That line is the entire design of this deployment. A language model asked which fund to buy will answer, because answering is what it does, and on a recorded call from a regulated firm that answer is the firm's advice.
So this one does not advise. It understands goals, horizon and broad risk comfort, books time with a qualified advisor, and is explicitly forbidden from recommending, projecting or reassuring.
The risk in this category runs in the opposite direction from most: the danger is saying too much, not too little.
An unqualified person fielding a question about returns is a compliance incident waiting to be recorded, and it happens because the enquiry needed answering and nobody else was available.
The most expensive person in the firm collecting goals, horizon and KYC status is the least efficient possible use of a qualified hour.
“It is basically safe” is the kind of thing people say to be kind. In this category it is a mis-selling exposure with a paper trail.
Investment decisions are deferred rather than declined, routinely for months. Almost nobody has a system that comes back at the right time.
A complete fact-find, so the advisor's first call starts at the useful part.
Three questions in this exchange are requests for advice. All three are declined and routed.
You had enquired about investment options on our website, so I am following up on that. Is now convenient?
Yes. Which fund should I put my money in?
That is genuinely a question for our advisor rather than for me, and I would be doing you a disservice by having a view on it. What I can do is understand what the money is for, so their time with you is useful. Is there a particular goal behind it?
Retirement mainly. What kind of returns can I expect?
I cannot project that. Returns are market-linked and not guaranteed, and anyone giving you a figure on a first call is guessing. Roughly how many years away is retirement for you?
About fifteen. Is it safe though?
It carries risk, and how much depends on the category, which is exactly what the advisor will take you through properly. Would Thursday evening work for a call with them?
No recommendation, no projection, no reassurance. Three refusals and a booked advisor call.
This list is not a feature. It is the reason the deployment is defensible.
It never gives personalised investment advice or recommends a scheme. Every such question is routed to a qualified advisor, explicitly and by name of role.
It never projects returns, and never describes anything as guaranteed or risk-free. If returns come up at all, it states that they are market-linked and not guaranteed.
It never invents NAVs, past performance, expense ratios or AUM figures. Nothing numerical enters the call unless you supplied and approved it.
General-purpose assistants are trained to be helpful, and helpfulness in this domain looks exactly like advice. Asked which fund suits a fifteen-year horizon, an unconstrained model will produce a reasonable-sounding answer. It will be reasonable-sounding on a recording, from your number, to someone who then acts on it.
The fix is not a better model, it is a narrower brief. This persona is defined by what it refuses: no recommendation, no projection, no characterisation of risk beyond routing it to a human. Those refusals are written out with the wording approved, so they are consistent on call four hundred as on call one.
The commercial argument runs the same way as the compliance one. A fact-find completed before the advisor picks up the phone means the qualified hour starts at the useful part. The agent is doing the work an advisor should not be doing, and stopping short of the work only an advisor may do.
Every persona has its own script, its own vocabulary and its own list of things it will not say.
The agent does not advise, recommend or solicit a specific product. It handles enquiry follow-up and fact-finding and books a call with your qualified people. Where your compliance function requires specific disclosures or recorded consent on outbound calls, those are written into the brief before launch. We build to your compliance team's wording rather than to our reading of the rules.
It keeps declining and keeps routing. It is not instructed to relent under pressure, which is a meaningful difference from a junior human handling the same call.
Only figures you have supplied and authorised, with their source and period. It will never estimate, and it will never present past performance as an indication of future returns.
Those are usually better handled by the insurance persona, which has a different objection set and different disclosure requirements. We would scope which one fits at the consult.
Yes, with full transcripts on the lead record and retention set to your policy. In a regulated category the record is generally an asset.
It will decline, twice if you push, and book you an advisor instead. That refusal is what you are actually buying.