How Advisers Use AI to Prepare for Annual Client Reviews

Advisers use AI for the reading and drafting in review prep: summarising last year's file notes and emails, flagging what may have changed, writing a personalised pre-review questionnaire and agenda, and turning platform reports into plain-English commentary. Valuations, performance calculations and any recommendation stay with the adviser and the platform.

The biggest gain isn't the pack itself. It's the "since we last spoke" brief: a page that pulls together everything the client told you over twelve months, from the review meeting, the phone call in March and the email about their daughter's wedding. That brief is what makes a review feel personal, and it's the part that used to get skipped when the diary was full.

Follow me on Instagram@sagnikteaches

What AI drafts and what the adviser owns

Settle the split before writing a single prompt, because it shapes everything that follows:

Connect on LinkedInSagnik Bhattacharya

The rule behind the table is simple: AI handles words about the client, never numbers about their money or judgements about what they should do. Language models are poor at arithmetic and fluent enough to make a wrong figure sound authoritative.

Subscribe on YouTube@codingliquids

Four weeks out: the "since we last spoke" brief

Export or copy the client's records from the past twelve months: last review's file note, any interim meeting notes, logged calls and relevant emails. Work in a business-plan assistant (ChatGPT Business, Claude Team, or Copilot inside your Microsoft 365 tenant) where content isn't used for training by default, and keep one client per chat or project. Mixing two families' details in one conversation is how a client ends up reading someone else's circumstances.

Below are my records for one client household from the last
12 months, in date order. Write a one-page brief for their
annual review.

Sections:
1. What they told us they wanted last year (quote them)
2. Events since then, with dates (job, health, family, property,
   money received or spent)
3. Things they said they would do, and whether we know if they did
4. Open questions: anything mentioned but never followed up
5. Possible changes to check at the review

Rules: use only these records. Give the date and source for every
point. If records conflict, show both versions. Do not suggest
advice or products.

Records:
[paste]

Section 4 is where the value hides. "Mentioned in a June call that her mother may move in" is exactly the sort of detail that changes a plan, and it's easy to lose between reviews.

For a couple in their early sixties with a year of fairly routine contact, the output comes back looking something like this (illustrative and abridged):

1. What they wanted last year
- "We'd like to both stop full-time work by 62 and still help
  with the grandchildren." (Review note, October)

2. Events since then
- February: Client 1's employer announced a restructure; he may
  be offered voluntary redundancy. (Call log, 9 Feb)
- June: Client 2 said her mother may move in with them.
  (Call log, 21 Jun)
- August: asked how to add their daughter to the joint savings
  account. (Email, 3 Aug)

3. Things they said they would do
- Update their wills after the grandchild's birth. Not known
  whether done.

4. Open questions
- Redundancy: nothing recorded after 9 Feb.
- Mother moving in: nothing recorded after 21 Jun.

5. Possible changes to check
- Client 1's employment; household costs; care needs.

What needs fixing? One line. The February call log actually said he had been offered voluntary redundancy and was deciding whether to take it. "May be offered" is softer, and it would have led to a vaguer question at the review. Opening the source line took thirty seconds, which is why the prompt insists on a date and source for every point. The August email is also filed under "events" when it's really an admin request; harmless, but it shows the model sorting by what was said, not by what matters.

A thin brief is information too. For a client with one logged contact all year, the review note itself, a well-behaved model returns a page that is mostly "no records": nothing under events, nothing under open questions. The temptation is to pad it from memory, or to let the AI "infer likely changes for someone of this age", which produces a plausible fiction. Leave the gaps visible. A year of silence from a client who used to call every quarter is worth a phone call before the questionnaire goes out, because the questionnaire assumes a relationship the records say may have gone quiet.

Three weeks out: a questionnaire built from last year's plans

Generic pre-review forms get generic answers. Ask AI to turn the brief into a short questionnaire that refers to the client's own plans:

• "Last year you were aiming to stop work at 62. Is that still the plan, or has it moved?"
• "You mentioned your son might start university next September. Is that going ahead, and are you expecting to help with costs?"
• "Has anything changed with your health, or anyone else's in the household, that we should know about?"
• "How did you feel when markets fell in the spring? Did it change how comfortable you are with ups and downs?"

Keep it to eight or ten questions, most answerable in a line. Read every question before it goes out: AI will sometimes reference an event from the wrong year, and a question about a "planned house move" that the client cancelled eight months ago tells them you weren't listening. When answers come back, paste them in and ask for a list of changes against the brief, each marked "confirmed change", "possible change" or "no change". An illustrative result for a single client in his fifties:

The model's habit is to upgrade uncertainty into fact. A client who says "probably" is telling you the conversation is still open, and the agenda should treat it that way.

Two weeks out: commentary around figures you supply

Run valuations and performance from your platform or back-office system as normal. Then give the AI those figures and ask it to explain them, not calculate them:

Write a plain-English explanation of this client's portfolio
year for their review pack. Reading age: a general newspaper.

Figures (from the platform report, do not recalculate):
- Value at last review: [ ]
- Value now: [ ]
- Contributions in: [ ]
- Withdrawals out: [ ]
- Return for the period, as reported: [ ]
- Asset mix now vs target: [ ]

Rules:
- Use only the figures above, exactly as written.
- Do not calculate any new figure. If you need one that isn't
  given, write [MISSING].
- Do not forecast, recommend or say what the client should do.
- Maximum 250 words.

Here's the kind of first draft that shows why the rules matter, followed by the version that went in the pack (illustrative figures). The platform reported a value of $412,000 at the last review, $441,000 now, $24,000 of contributions and a return of 1.2% for the period.

First draft: "Your portfolio delivered strong growth this year, rising from $412,000 to $441,000. This reflects the resilience of your balanced strategy, and you may wish to consider increasing your monthly contributions to build on this momentum."

Corrected: "Your portfolio was worth $412,000 at your last review and is worth $441,000 now. Most of that rise, $24,000, came from the contributions you paid in. The investments themselves returned 1.2% over the period, as reported by the platform, after a fall in the spring and a partial recovery since. We'll talk through what this means for your plans when we meet."

Three problems in the first draft: "strong growth" credits the investments with money the client paid in; "resilience of your balanced strategy" is a claim no figure supports; and the last sentence is a recommendation nobody made. All three read naturally, which is exactly why they'd survive a quick skim.

Then do the check that catches most problems: read each sentence and ask whether a number above supports it. A draft that says "strong growth" for a year that barely beat contributions, or "the portfolio recovered" when it hasn't, is a mis-statement in a client document. If you prefer to keep the maths in a spreadsheet, a reconciliation of opening value, contributions, withdrawals and closing value in Excel takes five minutes and belongs in the pack's working papers.

The week of the meeting: agenda and the questions only you can ask

Ask AI for a draft agenda from the brief and questionnaire answers, with a time for each item. Then add, by hand, the items it can't know about: the conversation you need to have about a beneficiary nomination, the fee you want to explain, the sense from the last phone call that one partner is more anxious than they let on.

For the couple above, the finished agenda might read like this, with the AI's draft in plain text and the adviser's additions marked:

Annual review (75 minutes)
1. Your year: what's changed (15 min)
   - Redundancy offer: decision and timing
   - Mother moving in: costs, care, space
2. Your portfolio year, in plain English (10 min)
3. Retirement at 62: still realistic? (20 min)
4. Wills and the new grandchild (5 min)
5. [Adviser added] Beneficiary nominations on both pensions:
   not updated since 2019 (10 min)
6. [Adviser added] Our ongoing fee: what it paid for this year
   (5 min)
7. Agreed actions (10 min)
[Adviser note, not shown to client: Client 2 quieter on the
last two calls; leave room for her to speak first on item 3.]

Item 6 is worth preparing with AI, carefully. Give it the fee figure from your back-office system and a list of what the firm actually did for the household this year, taken from the CRM: one review meeting, three phone calls, two portfolio rebalances, the beneficiary check, the email about adding their daughter to the savings account. The first draft opened with "Your fee covers continuous monitoring of your portfolio and proactive planning throughout the year." Neither phrase is on the list. The version for the pack names the six things that happened, in the client's terms, and leaves the fee figure exactly as the system reported it. Clients who question a fee are usually asking what it bought; a specific list answers that, and a generic one invites the question.

Two things AI prep can't do. It can't tell you what the client didn't say. And it can't notice that a client's answers have become shorter, vaguer or more dependent on a family member, which may be a vulnerability indicator. Read the questionnaire answers yourself before the meeting, not only the AI's summary of them.

After the meeting: note, actions and the review letter

If you use a note taker, the structured file note comes first; set that up as described in AI note takers for financial advisers. Adviser-built tools such as Jump and Zocks also include pre-meeting preparation features, so if your firm subscribes to one, test its prep output against the brief prompt above before building your own process.

From the signed note, AI can draft the review letter and a list of actions with owners. The letter summarises the meeting and, where advice was given, your recommendation and reasons. Treat that draft with the care in whether AI can draft a suitability report that passes compliance review: every figure checked, every reason one you actually gave.

The letter draft tends to resolve what the meeting left open. For the couple above, the signed note recorded that Client 1 was "still deciding on the redundancy offer; wants to see the figures with and without it before the deadline on 30 November". The AI's draft letter said: "We agreed that taking voluntary redundancy fits your plan to stop full-time work at 62, and we'll update your projections accordingly." That sentence turns an undecided client into a decided one and puts the adviser's name to a recommendation that wasn't made. The corrected paragraph: "You're still deciding whether to accept the redundancy offer. As agreed, we'll send projections with and without it by 14 November, so you can decide before the 30 November deadline." Read every "we agreed" in a draft letter against the note, because that's the phrase the model reaches for when it wants to tidy up an ending.

Running review season across a whole client bank

Review prep scales badly when it's done client by client in a rush. It scales well when it's batched by review month. Keep a simple review calendar (client, review month, adviser, last review date, prep status) and work a month ahead:

1. Week 1 of the month before: build briefs for next month's reviews in one sitting.
2. Week 2: send questionnaires and meeting invitations.
3. Week 3: run valuations, draft commentary, chase missing questionnaires.
4. Week 4: agendas and packs, final adviser read-through.

Batching has one risk of its own, and it shows up on the busiest morning. In an illustrative session building twelve briefs, the fourth brief listed "daughter's wedding in June" under events for a couple with no children. The line came from the third household's records, pasted earlier into the same chat. Nothing in the brief looked wrong except to someone who knew the clients, and the questionnaire built from it would have asked about a wedding that didn't exist. The fix is procedural: a new chat or project for every household, named with the client reference, and the brief's first line reading back the household members so a mismatch is obvious at a glance.

Here's how the numbers can look for a sole adviser with 140 review clients, about 12 a month. Suppose prep currently takes around three hours per review: an hour reading the file, an hour on the pack, an hour on the letter afterwards. With AI drafting the brief, commentary and letter, and the adviser reviewing each, a realistic figure is 90 to 110 minutes. That's roughly 14 to 18 hours a month back, from a tool costing somewhere between $20 and $30 a month on a business plan. The time only materialises if the review steps above are kept short and specific; an adviser who re-reads the whole file anyway saves little.

Where AI review prep misleads advisers

• Last year's facts presented as this year's. The brief says "employed as a teacher" because that's what the records say. It doesn't know she retired in May. Every "current" fact needs the date it was last confirmed.
• Suggestions creeping into client documents. Commentary drafts love a closing line such as "you may wish to consider increasing your contributions". That's a recommendation, and it shouldn't appear in a pack unless you made it. The prompt rule against suggestions reduces this; your read-through removes the rest.
• Conflicting records resolved silently. If one note says the client has two children and another says three, a model may pick one. The prompt asks it to show both; check it did.
• Packs that grow. Because AI makes writing cheap, packs swell to twenty pages. Clients read two. Cap the commentary and put detail in an appendix.
• Figures copied wrongly. Even when told not to calculate, a model can transpose digits when rewriting. Compare every figure in the commentary with the platform report.

If client records are scattered across email, a CRM and paper files, the brief will be only as complete as what you feed it. Fixing where notes are stored is often the first real job; the AI governance checklist for advice firms covers which tools and records should be in scope before you start.

Further reads

• Can Financial Advisers Use ChatGPT? What Compliance Allows — Which assistants and plans your compliance position allows.
• Using AI to Spot Remortgage Opportunities in Your Client Bank — The same client-bank scanning idea applied to mortgage clients.
• How to Turn Meeting Notes Into Tasks Automatically With AI — Turn review actions into tasks without retyping them.
• How to Review AI Call Transcripts for Quality and Compliance — How to sample AI transcripts for accuracy and compliance.
• Claude vs ChatGPT for Business Writing, Proposals and Reports — Which assistant writes clearer client letters and summaries.
• A Simple AI Risk Register for Small Businesses (With Template) — Log the risks of AI-assisted review prep in one place.
• Advice-Specific AI or ChatGPT: Which Should an Advice Firm Use? — What adviser-specific AI adds over a business ChatGPT plan, one meeting written up both ways, and how a four-adviser firm split its budget.
• Paraplanning With AI: What to Automate and What to Keep Human — Twelve paraplanning tasks sorted into automate, assist and keep human, with worked examples of data extraction, chasers and the checks that keep it safe.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Jump pricing page (pre-meeting preparation in the Meet plan); Zocks pricing page (meeting prep on all plans); facts on ChatGPT Business, Claude Team and Microsoft 365 Copilot Business plans checked September 2026.