Part one ended on a puzzle. Three quarters of firms using AI say their people got more productive. Seventy-seven per cent say their revenue did not move. If the tool works, where does the gain go?
I think the answer is in what these tools are actually good at, and what a broking firm actually sells.
It is very good at the typing
Look at what firms report using it for. Over 60% of larger businesses say the main use is improving operations. Not deciding things. Running the same things faster.
That fits what the technology is. These are language models. Give one a pile of words and ask for fewer words, or different words, or the same words in a tidier order, and it is genuinely remarkable. Drafting, summarising, restructuring, first passes. Real hours, really saved.
Now ask it whether this client can afford this loan. It will answer. It will answer in exactly the same confident register it used for the tidying, and that is the problem I wrote about last week: it is never unsure, whether it should be or not.
The typing is not the job. In a broking firm the job is the judgement, and the judgement is the bit that carries your name.
Which is not a sentiment. It is written down.
This is where the regulation stops being background and becomes the actual answer.
The FCA's AI Update says any use of AI in relation to an activity or business area of a firm falls within the scope of a Senior Manager's responsibilities. That is paragraph 3.40, and read it carefully, because the mechanism it describes, SMF24 and SMF4, applies to dual-regulated and Enhanced firms, and it says in terms "but not Core or Limited Scope SM&CR firms". If you are a solo-regulated broker it is easy to read that paragraph and conclude none of this reaches you.
Read the next one. Paragraph 3.41 says all Senior Managers in SM&CR firms, and it names solo-regulated Core and Limited Scope firms explicitly, must have a Statement of Responsibilities, and are subject to the Conduct Rules requiring them to take reasonable steps to ensure the business for which they are responsible is effectively controlled.
There is no version of that sentence where the tool becomes responsible. You can hand a machine the drafting. The reasonable steps stay with a person, and that person has a name on a document.
The Consumer Duty makes the same move from the other direction. Paragraph 3.43: a firm's board should review and approve an assessment of whether it is delivering good outcomes, evidenced with data. Evidenced. Not felt, not assumed because nobody complained.
So the ceiling is not technical
This is what I think the money has got wrong.
The valuations assume a technology that replaces the work. The regulation describes one that can only ever assist it, because the accountable part cannot be transferred to something that cannot be accountable. Not "should not". Cannot. There is nobody to name on the Statement of Responsibilities.
That is not a UK quirk or a temporary drag. It is the shape of the thing.
And it explains the revenue puzzle better than any story about implementation. If the part you can automate is the typing, and the typing was never what the client was paying for, then automating it makes you quicker without making you richer. Exactly as three quarters of firms reported.
What that means for anyone waiting to be replaced
The two brokers who asked me this were not being dramatic. It is the question everybody is quietly holding.
My honest view is that the thing coming for brokers is not a machine that does the job. It is other brokers who worked out what to point it at, while the rest were still deciding whether to be frightened of it.
The numbers say almost nobody has worked that out yet. Adoption tripled and depth barely moved. Half of firms report no change in headcount at all.
That is not the calm before the replacement. It is a wide open field, with a great deal of money betting loudly on the wrong outcome.