萃靈 Essevin
An insurance meeting rarely ends with a clean, structured record. An adviser may leave with notes about family responsibilities, a summary of existing cover, an unconfirmed budget, several voice-note fragments and a few questions that must be resolved at the next meeting.
Asking AI to “recommend the right policy” is the wrong shortcut. It asks a tool without the client’s full circumstances, the firm’s procedures or the complete product documents to make a professional judgement. A more useful role is administrative: organise approved, anonymised material into an internal Needs Analysis Brief.
The Brief is not an answer. It is a meeting-preparation map showing what is known, what is missing, where each statement came from and what the adviser still needs to ask.
Use AI as an organiser, not an insurance adviser
Think of the order slip in a busy Hong Kong cafe. It can record “less sugar”, “no ice” and “takeaway” so the kitchen does not miss a detail. It cannot choose the meal for the customer, and it should never guess about an allergy that was not recorded. In the same way, AI can organise sources, group needs and draft neutral questions. Suitability, coverage interpretation, product comparison conclusions and recommendations remain human responsibilities.
A usable Brief separates four things: supplied facts, the client’s own words, missing information and working assumptions. If an assumption is rewritten as a fact, every later step can move confidently in the wrong direction.
The five-step workflow
1. Prepare only the minimum necessary material
Follow your organisation’s data-classification, confidentiality and approved-tool rules before submitting any material. Remove names, telephone numbers, email addresses, identity numbers, policy numbers, bank details and identifiable family information when they are not needed for the task. Use labels such as “Client A” and “Dependant 1”.
Anonymisation is not a universal permission to upload. A combination of details may still identify someone, and the tool, storage location and access rights must still comply with the organisation’s rules.
Add a source marker to every input, such as Meeting Note 2026-08-20, paragraph 4 or Policy Summary A, page 2. A summary that cannot be traced back to its source is difficult to review.
2. Build a neutral fact table
Ask the model to group only the supplied material under headings such as family and financial responsibilities, protection objectives, existing arrangements, time horizon, budget statements, preferences or constraints, and information still to be confirmed. Every entry should retain its source marker. Blank fields must stay blank.
3. Convert gaps into follow-up questions
Turn each gap into a neutral information request. Instead of asking, “Do you need more cover?”, ask, “The current notes do not state the amount and duration of this responsibility. Which figures should be confirmed at the next meeting?” The question should collect facts, not steer the client towards a product.
4. Draft the follow-up message
The draft should summarise confirmed points, list documents or figures the client needs to provide, and suggest the next meeting arrangement. Any description of benefits, exclusions, returns, fees, coverage or eligibility must be checked against the current official product documents and reviewed by an authorised person before it is sent.
5. Review and record the decision
Return to the original notes and policy material. Check that the draft did not lose a negative, deadline, exception or expression of uncertainty. An authorised person decides which questions to ask, what may be sent to the client and what must go to compliance or a supervisor. Record the reviewer, date and version.
Reusable prompt
You are an insurance-meeting information organiser. Produce an internal
meeting-preparation draft only. Do not perform a suitability assessment,
recommend a product, interpret coverage, predict returns, or complete any
disclosure, application or approval.
Inputs:
1. Meeting notes anonymised and approved under the organisation's rules;
2. Approved summaries or comparisons of existing policies;
3. A source marker for every passage.
Produce:
A. Needs Analysis Brief: supplied facts and the client's own words only;
B. Information gaps: mark "not provided", "inconsistent" or
"requires human confirmation";
C. Follow-up questions: state which missing fact each question resolves;
D. Follow-up message draft: confirmed points, documents or figures required,
and the next meeting arrangement only;
E. Source Map: map every item to its original location.
Rules:
- Do not invent income, liabilities, health information, family needs,
risk preferences or protection objectives.
- Do not decide that any product is suitable, sufficient, better or required.
- Do not rewrite product terms or add benefits, exclusions, returns or fees.
- If sources conflict, quote both and do not choose one.
- Mark all client-facing text "Draft for authorised human review".Human review checklist
Can every fact be traced to a note, page or paragraph?
Are the client’s words, confirmed facts, model classifications and working assumptions clearly separated?
Does the draft still contain unnecessary or potentially re-identifying data?
Has every item of product information been checked against a current official source?
Are suitability, recommendation, disclosure and client communication handled by an authorised person?
Does the final version record its reviewer, date and next action?
Common traps
Treating anonymisation as permission to upload anything. Anonymisation is one control. The tool must still be approved, the data must be necessary, and storage and access must follow organisational rules.
Turning information gaps into sales prompts. The Brief identifies what is not yet known. It must not assume the client needs a particular type of cover.
Letting AI rewrite product terms. A smooth summary may omit a definition, exception or condition. Anything that may affect a client’s understanding or decision must be checked against the official document.
Sending the follow-up draft without review. The model does not know the organisation’s full requirements, the client’s complete circumstances or who is authorised to make a statement. Every client-facing output remains a draft.
Keep meeting time for the questions that matter
A well-structured Needs Analysis Brief does not make decisions for the adviser. It turns scattered material into a traceable working paper, reducing the chance of missing a question and making it easier for a reviewer to see where every statement came from.
Essevin AI Chat supports multiple models that can help organise these drafts within an organisation’s approved data-handling boundary. The tool, data and workflow must still be approved, and an authorised person must review every professional judgement and client-facing statement.
Information in this article is current as of 28 August 2026 and is provided for general reference only; it does not constitute advice of any kind. Third-party product features, pricing and policies are subject to their official announcements. Essevin service details are as shown on essevin.com and in the console.