TL;DR: We read 1,464 recorded client meetings (about 1,300 hours) from 12 UK advice firms and coded 3,822 recommendations with the client's response in the room. Two in five recommendations get no decision, and only 1.5% are declined outright. One adviser behaviour survived every control we applied: when the adviser ties a recommendation to an objective the client has already stated, agreement is 65.6% against 49.1%, a 16.5-point gap (client-clustered 95% CI +12.9 to +19.9). It holds in 11 of 12 firms and in every meeting type, and almost all of it comes from fewer non-committal answers, not fewer refusals. Advisers make this link on 45% of recommendations, from 23% to 67% between individual advisers. After the meeting, a third of money-in-motion moments and half of held-away assets are never followed up. The study is observational and model-coded, so these are associations, not proven causes.

Download the full whitepaper (PDF, 9 pages)

A UK advice firm can report how many annual reviews ran last quarter, how many suitability reports went out and how many files passed compliance. It cannot report what was said in the room: whether the adviser picked up on a need the client raised, or how a recommendation was framed. Meetings held, reviews completed and files checked all measure activity. None of them measures technique.

That gap sits at the point where outcomes are decided. A recommendation the client never decides on still carries the preparation behind it. Under COBS 9.4 a suitability report sets out why advice suits the client's demands, needs and objectives, and the Consumer Duty asks firms to evidence good outcomes. Each undecided recommendation has four to six hours of suitability work already sunk into it.

Advisers are also at capacity. Any lever that asks for more work will not be adopted, so this study looks for behaviours that change outcomes without adding effort. In this paper we:

1. Describe what happens to recommendations in 1,464 real client meetings

2. Test which measurable adviser behaviours are associated with an in-meeting decision

3. Show that raw per-adviser agreement rates are confounded by meeting mix

4. Measure what is heard after the meeting but not followed up

5. Price the cost of undecided recommendations in suitability hours

AdvisoryAI is reading its own customers' meetings through its own pipeline. We state that plainly because a sceptical reader would otherwise assume it.

The study

AdvisoryAI client firms have recorded more than 300,000 client meetings across about 400 UK advice practices (platform base as reported by AdvisoryAI). From this base we drew 1,464 meetings across 12 firms, 34 advisers and 1,212 distinct clients, dated 9 April 2025 to 6 August 2026. Four further firms were excluded: two had less than 12% of their meetings extracted, which is not a random sample of the firm, and two had no extraction at all.

The unit of analysis is the recommendation: a specific course of action the adviser puts to the client in the meeting, paired with the client's response in their own words. Each meeting produced 2.6 recommendations on average. Each response is coded agreed, deferred, declined or unclear (no answer on the transcript), and unclear counts as not agreed. Every recommendation is also coded for whether it was explicitly tied to an objective the client stated earlier.

Firms are split by total adviser headcount into sole advisers (1), small teams (2 to 4) and larger firms (5 or more), four firms each. Adviser and firm identities are replaced with codes, client identifiers were already pseudonymised, and no transcript text leaves the pipeline.

Two in five recommendations get no decision

Of 3,822 recommendations, 56.6% were agreed in the room, 10.5% deferred, 1.5% declined and 31.4% drew no clear answer. The gap is not refusal. It is silence: deferrals and non-committal responses that leave the suitability work behind the recommendation unused.

One behaviour moved the number

When the adviser tied the recommendation to an objective the client had already stated, the client agreed 65.6% of the time (1,136 of 1,732). When the adviser did not, 49.1% (1,027 of 2,090). The effect is +16.5 points, client-clustered 95% CI [+12.9, +19.9], with no bootstrap draw at or below zero. Weighting every firm equally it is +12.9 points, and the conservative firm-level bootstrap gives [+7.4, +18.3]. The effect is positive in 11 of 12 firms. On investment recommendations alone it is +19 points.

The table shows where the effect comes from. Declines barely move. What moves is the unclear share, which falls from 39.1% to 22.0%.

Conditional on an explicit response, agreement is 84.1% against 80.7%, a +3.3 point effect. The behaviour does not mainly make a recommendation more persuasive. It gets the client to decide out loud.

The effect holds in every meeting type. It is widest in first meetings with prospects (55% agreed when aligned against 32% when not) and narrowest in ad hoc conversations, which already run at 64% agreement. It is largest on product recommendations, at 23 points.

Everything else fell away

We tested every adviser behaviour the codebook measures. A behaviour survives only if it holds within meeting type, within firm, and is not explained by reverse causation.

Asking for business reads negative because asks concentrate in new-business meetings, which convert least. Agreeing a next step fails the reverse-causation check: a next step is recorded when something is unresolved, and "no next step" out-converts a vague one (59.5% against 38.9%). Talk time is flat across quartiles of adviser share of words.

By firm size

The effect holds in every segment: +17.8 points in small teams, +16.4 in larger firms and +9.5 for sole advisers. Sole advisers already get more agreement in the room (62.1% against 55% to 57%), so there is less silence left to convert. In every segment the shape is the same: the no-answer share falls and refusals barely move.

Advisers vary, and raw rates hide it

Pooled, 45.3% of recommendations are aligned. The rate runs from 31.9% to 62.6% between firms. Among advisers with 20 or more recommendations, it runs from 23% to 50% inside one firm. By segment the spread is 35% to 50% for sole advisers, 23% to 55% in larger firms and 23% to 67% in small teams. The highest firm rate belongs to a two-adviser practice. The lowest is one of the largest firms.

Raw per-adviser agreement rates mislead in both directions. One adviser agrees 50% of recommendations overall and 64% within annual reviews, because this adviser runs more initial meetings, which land at 45% against 64% for ad hoc conversations. Another reads 42% overall and 56% within reviews. A third reads 50% overall but 38% within reviews. Controlling for meeting mix widens the gap between advisers in that firm from 7 points to 30. This is a composition effect, and a firm reading raw rates off a dashboard would rank its advisers wrongly.

The same blind spot after the meeting

Clients mention money in motion in half of all meetings. Of 1,154 such moments, 762 (66.0%) were followed up. Of 2,636 assets held elsewhere, 1,264 (48.0%) were discussed for consolidation. Of 1,200 needs never quoted for, about one in four was advised on. Follow-up depends on the asset, not on chance:

Three in five clients (752 of 1,212) mention children or grandchildren, and the firm engaged one in ten of those (75). People who plainly need advice are named in about one meeting in three (873 people), while introductions were asked for in 15 of 1,464 meetings (1.0%). Sole advisers asked in none of their 168 meetings.

Manageable assets not followed up

Excluding property, business restructures, spouse accounts and uncategorised items, the unfollowed manageable money in motion is about £3.1m per adviser per year at type medians (£1.2m on stated values only). Undiscussed manageable held-away assets are about £3.2m (£4.0m stated only). These are round figures built from what clients said, not forecasts.

What clients raised, month by month

Across January to July 2026 (months with 25 or more meetings), inheritance tax planning and wills lead every month at 21 to 50 needs per 100 meetings. Long-term care falls from 23 per 100 in January to 9 by July. Protection rises from 6 in April to 14 in July. Planned retirement falls from 55 per 100 meetings to 21 among life events, while health issues hold at 25 to 42.

Against dated events, IHT planning ran at 35 to 44 per 100 meetings in the eight weeks before the 3 March 2026 Spring Statement and near 25 after. Protection rose from 9 to 14 across the change of prime minister (22 June to 20 July) and to 17 in the fortnight after. The Iran conflict, the oil price spike and the March mortgage-rate move left no separate mark. Windows overlap and the winter base is thin, so these readings are directional.

What the gap costs

On product recommendations alone, about 85 to 90 per adviser per year get no decision, carrying 370 to 560 hours of suitability work, or £6,500 to £17,000 at paraplanner rates. For a five-adviser firm that is £33,000 to £85,000 a year.

If every recommendation were aligned at the measured effect, roughly 35 more recommendations per adviser per year would be decided, about 17 of them product recommendations, freeing 70 to 100 hours (£1,200 to £3,100). Revenue per agreed recommendation is not in the transcripts, so we leave it to the reader: at £1,000 per agreed recommendation, 35 decisions would be worth up to £35,000 per adviser.

Model inputs: 4 to 6 hours per suitability report at £17.65 to £30.59 per paraplanner hour. This prices preparation cost, not lost revenue. Per-year figures are upper bounds.

What this means for advice firms

The lever is a sentence, not a skill

The difference between "I'd suggest moving this into a bond" and "you said you want the grandchildren looked after without a tax bill, so I'd suggest moving this into a bond" is one clause. It asks the client to decide on something they have already said they care about. On this data that clause is associated with clients answering at all, and most explicit answers are yes.

It costs no extra work. The client has already stated the objective and the adviser has already formed the recommendation. The same sentence that draws a quotable answer also evidences suitability on the file, so it is worth adopting even for an adviser who never looks at their agreement rate.

The opportunity is visibility

The behaviour appears on under half of recommendations and varies two- to three-fold between advisers in the same firm. Some advisers have already found it. Because raw agreement rates are confounded by meeting mix, a firm cannot see who they are without typing every meeting and attributing every recommendation, roughly 900 conversations a year for a typical firm.

AdvisoryAI's products can capture the meeting (Evie), make the behaviour queryable across a firm (Atlas), and check that the file evidences it (Colin). A firm that makes the behaviour visible will know within a quarter whether its own rate moved.

After the meeting

Follow-up is reliable on what sits in an adviser's own wheelhouse (pension access, old pensions) and weak on what is adjacent (inheritances, cash, property, next generation, referrals). For sole advisers the low referral rate is a capacity ceiling, not a skills gap: asking for referrals only pays once write-up and suitability work stop eating their evenings. For firms trying to grow, introductions are the cheapest new business not being asked for.

Limitations

Critical

• Correlational, not causal. Advisers may tie recommendations to stated objectives most often when the recommendation was always going to land. There is no counterfactual and the product was in use throughout. Every pound figure is an illustrative model, not a forecast.

• Model-coded, not human-validated. Alignment and client response are both judgements from the same model pass over the same transcript. The unclear-to-agreed boundary is the one most exposed to correlated coding error, and it carries most of the effect. A hand check of 100 codings is planned and has not yet been done.

• Provenance. AdvisoryAI is reading its own customers' meetings through its own pipeline. The platform base of 300,000 meetings across 400 firms is as reported by AdvisoryAI.

Notable

• Small, unbalanced firm sample. Twelve firms, four per segment. Two firms carry 42% of recommendations. Five firms have one adviser observed and cannot inform within-firm comparisons.

• Short windows and seasonality. Most firms are observed for about three months in spring review season. Per-year figures are upper bounds, especially for small teams.

• Unclear is large. 31% of responses pooled, and around half in the two lowest-agreement firms.

• Asset values are speech. Stated values are parsed from free text for about half of items and priced at type medians. The figures are orders of magnitude, not balances.

• Time series are thin. Seven usable months, overlapping event windows, and a winter base under 40 meetings for some windows.

• Exemplar assumption. The 12 firms are treated as approximating the platform. That is untested.

Future work. A before-and-after design on the same advisers, a quarter before and a quarter after the behaviour is made visible, with revenue per agreed recommendation from one firm's back office, would turn this into a test of EBITDA impact.

Conclusion

Across 1,464 UK advice meetings, two in five recommendations get no decision in the room, and almost none are refused. Tying the recommendation to an objective the client has already stated is associated with a 16.5-point higher agreement rate, mostly by turning silence into an explicit answer. It holds in almost every firm, every meeting type and every firm size, while every other behaviour we tested fell away. Advisers do it on under half of recommendations and no firm can currently see who does.

The primary implication: make this behaviour visible, then measure whether your own rate, and your undecided suitability work, moves within a quarter.

Download the full whitepaper (PDF, 9 pages)

Frequently asked questions

Which adviser behaviour has the biggest link to clients agreeing a recommendation?

Tying the recommendation to an objective the client has already stated. Across 3,822 recommendations, agreement was 65.6% when the adviser made that link and 49.1% when they did not, a 16.5-point gap that held in 11 of 12 firms and every meeting type.

Why do so many recommendations go undecided?

Mostly silence, not refusal. 31.4% of recommendations drew no clear answer and 10.5% were deferred, while only 1.5% were declined. Linking the recommendation to a stated objective cut the no-answer share from 39.1% to 22.0%.

Do discovery questions, asking for business or talk time make a difference?

Not on this data. More discovery questions read slightly negative, asking for business is confounded by meeting type, agreeing a next step fails the reverse-causation check, and talk time is flat across quartiles.

Can a firm rank its advisers by raw agreement rate?

No. Raw rates are driven by meeting mix. In one firm, controlling for meeting type widened the gap between advisers from 7 points to 30, and some advisers ranked higher or lower than their like-for-like performance.

How was the study done?

Every transcript was read in full by a large language model pipeline applying a versioned codebook, with each coded item carrying a transcript ID, a verbatim quote and its position in the meeting. Confidence intervals are cluster bootstraps resampling clients within firm, 5,000 draws. The design is observational, so results are associations, not causal effects.

Key terms glossary

Aligned recommendation: A recommendation the adviser explicitly ties to an objective the client stated earlier in the meeting.

Agreement rate: Recommendations the client agreed to in the meeting, over all recommendations. Unclear responses count as not agreed.

Unclear response: A recommendation with no answer from the client on the transcript.

Money in motion: A moment where the client mentions money about to move, such as pension access, a bonus, an inheritance or a property or business sale.

Held-away assets: Assets the client holds outside the firm, such as old pensions, other platforms, cash savings or property.

Composition effect: When an overall rate is driven by the mix of groups behind it rather than performance within each group, as with advisers who run more low-converting initial meetings.

Sources

• Cameron, A. C., Gelbach, J. B. and Miller, D. L. (2008). Bootstrap-based improvements for inference with clustered errors. The Review of Economics and Statistics, 90(3), 414–427.

• Efron, B. and Tibshirani, R. J. (1993). An Introduction to the Bootstrap. Chapman and Hall.

• Financial Conduct Authority. FCA Handbook, COBS 9.4: Suitability reports. Accessed October 2026.

• Financial Conduct Authority (2022). PS22/9: A new Consumer Duty.

• Simpson, E. H. (1951). The interpretation of interaction in contingency tables. Journal of the Royal Statistical Society: Series B, 13(2), 238–241.

Authors: Alan Gurung and Roshan Tamil Selvan, AdvisoryAI. Published 22 October 2026.