Introduction

Thanks for all being here this evening.

Who I Am and What This Talk Covers

A bit of background. I'm an independent financial advisor based in Bath, so I run my business there, and I cover the Bristol Bath area.

I thought I'd just quickly show this to start with before I really go into the talk. For those of you a bit more technical in the room, this is something I've been building for maybe a month or two primarily with Claude Code.

I've got a background in maths so I have I know I'm wearing a suit maybe look a bit economicsy but I have done some coding and that's helped with this which is put simply a guard railed financial advisor so if you ask it stuff like should I invest in 10k in Tesla stock it will hopefully tell you have you thought about paying off a his student loan first um so yeah just a quick one so if you've got any questions about that

more advanced stuff please feel free to come reach out at the end um the reason i mentioned that is because the talk i've got today is really at the base level i'd say so hopefully everyone here can take something away and yeah put it into action this week and really yeah i'll start with

Why This Talk: Getting Value for Money From AI

really covering why i wanted to put together this talk and it's really about getting value for money so currently as we'll see in a bit I think there's a lot of value you can get back using LLMs at the moment no matter your technicality and yeah we'll address

that now so I'll start with a bit of something I noticed from my work and as you'll notice it doesn't just apply to wealth management financial advice which what I do but also solicitors accountants the kind of people that all all of us will probably encounter at least once in our adult life.

The Promise of LLMs for Everyday Questions

So yeah, I'll just start with the conflict here between what AI, normal LLMs that we all have in our pockets can do right now, which is basically the benefits.

It's free, always good. I know some of us do pay, but it is fundamentally free. You can ask it tons of stupid questions, which you may not want to ask a financial advisor or solicitor,

maybe like, what's an ISA? or um and we'll come on to the next question uh slide in a second which is basically saying you don't want to ask those questions because you may be paying 300 quid plus um for their time

so why spend five minutes on that um but then of course there's the negatives so for example

The Risks: Errors, Accountability, and Regulation

i was reading this morning in preparation for this talk about this unfortunate lady i think think she lived up near Huddersfield she was earning 99k lucky person gross per ring for annum and she asked okay have I got a salary increase this year would I lose my basically your child free benefit which for those of you who don't know in the room that stops after 100k

basically meaning that you don't get child care for free in the UK now that is if you think about that that's a massive thing to lose for people that's hours of time free with their kid um and

the ai said no this increase wouldn't um you wouldn't lose your child free um allowance or anything like that so that's just a little example there there's no accountability so if the uh i shouldn't swear so the sh it hits the fan you can't go and sue sam altman or whatever um that

sayer so this brings us on to the biggest benefit about the human domain so someone like me is i i need to pay a lot of money to make sure i'm compliant and that if something goes wrong then you can basically tell come after me and get your money back so lots of us in the room

may have a cash isa with some bank that wasn't around 20 30 years ago tembo trading 212 someone like that and the reason you feel safe doing that is because you are protected most of the time up to 125k as of this tax year um so yeah that's great um of course the unfortunate thing is where

The Cost of Human Advice—and Where It Falls Short

we've already come on to now is the cost of the human um which is one of the reasons yeah lots of traditional meetings i'd say now um lots of admin data gathering um and you're paying 300 or 400 hundred quid at least um for that kind of thing so you may be i don't know i might get have an

annual review with you for example so we're looking at how your investments have performed over the last year um and i'm just saying oh yeah your isa has gone up by five or six percent um this is what's happened this is xyz and sort of guiding you through the report is that giving you any value not particularly it might be giving you some um but not as much

as it could be and this is really where it's coming on to next which is how you can use AI to really help you get value for money and yeah well I'll show you a little thing which only really takes about five minutes to basically promote your solicitor advisor from an intern into sort of a senior partner is there what I came up with so yeah let's crack on basic yeah because none of us want

on admin. Maybe the council do.

A Practical Workflow: Using AI to Prepare for Professional Meetings

Let's come on to how we can build something to actually help you.

You can take a picture of this. I'll show a QR code at the end and with a link to a few different documents which include this stuff.

I'm going to try and show you how to use this initially.

Stage 1: Build a Role to Improve Outputs

Stage one is basically building the role.

I tend to use, who here uses grok copilot deep seek basically what i'm saying is not claude gemini chat gbt you want to save your tokens if you don't know what tokens are there your usage probably for the important tasks and the sort of complex stuff which is why most of us in the room probably use claude gemini chat gbt so you'll see on copilot i never use copilot apart from for this and i have a whole

list of basically that step one which is building the role so the first step on here as you see list 20 words describing an XYZ specialist so I've done this on ChatGPT although I would advise you basically to do this step on something which you don't normally use so yeah here we go trustworthy knowledgeable Knowledgeable, strategic, all those good words which hopefully you may associate with me.

And then, yeah, I've got this prompt here which I normally keep literally in that folder there and I'd advise you just to keep that part somewhere easily accessible because I'm not expecting anyone here to remember all of those words in that exact sentence. But yeah, you see what it comes up with and suddenly you've got here your role.

and this is so that's the end of step one so you can see here you've now got your role and what we can do then is with the next prompt put in that role which moves us on to

Stage 2: Summarise the Document in Plain English

oh a bit too far there we go so the task so in the example that we used um i said an annual review is the kind of thing which you might be able to get a lot more value out of and i think for most people um here that you can really like push it and get a lot out so i've uploaded um let's have a

look at mr mindstone so mr mindstone is sam jones and you've got his annual review which is basically how your investments performed um a few bits and bobs like that and investment experience where where your attitude to risk is.

Now, I will say here, and I'll come on to the end, I would 100%, 200 % recommend all of you to take out some, you know, date of birth, your name, a few other things,

and preferably as well, turn off the training. So I don't know how many of you know here, but you can turn off whether the LLM uses basically your data to train the model. So yeah, at least do that, definitely.

But when you plug in Mr. Mindstone, what you may want to do is like copy and paste, as I've done here, the content. So there we go, SAM, and what I've done here is that

prompt, which is on the slide, which is summarise in plain English what the three things to be most concerned about, what's in this report, what can I get out of it, basically a summary, and it pops out once we've gone through that document, which is rather long. I will see if we can skip a bit.

hopefully it'll pop out with here we go so three things you'll be most concerned about so your charges per year um here we go yeah your charges compound largely over time big point to be

aware of um sam mr meinstein has been classified as a four to five moderate to invest adventurous investor generally in my industry it's all either like a scale of one to seven or one to ten and generally people were sort of somewhere in there depending on your risk appetite basically if you want to sleep at night what can you deal with volatility in the stock market

and yeah it goes through those things which is brilliant now where does it come to yet an action

Stage 3: Generate a One-Page Meeting Brief and Questions

plan and here we go so that this is where we then come on to the third thing which is generating what you can actually use in a meeting so you can't take um i mean i suppose you could but you probably don't want to take a 10 page chat gpt output into a meeting you can't look at it you can't really think about it and you want to be able to have something really clear to focus on

So this prompt here, which is a bit smaller, so apologies if you can't read it. I'll just click on slideshow So key questions need to ask my financial advisor

Of course you can see on these prompts here are plugged in that role and that's really important to basically improve the output you get I know some more technical people in the room probably have prompts with tons of different things like role barriers all kinds of Yeah constraints as I think Rob mentioned a bit there

so yeah when you plug this in to this output you get oh it did load fantastic you'll get something like this so this is a one -page meeting brief which has gone through um yeah how to ask constructively so i think one of the things that it popped out with is basically your charges are 1 .96 percent and then it would prompt you to basically say ask my advisor uh what am i getting getting value for this money?

Well, you get what I'm saying. Of course, that's quite a blunt question, and you might want to smooth that over a bit.

Using AI Safely: Self-Awareness, Privacy, and Verification

And one thing which is really important about this, and it's something we'll come on to, is looking at your weaknesses or at least areas to improve in. So when you build something, when you're using AI, I'd I'd encourage all of you to try and identify yourself.

Know Your Weak Spots Before You Trust the Output

So I've got here the four types of people I tend to work with. Of course, everyone doesn't fit into these categories, but it broadly really helps you identify how you use AI and where you might be needing some improvement or at least someone to check where you are.

So for example, let's say I hope all of you by the end will be the milestone of 10D, which is basically when you generate a plan like that, your biggest risk, I would say at the moment, is actually just accepting it straight away or not actually checking the output before you rock up to the meeting. Actually have a look at it and see whether it's right.

And you can see all the other, I don't need to go through all of that. Human safety net to connect to the 20K grenade.

Yeah, that's something I saw about a tax -free lump sum allowance. allowance um another story this guy took out thought he could take out a tax reliance on from his pension he couldn't actually give him a massive tax liability so yeah something to avoid

um but yeah when i built the prompt i think it will say on here um here we go my achilles heel so there's something for you to add in as a check and it adds it to that document i think and i've been told to avoid going down but if you see down there um is to try and put that in

to help you um and i don't know if it says on the slide but um i definitely say like here we go the bottom self -awareness is yeah really is the cheapest form of insurance because it allows you

Common Failure Modes: Hallucinations and Overreliance

to really get the most out of it um yeah and here's just a few risks to be aware of i've mentioned the privacy before of course a hallucination so that's why i've come back here which is falling back into just accepting what it comes out with um example of that in

in the real world at the moment is I know lots of cybersecurity testing teams at the moment are basically being turned away because analysts of code are basically just coming back with AI audits, and they're not any good. So that's a little example there.

Resources and Closing

So if I come onto the QR code, yeah, that workflow is on there, Linktree bio. So if you want access to the slides, video, the prompts, so you can just keep them tucked away. Yeah, that's all on there.

If you've got any questions, yeah, feel free, whether it's about Trevor, this prompt thing, or maybe just AI generally in the financial services space, feel free just to give me a question. I'd love some.

Thanks very much, Henry. No worries.

Q&A: Industry Adoption and Prompting for Consistent Results

Question over here. Thanks again, really interesting.

How are your colleagues feeling about the whole AI wave? are they embracing people preparing before meetings or are they finding it annoying

and you obviously how's the industry responding i think generally still a bit in shock um a bit slow to catch on i know particularly in the financial services a bit like yeah we're just going to implement it here and there we'll help um generate some reports but i think it's

it's very much one or the other of either we're not going to use it, we don't trust it, and the flip side of that is we are using it, but we don't know how to use it, which from my point of view just adds in a huge amount of risk.

So I know I've been coming to MindStone for six, seven months. Adrian doesn't actually know this, but you're the one who convinced me to first get a Claude Pro subscription. I don't know whether that's a good or a bad thing.

It's a great thing. It's a great thing. But yeah, I don't think that adoption, particularly safe adoption is really being done much at the moment and yeah you can already see

i know some companies yeah read the ft every day are laying off i think standard chartered um the last day or two said they were laying off something like uh 5 000 people for ai roles um

big four accountancy firms at the moment in the last oh who was it big four so you've got ey UI, Deloitte, PwC, and another one, have actually now got more AI -listed roles than auditing roles,

which is kind of incredible, but whether they know how to actually implement it, yeah.

I'll just draw a little thread through what Robert presented on and what Henry's just presented on in terms of if you remember the rectangle of steps

in order to achieve a consistent output and for those who are perhaps less technical and less aware this is another really good example of if you just put something into AI and then expect it to give you an answer it's highly unlikely to work probability is not with you but if you follow Robert and Henry's

examples of breaking it down and making sure that what you're inputting is the environment is tailored then you're more likely to get something that you can then go and discuss with an appropriately qualified person and get more out of it so yeah really appreciate that kind of loop running through there

and do check out henry's steps um because even if it's not for this workflow it will give you ideas on how to then repurpose the workflows um and get more predictable outcomes is that fair

um any thoughts on that before we end yeah i mean i think prompt engineering is i mean that's basically what this simply is is like a little bit of prompt engineering um but you can really really go off the rails. In a good way, I would say that.

With something like this, these are really, I mean, for me, that's quite a small prompt. And if you want to add in the constraints, specific tasks, and narrow down your output,

I think I just really emphasise that AI will give back a multiple of what you give it. And the more you give it, the more you get back. So do try and invest in learning the steps, learning the process, and you'll get a lot more use out of it.

Conclusion

Brilliant. Thanks very much, Henry. Can we give it up for Henry? Thank you.