AI Software for Wealth Management: A Small Firm's Evaluation Checklist

Check six things before buying AI software for wealth management: where client data is stored and whether it trains models, independent security evidence, how notes reach your compliance archive, whether an adviser must approve everything before it reaches a client or the CRM, accuracy of prep packs, notes and reports on your own meetings, and the full cost including archiving.

Most small firms evaluate on the demo, which shows the best case. Run it in three passes instead: knockout questions on a 20-minute call, tests on your own meetings and reports, then the full cost. Prices below are list prices from the vendors' pricing pages, re-checked on 4 November 2026. Those pages change, so confirm every figure in your quote.

Follow me on Instagram@sagnikteaches

Three passes: knockouts first, then job tests, then the full bill

The order saves time. Pass one takes about 20 minutes per vendor and removes anything that cannot meet your data or compliance needs, so you never spend a week testing a tool you cannot buy. Pass two takes two to three hours of adviser time per candidate and measures what matters to clients: whether the prep pack, the notes and the report text are right. Pass three is an hour with a spreadsheet and turns list prices into a year-one cost that includes the extras sales pages tend to leave out.

Connect on LinkedInSagnik Bhattacharya

Bring three people to the first call if you have them: an adviser who will use it, whoever runs operations, and whoever signs off compliance. Questions that sound technical to one of them are usually obvious to another.

Subscribe on YouTube@sagnikteaches

Pass one: eight knockout questions for a 20-minute call

1. Does it record audio, or only transcribe, and where do those files live? Zocks' pricing page lists "no meeting recordings" on all three of its plans. Jump's security page speaks of meeting recordings, transcripts and summaries, so ask which of the three it keeps in your set-up and for how long.
2. Where is the data stored, and can you choose? Zocks' pricing page states that data is stored in one fixed country and treats storage in a custom geography as a contact-us item. Jump's security page says data, backups included, is stored in the region you designate. Neither answer is wrong; what matters is whether it fits your clients and your regulator.
3. Is client data used to train models? Jump's security page says meeting recordings, transcripts and summaries are never used for training or fine-tuning. The Zocks pricing page I read does not address training, so ask for the answer and the contract clause. Wealthbox's pricing page, as I read it, makes no statement on how customer data is used for AI either.
4. What independent security evidence exists? All three vendors say SOC 2 Type II on their own pages. Jump describes an independent assessment across a continuous 12-month period. A claim is not a report: ask for the report, the period it covers and any exceptions, and check that the AI features are within scope.
5. How do notes reach your archive, and what does that cost? Zocks' pricing page lists archive integrations (Smarsh, Proofpoint, Global Relay and an alternative archive) against its Professional and Ultimate plans rather than Essentials, though its comparison tables do not agree on every row, so confirm in the quote. Jump's pricing page, as I read it, does not mention archiving. Wealthbox lists "Compliant Note Archiving" among its compliance features without showing per-plan detail in the text I could read. Whichever you choose, add your archive provider's own fee to the bill.
6. Does anything reach the CRM or a client before an adviser approves it? Jump's security page says nothing reaches a client or system of record until an advisor has reviewed and approved it. Whatever the vendor promises, try to break the rule in the trial by letting a note sit unapproved and seeing whether it syncs anyway.
7. How does it evidence consent? Look for per-client tracking of who was told, a way to switch capture off for one meeting, and an admin view of the record. Jump's security page says customers can manage recording disclosures and consent. Ask every vendor to show the screen, not describe it.
8. What do you get on the way out? Notes, transcripts, action items, templates and the audit trail, in which format, within how many days, and how deletion is confirmed in writing.

Set one rule before the call: a "no" or "we'll come back to you" on questions one to six pauses the trial until the answer arrives in writing. Questions seven and eight can wait until pass three.

Pass two: test meeting prep, notes and reporting on your own work

Run the same six meetings through each candidate: two routine reviews, two complex ones with several topics and a lot of numbers, and two messy ones. Messy means a couple talking over each other, product names that sound alike, and an acronym nobody defines. Until pass one is cleared, use role-played meetings with colleagues acting as clients, so no real client data goes in. Where a tool accepts an uploaded audio file, feed the same file to both candidates; otherwise run the two tools side by side on separate devices.

These pass marks are my starting suggestions, not an industry standard. Tighten them if your compliance lead wants zero tolerance on figures, and write the final numbers down before the first test so nobody adjusts them after seeing a favourite fail.

Catching what an AI notetaker gets wrong with figures and advice

Two failure types cause the most compliance trouble: advice attributed to the wrong person, and figures that are close but wrong. The first shows up like this in an illustrative test meeting.

That draft records advice the adviser never gave, and it reads fluently, which is why it survives a quick skim. Train reviewers to stop on every sentence containing "recommended", "advised", "agreed" or "decided" and check who said it. For the second type, use a number check that works with any assistant, including the one inside your CRM:

Below are (1) a draft paragraph of client report commentary and (2) the figures from the portfolio system for the same period. List every number, percentage and date in the draft. For each one, show the matching figure from the system or write NOT FOUND. Do not correct anything and do not calculate new figures.

DRAFT:
[paste the commentary]

SYSTEM FIGURES:
[paste the table]

A realistic sample output, with the problems it should surface:

Three mismatches in five figures is an exaggerated sample, but the transposed cash figure is the kind of slip people really make. Do the sum on your own volumes. If a note carries 25 figures and the tool gets 98% of them right, the expected number of wrong figures is 25 x 0.02 = 0.5 per note, and about four notes in ten (1 minus 0.98 to the power of 25) will contain at least one. That is why the approval step is not optional, and why a tool that writes numbers from structured data beats one that retypes them from speech. For the general version of this problem, see catching wrong owners and invented decisions in AI meeting notes.

Prep packs and report commentary: allow drafting, not calculating

Prep is where AI earns its keep most visibly and where an unchecked line does the least harm, provided every line names its source. A good pack for a review meeting is a one-page list the adviser can verify in three minutes. In the illustrative extract below, each row says where the fact came from, so a checker knows where to look.

The last row shows the rule to enforce: a line that cannot name its source is a draft, not a fact. Report commentary needs the same discipline, plus a ban on arithmetic. Let the assistant phrase what the system has already calculated, and tell it so:

Draft two short paragraphs of client report commentary using only the figures below. Do not calculate or estimate any figure, do not mention periods I have not supplied, do not forecast, and avoid words such as strong, safe or guaranteed. If something cannot be said from the data, write FLAG and say what is missing.

FIGURES:
12-month time-weighted return 6.7%; benchmark 6.1%; contributions $24,000; equities 60.4% against a 60% target; period end 30 June.

A realistic output, and the edit it needs, would read as follows. Draft: "Over the 12 months to 30 June the portfolio returned 6.7% against 6.1% for the benchmark. Contributions of $24,000 were invested and equities ended the period at 60.4%, close to the 60% target. FLAG: no figure supplied for volatility, so no statement on risk is made." Before it goes anywhere, the adviser confirms the benchmark's name and the return period, and whoever reviews performance wording at your firm reads the sentence comparing the portfolio with the benchmark. The FLAG line is the behaviour to look for in a trial: a tool that fills the gap with a confident paragraph about risk fails the test, and one that leaves a visible hole passes it.

When the AI sits inside your CRM: what to ask before adding it

Built-in AI is often priced separately from the plan it lives in. Wealthbox's pricing page lists Basic at $59, Pro at $75 and Premier at $99 per user per month, and shows its AI Notetaker as a $49 per user per month add-on on every plan, so a Pro seat with the notetaker comes to $124. A separate Vault add-on is $29 per user per month. For a deeper comparison of the add-on against standalone tools, see the Wealthbox notetaker cost comparison, and for the wider built-in versus bolted-on question, what is native and what is an add-on in adviser CRMs.

• What can the AI see? It should see only what the signed-in user can see. Test it with a restricted client record.
• What can it write? List the fields it fills, whether it overwrites existing entries, and whether AI-written entries are marked as such in the audit trail.
• Can you switch it off per user? You will want that for associates before they are trained.
• Does it act on its own? Overnight task creation and "next best action" features need the same approval gate as notes.

An illustration of why the last point matters: a CRM feature creates follow-up tasks overnight from a meeting summary. On Monday an associate phones a client about a pension transfer that was only discussed as a hypothetical. Nobody approved the task because nobody knew the feature was on. The fix was a setting, and a line in the approved-use statement below.

Pass three: the full cost for a five-person firm

Price the first year at list for a firm with three advisers and two client-service associates, using annual billing where the vendor offers a lower rate. Prices come from Zocks, Jump and the Wealthbox page above, all read on the same day.

Now put a price on a meeting. Assume each adviser holds 10 client meetings a week for 46 weeks, which is 460 a year, or 1,380 for three advisers. The Zocks Professional total of $4,812 comes to $3.49 a meeting, Jump at list comes to $2.61, and the Wealthbox add-on to $1.28. Value an adviser's time at $120 an hour, which is $2 a minute, and the break-even is 1.7 minutes saved per meeting for Zocks Professional, 1.3 for Jump and 0.6 for the add-on. The licence is rarely the real cost. The real costs are the review minutes, the archive fee and a compliance process nobody has written down.

Add the lines that rarely appear on a pricing page: the archive provider's quote, template set-up time (budget a few hours per adviser), staff training, a price cap at renewal and any usage limits. Ask whether the price you are quoted is locked for the first 12 months. A tool that costs $1,764 a year and takes 30 hours of compliance time to approve is not the cheapest option.

Reading the results: a decision rule for two candidates

Two candidates, X and Y, were each run on the same 12 role-played meetings. These figures are illustrative.

Apply the rules written before the tests. Any invented recommendation is an automatic fail until the vendor explains and fixes it, so X fails. Wrong figures must stay under 1 in 100: X is at 2.7%, Y at 0.7%. Actions captured must reach 9 in 10, meaning at least 39 of 43: X captured 42, Y captured 40 (93%), so both pass. Phantom tasks must stay at one or fewer in 12 notes: X had five. Y passes all four rules. Then check time. If your manual notes take 15 minutes, Y saves nine minutes per meeting and X six, and both clear the 1.3 to 1.7 minute break-even above comfortably. X's higher capture of real actions is real, but it does not outweigh recommendations that were never made.

A one-page approved-use statement your compliance lead can sign

Writing the statement forces the decisions the tool will not make for you. Keep it to a page, name an owner, and review it when the vendor changes anything material. Wording that works for a small firm:

AI MEETING NOTES: APPROVED USE (reviewed [date], next review [date])
Tool and plan: [name], [plan]
Who may use it: advisers; associates only with an adviser's approval
Clients told how: a sentence in the meeting invitation, repeated at the start of the meeting; opt-outs recorded in the CRM
What the tool captures: transcript and summary; audio [kept / not kept]
Review step: the adviser reads and edits every note and approves it before it is saved to the client file or sent to anyone
Automatic features: [task creation / next best action] switched [on / off]
Where notes live: client file in [CRM]; copy sent to [archive provider] within [24 hours]
Retention: [period agreed with compliance], then deleted from the tool and the archive
Never: record a client who has opted out; paste notes into any other AI tool
Owner: [name]. Compliance sign-off: [name], [date]

A short disclosure sentence for invitations: "We use an AI note-taking tool to write up our meetings. An adviser reviews every note before it is saved to your file. Tell us if you would rather we did not." Whether you must say more depends on your regulator and on whether audio is recorded, so check with your compliance lead. Whether AI summaries and transcripts count as records you must keep is the same kind of question; a tutorial on archiving AI notes and chats sets out what to ask, and a comparison of two notetakers on compliance fit shows how the answers differ between tools.

Warning signs during the trial

• The vendor will not share a SOC 2 report or its scope, or answers only with a badge.
• Notes reach the CRM or a client email before anyone has approved them.
• The archive link is "coming soon", or works only on a plan you were not quoted.
• Demo data is canned, and every request to try your own template is postponed.
• Numbers in reports are retyped by the AI rather than pulled from the system.
• Pricing changes between the demo, the quote and the contract, with no cap at renewal.
• Reviewers start approving notes in under a minute. That usually means the review has become a click, so sample a few approved notes against the transcript each week.

The last point deserves a number. If reviewers approve notes in 40 seconds that took 6 minutes in week one, the errors have not vanished; the checking has. Re-run the number check on five approved notes every week for the first month, and keep the results with your approved-use statement as evidence that supervision is happening.

Questions advisers ask before a trial

Do clients have to agree before we use an AI notetaker?

The rules depend on whether audio is recorded, where participants are and what your regulator expects, so ask your compliance lead rather than guess. In practice most firms tell clients in the meeting invitation, repeat it at the start of the meeting, let clients opt out, and keep a record of both. Choose a tool that can show that record per client.

Should associates get their own notetaker seats?

Usually not at first. Zocks' pricing page lists a dedicated admin assistant seat at $25 a month, limited to five meetings a month, while Jump's page prices per adviser, so ask each vendor how support staff are licensed. Start with advisers, get the approval flow working, then add support staff once you know how many meetings they attend.

How long should the trial of an AI notetaker run?

Two to three weeks is enough if you plan the tests. Spend week one on role-played meetings with colleagues, which avoids live client data before the contract and consent are settled. Spend weeks two and three on real meetings with consenting clients, then count errors and review minutes before anyone sees a renewal quote.

Is a CRM's built-in notetaker enough, or do we need a standalone one?

Run both through the same tests. A built-in notetaker keeps notes beside the client record and avoids a sync to maintain, but it may be an add-on, as on Wealthbox's page at $49 per user per month on top of the plan, and its archiving detail needs confirming per plan. A standalone tool adds a sync but may offer different compliance features. The pass-two results should decide.

Further reads

• Zocks Pricing for Financial Advisers: What to Ask Before You Sign — Questions to put to one vendor about price and archive links.
• Can AI Write File Notes From Adviser Meetings? Draft, Check, Sign Off — A draft, check and sign-off routine for file notes.
• AI for Wealth Management Firms: A 90-Day Rollout Plan for Small Firms — A rollout plan once you have chosen a tool.
• Jump AI Set-Up for a Three-Adviser Firm: Templates, Consent, Review — Templates, consent and review set up in one tool.
• AI in KYC for a Small Wealth or Advice Firm: Checks That Stay Human — Which onboarding checks should never be left to AI.
• AI Agents for Financial Advisers: What They Can Do Without Advising — Where agents help with admin without straying into advice.
• AI Note Takers for Financial Advisers: Compliant Meeting Records — A wider look at compliant meeting records.
• Best AI App for Wealth Management: Jump, Zocks and Wealthbox — Jump, Zocks and the Wealthbox notetaker compared on price, recording and archiving, with a break-even sum and a pilot a small firm can run.
• FP Alpha vs Holistiplan: Tax Report Tools for Planners Compared — Per-household against per-credit pricing, a feature table from both vendors' pages, and a test you can run on one synthetic return before paying.
• Holistiplan Pricing: Basic vs Premium Tax and Who Needs Which — Holistiplan Basic Tax vs Premium Tax priced by household tier from 30 to 750, with a Premium break-even, the billing rules and a 7-day trial checklist.
• AI Tools for Asset Management: Factsheet Commentary With Sign-Off — A month-end workflow for factsheet commentary: source-only drafting, a check on every figure, and a named approver who signs a log.
• AI Tools for Investment Managers: Client Reports and Fact Sheets — How a small investment manager can draft quarterly reports and fact sheets with AI while the numbers stay locked and compliance gets a reviewable packet.
• AI Tools for UHNW Wealth Managers: Seven Jobs With Privacy Rules — A job-by-job guide to AI in a small wealth team, including access limits, sample outputs and review gates.
• AI for Family Offices: Reporting, Document Filing and Privacy Limits — How a two-to-ten person family office can use AI for filing and reporting commentary, with a privacy tier system that decides what the tools may see.
• Best AI Tools for Investment Management: Committee Packs and Minutes — Which AI tools help a small investment firm assemble committee packs and draft minutes, with the record-keeping and vendor-terms checks that stay human.
• AI in Wealth Management Examples: Five Admin Jobs Worth Trying — Meeting notes, follow-ups, review packs, onboarding checklists and newsletters: five admin jobs a small wealth firm can give AI, plus the records caveat.
• Can AI Write an Investment Proposal? Draft Narrative, Check Numbers — A practical adviser review of statement extraction, portfolio arithmetic, proposal wording and sign-off.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the pricing pages of Zocks, Jump and Wealthbox and the security pages of Zocks and Jump, re-checked on 4 November 2026.