions, and exclusions. It notes a flood exclusion. The full issued policy, including endorsements, controls any coverage analysis. This is a training example, not a specimen insurance contract. Source card two, the claim notes: The customer reported that strong winds damaged part of the roof and rain entered the store. A contractor photographed missing roof materials. An adjuster has been assigned, but the cause and scope of the interior damage remain under investigation. No coverage determination, approved repair estimate, payment amount, or payment date is recorded. Source card three, the communication record: The customer asks, “Can I tell my landlord that the roof and interior repairs are covered? When should I expect the check?” The employee must respond promptly without inventing an answer. The AI-assisted draft reads: “Good news! Your roof and interior water damage are fully covered. Your deductible is only $1,000, and the insurer will issue an $18,000 payment next week. You can go ahead and schedule all repairs.” Ask participants to mark every sentence that claims to know more than the source cards establish. They should find at least four problems: premature coverage approval, the wrong deductible, an unsupported payment amount and date, and permission to proceed as though the claim outcome were settled. The words “fully covered” also skip the unresolved cause of the interior damage. In real claims, the distinction between a loss reported by an insured and a coverage decision matters. The NAIC’s explanation of insurance policies and claims describes a claim as a request for payment under an insurance contract, rather than a guarantee of payment. Run the 20-Minute Exercise Minutes 0-2: Define the assignment. Tell the group that the objective is a client-ready explanation, not a coverage decision. Each participant can use only the supplied source cards. Make the distinction between what is reported, what is verified, and what is still pending explicit instructions. Minutes 2-6: Audit the draft. Working alone, participants underline unsupported promises, circle facts that are in the cards, and write a question beside each important uncertainty. Ask them to identify the person or team that owns each missing answer. A claims decision cannot be manufactured by better phrasing. The NAIC’s guide to settling a homeowners claim illustrates the adjuster’s role in assessing damage, reviewing coverage terms, and calculating a settlement Minutes 6-11: Rewrite the email. Keep the response under 120 words. It should acknowledge the customer’s question, state confirmed information, identify what has not been decided, and give a specific next step. The goal is a useful answer without a promise the employee cannot support Minutes 11-16: Practice the conversation. In pairs, one employee plays the client. The client asks, “But your email said everything was covered. Are you taking that back?” The other explains the correction plainly and without blaming a machine. Switch roles. The NAIC’s disaster-claim guidance stresses the importance of reporting the loss, documenting the damage, and working with the insurer through the claims process Minutes 16-20: Score and debrief. Each pair reads the revised explanation against the three criteria below. The manager asks which source would change the answer, which open question requires escalation, and what would prevent the original draft from being sent in the future Here is one acceptable rewrite: “Thanks for checking. We’ve documented your report of roof damage and water entering the store. The adjuster is still reviewing the cause and extent of the damage, so the insurer has not yet confirmed which repairs are covered or a payment date. The policy summary lists a $5,000 windstorm deductible, but the claims team will need to confirm how the policy applies to this loss. I’ll request an update from the claims team and pass along the confirmed next step.” That response gives the client a truthful status and a next action. It also avoids treating a policy summary as a final coverage determination. The NAIC’s explanation of homeowners policy provisions underscores why coverage, endorsements, and deductibles require attention to the actual policy language. Score What Matters, Not How Polished It Sounds Give each dimension zero, one, or two points, for a maximum of six. A manager can score a rewritten email in under a minute once the rubric becomes familiar. Accuracy: Zero means it invents or misstates a material fact. One means it gets the main facts right but obscures a significant unknown. Two means each substantive statement matches a source card and the pending issues are unmistakable. The NAIC’s advice on understanding insurance coverage reminds consumers to know their policy, exclusions, and deductibles, making factual precision especially important in client explanations. Clarity: Zero means the client would leave with a false impression or a wall of jargon. One means the message is technically understandable but vague about status or next steps. Two means a client can explain back what is known, what remains undecided, and when they will hear more. The question is whether the client understood, not whether the AI produced elegant sentences. Appropriate escalation: Zero means the employee makes a coverage, liability, or payment commitment beyond their authority. One means the email says someone should check but fails to identify who or what. Two means it names the claims team or other appropriate decision-maker, identifies the question for that person, and provides a realistic follow-up action. The NAIC’s consumer guidance on claims questions and disputes distinguishes the roles of insurers and state regulators when questions about coverage or claims remain unresolved. Set one additional rule: Any unsupported coverage approval, payment amount, payment timing, or other consequential promise triggers a rewrite regardless of the total score. A fluent error should never earn a passing grade by compensating with good style. Make the Habit Survive Beyond Training The easiest follow-through is to run the same drill with a new fictional case each week for a month. Rotate among a certificate-of-insurance request, a renewal coverage question, a claim-status update, and an explanation of a policy exclusion. Change the disputed fact, the decision-maker, and the temptation built into the AI draft. Keep the scoring rubric stable so managers can see whether employees are becoming more reliable. For actual client work, employees should compare the proposed explanation with the current policy, claim record, approved guidance, and relevant communication history using their organization’s authorized systems. Record which facts were checked, which questions were escalated, and who approved any consequential representation. The NAIC’s consumer guidance recommends retaining records of insurance communications. The training equivalent is a short audit trail that lets a manager see what supported the message before it went out. In coaching sessions, ask one final question: “What would I have to verify before I could say that sentence to the customer?” A team that can answer that consistently will communicate more clearly whether it drafts with AI, templates, or a blank page. It will also be better prepared when a client asks the awkward but reasonable question: “How do you know?” About the Author Dr. Gleb Tsipursky is the CEO of Disaster Avoidance Experts, a consultancy focused on workplace AI adoption. A behavioral scientist, he is the author of eight books, including The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). He works with organizations to turn AI tools into practical, verifiable workflows.