Meeting Prep, Notes, and Client-Ready Outputs

The annual or quarterly review is the core recurring event in an advisory relationship, and preparation for it is a well-defined assembly problem rather than a creative one.

What an advisor needs in front of them before the client sits down. What the portfolio has done since the last conversation and why. What was agreed last time and whether it happened. What is outstanding from either side. What has changed in the client's stated circumstances. Any life events mentioned in passing that were never followed up. And what needs a decision at this meeting.

Assembling that manually means opening the portfolio system, the CRM, the last meeting note, the file of correspondence since, and reconstructing the thread. Twenty to forty minutes per client, done in the gap before the meeting, and frequently done badly because the gap was fifteen minutes.

What a well-structured preparation workflow produces is a one-page brief from those same sources, with the outstanding items and the unfollowed threads surfaced explicitly.

The second item is the one that changes the meeting. A client who mentioned eight months ago that their daughter was starting university, and is asked about it, experiences an advisor who was paying attention. That detail was always in the file. It was just never retrieved, because retrieving it cost more than the meeting had time for.

And the risk profile here is low, because the output is an internal brief that never leaves your desk.

Meeting capture is the highest-value workflow in the practice, and it needs to be set up correctly to be safe.

Consent comes first. Recording a client conversation requires the client's agreement, and in several jurisdictions and US states, the agreement of all parties. This is a one-sentence ask at the start of the meeting and it is not optional.

What then becomes available. A transcript. A structured meeting note. The action items with owners. A follow-up letter confirming what was discussed. And the suitability documentation, populated from what the client actually said rather than from memory three days later.

The compliance argument for this is stronger than the efficiency argument, which surprises people. A contemporaneous record derived from the actual conversation is better evidence of what was discussed and agreed than a note written from memory later in the week. Where the question is whether the client was told about a risk, the transcript answers it.

The conditions. The recording and transcript are client records and belong inside the firm's retention and supervision system, not on a consumer transcription service. The generated note is a draft the advisor reviews, because a model summarising a long conversation will occasionally attribute a position to the wrong party or compress away a caveat that mattered.

And one behavioural point. Advisors who stop taking notes during meetings report that the quality of the conversation changes, because they are looking at the client rather than at a pad. That is a real effect and it is arguably worth more than the time saved.

The pipeline that converts a conversation into the four artefacts it should produce.

Start with consent, which gates everything downstream. Without it the pipeline does not run.

The recording produces a transcript, which is itself a record and goes into retention.

From the transcript, four generated drafts. The meeting note for the file. The action list with owners and dates. The client follow-up letter. And the suitability or review documentation.

Every one of those four passes through advisor review before it is filed or sent, and the review is looking for specific things: a position attributed to the wrong party, a caveat that was compressed away, a number that should have come from the planning system, and anything the client did not actually say.

Only then do they split. The internal artefacts go to the file. The client letter goes out, and once it goes out the advisor is asserting it.

The structure matters because the temptation is to file the generated note directly. It is usually good, which is exactly why the occasional misattribution goes unnoticed.

Client-ready written output is where generation is genuinely strong, for a reason that is worth naming.

The hard part of explaining a financial concept is not knowing it. It is finding the version of it that lands with this particular person: their vocabulary, their prior knowledge, the analogy that works for them, and the length they will actually read.

An advisor knows the concept perfectly and has one or two explanations they have refined over years. A model can produce eight variants at different levels of technicality in seconds, and the advisor picks the one that fits the client.

Where this helps most. Explaining sequence-of-returns risk to someone approaching retirement. Explaining why a diversified portfolio underperformed the index that was in the news. Explaining a fee structure clearly enough that the client genuinely understands what they pay. Explaining a tax consequence without producing a document that needs its own explanation.

The last of these is a real problem in the profession. Written client communication in financial services drifts toward defensive completeness, which produces documents that are technically accurate and functionally unread. A model asked to produce a shorter, plainer version tends to make it better, not worse, because the length was serving the writer rather than the reader.

The review discipline. Every number verified against the system it came from. Every product fact verified against the document. And a check that simplification has not become inaccuracy, which is the specific failure mode when you ask for plainer language.

There is a threshold in client-facing content that changes the rules, and advisors cross it without noticing.

In the United States, the Investment Advisers Act Marketing Rule, Rule 206(4)-1, governs advertisements by registered investment advisers, with a compliance date of November 2022. It defines an advertisement broadly, reaching communications offering advisory services to prospective clients and, importantly, certain communications to existing clients that offer new or additional services.

So a personalised review letter to an existing client about their existing arrangement is generally not an advertisement. A newsletter, a market commentary distributed to a list, a social post, or a letter proposing an additional service can be.

What that means for generated content. The Marketing Rule prohibits untrue statements of material fact, unsubstantiated claims, and material statements the adviser cannot substantiate upon demand by the Commission. A model producing market commentary will readily generate confident claims about performance, market direction and product characteristics, and the substantiation requirement means the burden of proving each one sits with the adviser.

And performance presentation is heavily specified: the rule imposes requirements on net and gross performance, prescribed time periods, and related and extracted performance. This is not territory for generated prose at all.

The practical division. Explanation to an existing client about their own situation: generate and review. Anything distributed to a list or reaching prospects: generate the draft and route it through the firm's normal marketing review, unchanged from how any other advertisement is handled.

Digesting research is an obvious use and the one where verification discipline matters most, because the output is factual claims about specific things.

What works. Summarising a long fund prospectus or annual report for your own understanding. Comparing the stated strategies of several funds. Condensing a regulatory update into what changes for your practice. Extracting the assumptions behind a piece of market commentary so you can judge whether you accept them.

What fails, specifically and predictably. Expense ratios, historical returns, fund holdings, manager tenure, minimum investments and regulatory citations are all things a model will produce fluently and sometimes incorrectly. These are precisely the details a client acts on.

The pattern that makes this safe is retrieval rather than recall. Supply the actual document and ask questions against it, rather than asking the model what it knows about a fund. The failure rate on the first is far lower, and when it does fail you can find the answer in the document you supplied.

A second discipline for anything that will be repeated to a client or entered into a file: check it at source. Not because models are unusually unreliable here, but because the cost of a wrong expense ratio in a client recommendation is disproportionate to the ten seconds of checking.

And note the asymmetry that makes this workflow worth doing anyway. Reading the whole prospectus was never happening. A summary you spot-check is more coverage than most advisors previously had, not less.

One workflow is conspicuously absent from this lesson, and the omission is deliberate.

Advisors are frequently pitched tools that hold the client conversation: a chat interface that answers client questions, a system that conducts the fact-find, an assistant that handles routine enquiries. The pitch is that this frees the advisor for higher-value work.

The difficulty is that the conversation is the higher-value work, for a specific reason.

A large part of what an advisor does in a meeting is not information transfer. It is noticing. The client says they are comfortable with the risk level and their body language says otherwise. They mention their brother's business in passing and it turns out to be a substantial contingent liability. They say they want to retire at sixty and what they mean is that they want to stop doing this particular job. The stated objective and the actual objective differ, and reconciling them is the skill.

Behavioural finance research since Kahneman and Tversky's work on judgement under uncertainty has documented how poorly stated preferences predict actual behaviour under stress, which is the entire reason an advisor talks a client out of selling at the bottom.

None of that is available to a system taking a structured fact-find, because the client answers the questions asked.

So the honest framing. Automate the preparation for the conversation and the documentation after it. The conversation itself is not overhead to be optimised away. It is the product.

Pulling the review disciplines together, since each workflow has a different one and applying a generic check catches none of them.

Meeting preparation. Internal only, so the review is light. Confirm the brief has not invented a client circumstance, which is the one failure mode that would carry into the meeting.

Meeting notes and follow-ups. Check for misattribution, a party's position assigned to the wrong person, and for compressed caveats, where a qualified statement in the conversation became unqualified in the note. Both are common and neither looks wrong on the page.

Client explanations. Every number traced to the system that computed it. Every product fact traced to the document. And a specific check that plainer language has not become inaccurate language.

Anything that meets the definition of an advertisement. Normal marketing review, plus substantiation for every material claim, because the burden of substantiating on demand sits with the adviser.

Research summaries. Retrieval over recall, and source verification for anything that will be repeated to a client.

The general shape of it. The review is not a proofread. It is a targeted search for the two or three specific ways this particular workflow fails, which takes a couple of minutes when you know what you are looking for and produces nothing when you do not.

And across all of it, the adoption principle: what goes into the file or out to the client is the advisor's statement, whatever produced the draft.