AI for Asset Managers: Client Reports, Meeting Notes and Records

A small asset manager can safely use AI to draft and tidy client letters, commentary and meeting notes from material it already owns, summarise public research, and prepare agendas, as long as a qualified person checks every figure and claim. Keep the final client document, the review note and a log of what AI touched. Leave judgement calls and anything involving suitability to people.

The records question is the one that gets skipped. A registered adviser must already keep written communications about advice and recommendations, and an AI-drafted letter is still a written communication. The difficult part is rarely the tool. It is being able to show, a year later, how a client report was produced and who stood behind it. For a firm of two to twenty people, the answer is a short safe-use list, one review habit and four kinds of record, not an enterprise AI strategy.

Follow me on Instagram@sagnikteaches

The safe-use list, in three bands

Sort every task by two questions: does client or portfolio data go in, and would a wrong output reach a client or regulator unchecked? That gives three bands.

Connect on LinkedInSagnik Bhattacharya

Print the table and pin it near the team. Its job is to make the default answer to "can I use AI for this?" quick and consistent, instead of left to whoever is nearest a free chatbot.

Subscribe on YouTube@sagnikteaches

Meeting notes: the highest-value amber task

Client meetings are where a small firm loses the most time, because each one produces notes, a file entry, follow-up emails and CRM updates. AI note takers cut that to a review. On their pricing pages in November 2026, Zocks lists Essentials, Professional and Ultimate at $67, $117 and $184 per user per month billed annually ($80, $140 and $220 billed monthly), with transcript and meeting-result archiving from Professional upwards; Wealthbox lists its CRM at $59 (Basic), $75 (Pro) and $99 (Premier) per user per month, with the AI Notetaker as a $49 per user add-on; and Jump lists Meet at $100 per adviser per month, with up to 20 per cent off for annual payment. Prices change, so check the vendor page before you buy, and see the Jump versus Zocks comparison for how the two differ on compliance features.

Whichever tool you choose, three settings decide whether the notes are a record or a liability.

1. Consent to record. Tell clients at the start and note it. Wording is in the consent wording tutorial.
2. Where the output lands. Notes should sync into your CRM or archive, not live only in the vendor's app. If the vendor goes away, your record must not. Ask whether the tool feeds the archive you already use for email: Zocks, for example, says it integrates with archives such as Smarsh, Global Relay and Proofpoint so transcripts can be archived and supervised the same way as email.
3. Retention. Decide whether the recording and raw transcript are kept, for how long, and who can delete them.

A before and after from one meeting

Say a four-person firm meets a client about moving $400,000 from cash into a balanced portfolio. The adviser's old process is 20 minutes of typing afterwards from memory. With a note taker, the draft summary arrives within minutes. Illustrative draft output:

Summary: Client wishes to invest $400,000 held in cash. Agreed in principle to a
balanced allocation. Client concerned about a fall in the first year.
Actions: (1) Adviser to send risk-profile questionnaire by Friday.
(2) Client to confirm source-of-funds documents.
(3) Follow-up meeting in two weeks.

What the reviewer fixes: the client actually said "I could live with a 10 per cent fall, not 20", which the summary softened into "concerned". That sentence is exactly what a later suitability question turns on, so it goes back in verbatim. Also missing: the client's mention of a house purchase in 18 months. Review takes about five minutes instead of twenty, and the note is better than the memory-based one.

Client reports and letters: draft with data held back

Quarterly letters and commentary are the second amber task. The safest pattern keeps the numbers out of the prompt and puts them in afterwards.

Write a 180-word quarterly letter for a client with a balanced portfolio.
Structure: one sentence on markets in general terms, one on how the portfolio
behaved, one on what we did, one on what we are watching, one on next steps.
Leave every number as [PERF], [BENCHMARK], [FEE] or [DATE]. Do not predict
returns, do not use the words "guarantee" or "outperform", and do not name
any individual security.

The placeholders mean the model cannot misstate a figure, and the banned words keep the draft clear of promissory language. You fill the numbers from your portfolio system. The reviewer then reads the finished letter as though it were an advertisement, because commentary that is later reused on your website or in pitch material can become one, and checks every figure against the system. Illustrative mistakes this catches: the draft says "markets fell modestly" in a quarter your own report shows down 6 per cent, or describes a holding as "defensive" when the fund factsheet calls it growth-oriented.

For factsheet-style commentary there is a dedicated walk-through in the factsheet commentary tutorial, and the annual review tutorial covers preparing the review pack.

Five more concrete uses, with the check that goes with each

• Explaining a fund in plain English. Paste the fund's public factsheet and ask for a 100-word explanation at the level of a non-specialist. The check: does every statement appear in the factsheet? Remove any the model added from its own background knowledge.
• CRM clean-up. Ask the CRM's built-in assistant to turn a messy two-page note into five headed fields. The check: names, dates and amounts against the source note.
• Prospect call preparation. Feed in only public information about a prospect's business and ask for ten discovery questions. No client data, so green band.
• Compliance calendar. Give the model your list of recurring obligations and ask for a quarterly calendar with owner columns. The check: your compliance consultant reads it against your actual manual.
• Research digest. Summarise a public research report into a one-page note for the team. The check: skim the original for anything the digest left out that changes the message, and never forward the digest as if it were your own research.

Notice the pattern: the more the task depends on the model's own knowledge, the more checking it needs. Tasks where you supply the source and ask for restructuring are the best value.

The records: what to keep and how long

A registered investment adviser keeps books and records under SEC Rule 204-2, available on the Cornell Legal Information Institute site. Two parts matter here. Paragraph (a)(7) covers written communications received and sent that relate to recommendations or advice given or proposed. Paragraph (a)(11) covers advertisements. Paragraph (a)(16) covers the records needed to show how any performance figure presented in an advertisement, investment letter or similar communication was calculated, so a return quoted in an AI-drafted piece must trace back to your own workings. The general retention period is five years from the end of the fiscal year in which the last entry was made, with the first two years in an appropriate office of the adviser. I cannot tell you how the rule applies to your firm, so confirm it with your compliance consultant. Firms outside the SEC regime should check their own regulator's rules and their professional indemnity policy.

For AI work, a sensible record set has four parts.

1. The final document. The letter, report or note as sent or filed. This is the legal record.
2. The review note. Who reviewed, when, what was changed.
3. The AI-use log. One line per use, covering tool, plan, task and data type.
4. The vendor file. The tool's contract, data terms and settings screenshot, dated.

Whether you also keep prompts and raw AI drafts is a policy choice. Keeping them costs storage and creates more material to produce if there is an inquiry. Not keeping them means you cannot show the first draft. Many small firms keep the final and the review note, and keep drafts only for the highest-risk documents. Write the choice into your policy and follow it consistently.

A one-line log entry, filled in

Date: [date] | Task: Q3 client letters (18 clients) | Tool: [chat tool], team plan
Data in: placeholders only, no client data | Reviewed by: [name], CCO
Changes: 11 of 18 letters had performance wording softened; 2 had the wrong benchmark
Stored: archive folder Q3/Letters, review notes alongside

Illustrative, and it takes under a minute. Over a quarter, the "changes" column also tells you whether the tool is helping or just adding checking work.

Vendor questions that matter more for an asset manager

Portfolio holdings, account numbers and client net worth are the most sensitive material a small firm holds, so the vendor check is stricter than for an ordinary business. Ask each vendor the following and file the answers with a date.

1. Is our content used to train models? State the plan you are buying, since consumer and business plans differ.
2. Can an administrator control who sees shared chats, recordings and transcripts, and remove a leaver immediately?
3. How long are recordings, transcripts and uploaded files kept, and can we set or delete them? Zocks, for instance, lists archiving as a feature from its Professional plan upwards, so the plan tier changes your record-keeping options.
4. Does the vendor hold an independent security attestation? Zocks's pricing-page FAQ, for example, says it is SOC 2 Type II certified; check any vendor's current claim and ask for the report itself, not the badge.
5. Does the output sync to our CRM, and in what format? A record that cannot leave the vendor is a dependency.
6. What happens to our data when we cancel?

A firm that cannot get clear answers to questions 1 to 3 should keep that tool in the green band, no client data, until it can.

Claims about your own AI: keep the brochure honest

The second records risk is outward-facing. If your website says "AI-powered portfolio analysis" and the truth is that an analyst uses a chatbot to tidy commentary, that sentence is a representation you must be able to support. The SEC's examination priorities for fiscal year 2026 say examiners will review registrants' representations about their AI capabilities for accuracy, and a marketing claim in a brochure is also an advertisement to retain under paragraph (a)(11). Use plain, true language such as "we use AI tools to help draft documents, and every document is reviewed by a member of the team". Avoid words that imply the technology makes investment decisions. Run any new website sentence about AI past your compliance consultant before it goes live.

A research digest, before and after

A quick example of the check that catches the quiet errors. An analyst pastes a public 12-page research note on a sector and asks for a one-page digest for the team. Illustrative output includes the sentence "The report expects margins to expand in the second half." The reviewer searches the original and finds it says margins "may stabilise if input costs ease". The model turned a conditional into a forecast. That is the typical failure: confident, tidy and slightly wrong. The fix is a standing instruction in the prompt ("quote hedged language as hedged; never turn a conditional into a prediction") and a reviewer who checks the three most important sentences against the source. For anything that will inform a client decision, the original is read in full, and the digest is just a map.

Staff rules that fit on half a page

A policy nobody reads protects nobody. For a small firm, these seven lines cover most of the risk and can sit inside your wider compliance manual.

• Only approved tools, listed with the permitted band for each.
• No client names, account numbers or holdings in any tool outside the approved amber list.
• Every client-facing document is reviewed by a named person before it is sent.
• Every figure is traced to the portfolio system or a source document.
• No predictions, guarantees or "outperform" language in AI-drafted text.
• One log line per use of AI on a client document.
• Report any slip, such as pasting the wrong thing into the wrong tool, to compliance the same day. Nobody is disciplined for reporting.

The acceptable-use policy tutorial has a longer template to adapt, and it is worth a quarterly read-through with the whole team, because the tools will have changed by then.

What the SEC's 2026 examination priorities say about AI

The Division of Examinations' Fiscal Year 2026 Examination Priorities place AI under emerging financial technology. Examiners will review for accuracy registrants' representations about their AI capabilities, and assess whether firms have adequate policies and procedures to monitor or supervise their use of AI technologies, including in back-office operations. The same document covers Regulation S-P: exams will look at policies and procedures, internal controls, oversight of third-party vendors and governance, and, once the compliance dates for the amended rule have passed, the administrative, technical and physical safeguards for customer information. An AI note taker or chat tool that holds client data is exactly the kind of third-party vendor that oversight covers. The document is a staff statement with no legal force, but it tells you what an examiner is likely to ask. The practical reading for a small manager is simple. Do not claim AI capabilities you do not have in a brochure or on your website, protect client information in every tool, and be able to show a written policy that matches what the team really does.

A worked example: a boutique firm adds AI in six weeks

Take a six-person firm with 85 client households and about $150 million under management. The principal, two advisers, an analyst, an operations lead and a part-time compliance consultant want to cut reporting time without a compliance scare.

• Week 1. The operations lead lists every AI tool already in use. Two staff use free chatbots for email drafts; one adviser uses a note taker on personal licence. The compliance consultant rates them against the three bands.
• Week 2. The firm picks one team-plan chat tool for green and amber uses and trials one note taker for three advisers. It requests the data agreement for each, checks training and retention terms and files them.
• Week 3. The safe-use table and a one-page policy go to staff. Free chatbots are banned for anything touching a client or portfolio.
• Week 4. Advisers run ten real meetings through the note taker with consent. Each summary is reviewed. The average review takes six minutes, against about twenty to type notes from scratch; across 10 meetings that frees roughly two and a half hours.
• Week 5. The quarterly letter process moves to the placeholder prompt. The analyst fills numbers from the portfolio system, an adviser reviews and the compliance consultant samples five letters.
• Week 6. The log and review notes are checked, the policy is amended in light of what went wrong (one summary had misattributed a client's words to a spouse) and the firm decides which tool stays.

The saving is modest per meeting and large across a year, and the review habit is what makes the output safe to rely on. For a set-up walk-through on one specific tool see the Jump set-up tutorial.

Failure modes to expect

These are the problems a firm like the one above meets in the first quarter. Most are caught by the review step, which is why it cannot be optional or delegated to the person who wrote the prompt.

• Invented specifics. A summary adds a figure or a product name the client never mentioned. Check names, numbers and dates against the recording or your own notes.
• Misattributed statements. Two voices in a meeting get merged. This matters when one is the client and one a family member.
• Softened risk language. The model smooths the client's real worry into something bland. Restore the client's own words on risk tolerance and objectives.
• Promissory phrasing. Draft commentary drifts towards "should deliver" or "protects". Strike it.
• Shadow tools. Someone keeps using a personal app. The fix is an approved tool that is as easy to reach, plus a short conversation, not a lecture.

If you would rather map the safe-use list and the record set with someone, that is what my AI implementation consultation covers. Either way, start with the green band this week, put the review habit in before you move to amber, and keep the log from day one.

Questions small asset managers ask about AI

Do AI-generated meeting notes count as books and records?

If the notes record advice, recommendations or client instructions, they are the kind of record a registered adviser keeps, whoever or whatever wrote them. Keep the reviewed, final version in your archive and decide in your policy whether the raw transcript and AI summary are kept too. Ask your compliance consultant to confirm the rule for your firm.

Can I paste portfolio data into ChatGPT or Claude?

Only on a plan whose contract you have reviewed, and never client-identifying data on a free consumer plan. ChatGPT Business and Claude Team state they do not train on business content by default. Confirm that for your plan, get a data agreement where available and test with anonymised figures first.

Which tasks are safest to start with?

Start with tasks that use no client data or only your own published material: rewriting commentary, drafting a meeting agenda, summarising public research, and turning your own bullet points into a letter. Move to notes and reports from client data only after the vendor checks and a compliance review step are in place.

Should the manager or an analyst review AI output?

Someone who could have written it themselves. Every figure should be traced to the portfolio system or a source document, and every statement about performance, risk or outlook read as if it were a marketing claim. Record who reviewed it and what changed.

What does an AI-use log need to contain?

Date, task, tool and plan, who reviewed, what data went in, what was changed and where the final output is stored. One line per use in a shared sheet is enough for a small firm. Its job is to let you reconstruct how a client document was produced if anyone asks.

Further reads

• AI Financial Planning Software for Advisers: What to Test First — What to test first in planning software with AI.
• AI Note Takers for Financial Advisers: Compliant Meeting Records — Compliant meeting records with an AI note taker.
• Can Financial Advisers Use ChatGPT? What Compliance Allows — What compliance allows with a general chatbot.
• HubSpot for Financial Advisers: Pipeline and Where Breeze Stops — Where a CRM assistant stops being useful.
• Claude vs ChatGPT for Financial Advisers: Meeting Prep and Records — Two assistants compared on prep and records.
• Claude for Financial Advisers: Plugin, Connectors and Plan Needed — What a financial-adviser plugin for Claude needs.
• AI CRM for Financial Advisers: What Is Built In and What Is Bolted On — Built-in versus bolt-on AI in your CRM.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: SEC Rule 204-2 (books and records of investment advisers, via the Cornell Legal Information Institute); SEC Division of Examinations, Fiscal Year 2026 Examination Priorities; Zocks, Wealthbox and Jump pricing pages; ChatGPT Business and Claude Team pricing pages.