BlackRock reports that 68% of advisory firms now use AI in some capacity, and roughly half of them are still piloting. An adoption percentage measures how many firms bought something. It tells you nothing about what any of them changed. The number that matters is cost to serve: what it costs your firm to look after 1 client for 1 year, because that number quietly decides who gets a yes and who gets turned away. Watch that figure, the minimum portfolio size at which a new client becomes economical, and the hours your team spends on work that never touches human judgement.
Key takeaways
- Adoption is an input, cost to serve is the output. A tool sitting in a workflow and a tool restructuring the economics of that workflow are 2 very different states of the world.
- Advisers already know where the leverage is: back office automation scored 7.72 out of 10 for expected impact in the T3 survey of 2,906 advisers, client facing AI scored 4.37.
- Saving time is a cost story. Changing who you can serve is a growth story. The same capability produces both, and the difference is what you do with the recovered capacity.
- The UK advice industry serves around 9% of the population. Reworked properly, the economics support closer to 25%.
- Exactness stays where exactness is guaranteed. A language model is the wrong tool for exact financial computation, and nothing should leave the firm without a human sign-off.
What does an adoption percentage actually measure?
I have watched the 68% figure circulate through the industry as proof that wealth management has crossed into its AI era. I read it differently.
When I dig into the data behind the headline, the picture gets more honest. Roughly half of the firms using AI are still in piloting stages, kicking the tyres rather than rebuilding the engine. A tool sitting in a workflow and a tool restructuring the economics of that workflow are 2 very different things, and most of the industry is currently in the first one.
I came into UK wealth management as an outsider, a mechanical engineer by training, and the first thing I did was follow the unit economics down to the single client. That exercise reshaped how I read every AI headline in this industry. The number that matters is what it costs a firm to serve 1 client for 1 year, because that number quietly decides who gets a yes and who gets turned away.
At Avagance, we have modelled service delivery costs dropping from £237 per client to £34. That reduction is the actual story.
When serving a client costs more than that client can reasonably pay, the firm declines the relationship, and the person walks away without advice. The cost structure made that decision, and nobody in the room said it out loud.
I know this personally. As a young founder, I could not access financial advice because my situation was not worth the admin it would generate. The maths excluded me before any human made a judgement call. That experience is why I treat cost to serve as the central metric of this entire transition, and why adoption percentages leave me unmoved.
The UK advice industry currently serves around 9% of the population. The economics, reworked properly, support 25%.
That gap is where the 68% figure becomes interesting, because it tells us the tooling is now in the building. The question is whether firms will use it to trim minutes or to rewrite who they can afford to say yes to.
Where is the work actually moving?
The pattern I keep observing is that advisers themselves already understand where the real leverage sits. The T3 and Inside Information survey of 2,906 advisers scored back office automation at 7.72 out of 10 for expected impact, while client facing AI scored just 4.37. Practitioners expect the machine to change the work behind the scenes far more than the work in front of the client.
I find that instinct exactly right, and I will say the uncomfortable part plainly. The adviser is irreplaceable. The admin is not.
Human judgement, the ability to sit with someone through a difficult decision, the earned credibility of being believed, these things stay scarce no matter how capable the models get. Suitability reports, compliance documentation, meeting prep, data re-entry across 5 disconnected systems, none of that requires human judgement. It requires accuracy and throughput, and those are automation problems.
None of it removes the sign-off. A draft is still checked and approved by a person, because the responsibility sits with the firm and always will.
This is already happening in the field. Some firms have phased out the paraplanning role entirely, with work that took 4 hours now completing in minutes around client meetings. I understand why the industry avoids saying this aloud. I think avoiding it does everyone a disservice, including the paraplanners, who deserve an honest read on where their role is heading rather than reassurance that expires.
Why efficiency that stays efficiency is a wasted asset
Here is the distinction I would encourage you to test against your own firm. Saving time is a cost story. Changing who you can serve is a growth story. The same automated capability produces both outcomes, and the difference is entirely in what you do with the recovered capacity.
The Kitces data shows this compounding in practice. In 2022, firms with 1 support hire serviced 86 clients on roughly $517,500 in revenue. By 2024, similar firms managed 111 clients and $591,000 with the same team. Same headcount, more relationships, more revenue. The capacity ceiling moved.
Now extend that curve. When your cost to serve drops by an order of magnitude, the client segments that were previously uneconomical become viable. The 30 year old with £40,000 and a pension question. The family below the wealth threshold your fee model quietly enforces. These people want human advice, and until now the maths said no on your behalf.
The BlackRock report points at the same opportunity from the demand side. AI tools that analyse trusts and tax documents, model scenarios, and surface issues early let you bring sophisticated planning conversations to clients who would never have justified the manual hours. Advanced planning stops being a premium reserved for the largest accounts and becomes something you can deliver across your entire book.
This matters enormously for the generational wealth transfer already underway. Younger inheritors expect digital fluency and they expect a relationship. The firms that lowered their cost to serve early will be positioned to start those relationships years before the assets arrive. The firms still piloting will meet those clients after someone else already did.
What constraints deserve respect?
I want to be careful here, because I have zero patience for utopian tech worship in a regulated industry.
Not every task belongs to a probabilistic engine. A language model that generates plausible text is the wrong tool for exact financial computation, and pretending otherwise creates regulatory exposure that no efficiency gain justifies. Exactness stays where exactness is guaranteed, and the model handles the work that tolerates it: absorbing dense documents, drafting, modelling scenarios for human review.
Infrastructure choices carry the same weight. When we built Avagance, wrapping someone else's API would have been faster to market. We chose to run open source models on our own servers instead, because in a regulated industry, controlling your data infrastructure is the difference between a wrapper and a platform. Client financial data moving through third party endpoints is a structural fragility, and I do not build on structural fragility.
I will also admit what I do not know. Nobody has fully mapped how the FCA will treat autonomous compliance workflows at scale, and anyone projecting certainty there is selling something. I would rather explore that terrain openly with the firms living inside it, which is why we recruited founding advisers directly from our customer profile when a reviewer showed me my blind spots. Outsider perspective questions the constraints. Insider expertise tells you which constraints are real.
What should you watch instead of the adoption number?
If you run a firm, I would suggest tracking 3 numbers over the next 24 months.
- Your fully loaded cost to serve 1 client for 1 year.
- The minimum portfolio size at which a new client becomes economical.
- The number of hours your team spends on work that never touches human judgement.
Those 3 numbers tell you whether AI changed your business or merely entered it. The 68% statistic tells you neither.
My conviction is that this transition rewards patient, compound improvement over dramatic overhauls. You plant the automation in 1 workflow, water it until it holds, then extend it to the next. Firms that rush a full transformation in a quarter tend to break their compliance processes and retreat. Firms that build the operational layer piece by piece end up with something structurally sound.
The endpoint I am building toward is an industry where a human adviser economically serves people the current model excludes. Where the cost structure stops making the decision and the adviser makes it instead. The tooling for that future already exists. The 68% proves firms are willing to try it. What remains unproven is how many will follow the economics all the way down to the client who used to get a no.
So here is what I keep turning over, and I would genuinely like your answer: when your cost to serve a client drops by 10x, who will you say yes to that you could not before?
Frequently asked questions
What does the 68% AI adoption figure actually mean?
It means 68% of advisory firms report using AI in some capacity, which is a measure of purchase rather than change. Roughly half of those firms are still piloting. Adoption tells you the tooling is in the building. It does not tell you whether any workflow, any cost line or any client decision is different as a result.
What is cost to serve in a financial advice firm?
It is the fully loaded cost of looking after 1 client for 1 year: adviser and support time, software, compliance, file checking and the admin nobody bills for. It matters because it sets the minimum portfolio size at which a new client becomes economical, which is the real, unspoken rule about who your firm can say yes to.
Will AI replace paraplanners?
The honest answer is that the admin is automatable and the judgement is not. Some firms have already phased out the paraplanning role, and work that took 4 hours now finishes in minutes. Others are redeploying that capacity into client work and more cases per adviser. Either way the checking and the sign-off stay human, because the responsibility sits with the firm.
Where does AI belong in an advice firm, and where does it not?
It belongs where the work tolerates review: reading dense documents, drafting, preparing meetings, surfacing issues early. It does not belong in exact financial computation, where a plausible number is worse than no number. Keep deterministic maths deterministic, and keep a human sign-off on anything that reaches a client or a regulator.
How do I measure whether AI actually changed my firm?
Track 3 numbers over 24 months: fully loaded cost to serve 1 client for 1 year, the minimum portfolio size at which a new client becomes economical, and hours spent on work that never touches human judgement. If the first 2 fall, the economics changed. If only the third falls, you bought time back and spent it on the same clients.