What AI tools or services are you using or looking at?
richallum
Administrator
It would be good to get a list running here.
Paraplanner. F1, Apple, Nutella, ice cream. No trite motivational quotes. Turning a bit northern.
Comments
Microsoft - CoPilot for Microsoft 365 (within my existing 365 environment)
Anthropic - Claude Pro (not for personal data and not planning to be)
OpenAI - ChatGPT Teams (not for personal data and annoyingly I have now identified a shortcoming in the data processing addendum that means you can't use it for data processing if you aren't the data controller)
Costs (inc VAT) per user:
MS - £355 annual
Claude - £21.60 monthly
Chat GPT Teams - $36 monthly but minimum two users so $72 ($60 if you pay annually)
Check out the assembly I did a couple of weeks ago with Richard and Jonny Stubbs.
TL:DR
Hyper-charged research assistant and analyst.
@Wildparaplanner here's the link to the assembly: https://paraplannersassembly.co.uk/event/paraplanning-in-an-age-of-ai/
Paraplanner. F1, Apple, Nutella, ice cream. No trite motivational quotes. Turning a bit northern.
thanks both, will take a look
Just watched the assembly link. Good stuff.
Few questions I've got and hopefully the guests will come back:
Jonny - what do advisers do in face to face meetings, are they taking a voice recorder in to the meeting with them? I take it the software is able to distinguish between who is speaking (adviser/client)? How do you deal with special category data (e.g. health/medical conditions) which might get incidentally recorded in the meeting audio / meeting notes?
**Benjamin **- when you talk about "building your own GPTs" - in say ChatGPT (v3.5) - is that just starting a "new chat" and feeding stuff into that particular chat? If i then want to have a conversation with it about a day out in London, I would just open a new chat and start fresh?
Ignoring the data protection issues for the moment, is there any software (free ideally / paid for) that you can dump multiple word documents/pdfs into it to analyse data without having to carve it out manually to then throw into the system? An example of this I guess would be a combination of 1) A cost comparison template 2) Multiple plan information packs - and then tell it to collate the charges and populate the template?
Hi @Wildparaplanner Good question. We're still trialling a number of solutions for face to face meetings as the quality of the transcription is definitely not as good as for actual Teams meetings. This is down to the ability of the AI to be able to, like you've hinted at, distinguish between different voices. It is getting better all the time though and we're having some success with using dedicated external microphones.
We treat the special category data being recorded as just the same as if it was recorded in a written meeting note.
No. Building a GPT is creating a specific GPT and requires a paid solution. In ChatGPT it's really intuitive with natural language, in CoPilot it needs some deeper grounding in software (using CoPilot Studio), which I don't have!
I have built GPTs to do specific things. EG I have a CPD log writer. I have fed it the CISI and CII CPD schemes, with all the terms, and the specimen entries, plus blank forms. I've also given it some real past entries I've done. Now I can give it a CPD certificate and the slide deck, or promo for an event I've attended, and in about 10 seconds it writes the CPD log entry. Because it also knows my job role, it's good at a first draft of the reflective statement.
I have another one that I'm stuttering with that will produce an investment analyst style output for a portfolio. High level - I feed it a Longscan from FE and ask it to review against the CIP for a given risk profile. It already has the CIP, including the main document itself, all asset allocations, funds, KIIDs, costs, weightings, etc. It has a template for how to structure the output (which it doesn't like sticking to!). Results from this aren't as good as the CPD one. I feel like this is where CoPilot would potentially thrive as it could automatically have access to a siloed sharepoint point site of all the past analyses I've written.
@parawhat Which options are you trialling for meeting note recording?
We have a large number of historic voice recordings of meetings and I would (a) like to get them transcribed - & for now, not too fussed about it identifying different speakers and ideally (b) get a summary of the meeting.
@benjaminfabi Having read your post & seen the Assembly, I am more than a bit in awe of what you are doing with AI. Any tips on how to learn how to start on this? I am particularly interested in both the FE analytics and CPD processes you have built.
@richardgough just get stuck in to the tools and start consuming video tutorials on how to use them. Like anything, the more you do it the better you'll be. I'd start by having it help you with writing tasks. I get lots of value by simply having it reduce the word count of stuff I write without losing the message or style.
In terms of transcription, you can try otter.ai and fireflies.ai. The paid versions of these both offer a certain level of customisation in output where they'll look for keywords and surface talking points on those words.
@benjaminfabi Thank you.
Just found an AI meeting notes function on zoom.
Great for getting advisers to actually write up their notes in a coherent and logical manner!
Based on receiving the outputs from several planners using various AI over the past few months, ChatGPT is hands down the best tool for putting together meeting notes summaries. I have a client who uses Fireflies.ai to get a transcript and then loads it through a custom GPT he has built.
The quality of the client file I get from him is significantly higher than anyone else. It helps that he is a very good planner and understands what he needs to talk about. But if you're looking for ways to improve how advisers communicate with clients this will highlight weaknesses so fast.
Firms using this properly, and as a training tool, will see massive levels of improvement.
Hi all, thought I'd bump this topic as I imagine there's been more use/progress in the last 12 months.
We have trialled and have now fully implemented Saturn.AI into our processes. Its a fantastic bit of kit that transcribes either phone call or team/zoom or face to face meetings into a really neat, bullet pointed summary and as its financial planning based, it puts things into really handy sections and knows the lingo!
Mainly used for client review meetings, onboarding/discovery meetings and you just plug in the names and it feeds it all through. Big time saver, rather than the adviser taking notes and providing summary to paraplanners and really drills down the detail. It also specifically looks for things in relation to vulnerability and the key pillars to match it too.
It has different templates/categories depending on which meeting type you pick and I have also started using this in non client meeting teams calls to take the detailed notes accordingly. For the time saved vs cost feel its a great addition to our tech stack.
We had a demo of Planner Pal and have another with Advisory AI shortly. Planner Pal certainly looked good. Anyone have any feedback either of them, or on Aveni which I am also looking at?
Hello Tom, last year we looked at Planner Pal, Advisory AI, Aveni and Saturn for meeting notes. A very high level summary of our feedback was:
PlannerPal - Nortes were too basic and lacked detail
Aveni - Notes we detailed but lacked accuracy.
Advisory AI - The most impressive, however also the most expensive. They also demo'd their
Suitability letter writing module, which has bags of potential but will take a while to get right. We need to get our Xplan data right first before we can consider that an option.
Saturn - We opted for Saturn in the end. Meeting notes were detailed and accurate and Saturn seem to be moving at a pace more suited to us. They are working on Suitability Letters and i'll meet with them to discuss this in April. @Nath Have you spoken to them about development here? I want to automate our ASRs if possible.
I also want to look at Templi.ai. Does anyone have any experience of this?
Thanks for this Gus, very useful.
Thanks for the updated info. Are any of them useful for data cleansing?
@Gustavo_Fring I haven't spoke to them on this yet. We are early adopters but keeping a close eye on developments via the Saturn community feed that we have a login for. If its up to scratch we would consider this but I feel it could be while before it doesn't involve too much manual intervention.
thanks for the info - are any of them useful for data cleansing?
I agree with expectations on timescales. Listening to a meeting and compiling summarised notes is one thing - a comparatively simple task (one data file from one data source, into one transformed output as one file) where the AI learning can take place across literally millions of users globally every day. For example I use AI to transform dictations almost every day and it's good at it.
But for anything that involves:
Well you very quickly get into an insane number of interdependent processes.
Paraplanning is one such function where the complexity becomes dauntingly vast if you try and break down all the component steps needed to build a finished output. Added to which, to get them right, you need the very best technical experts checking and feeding the learning.
And this final point is still the key - we still cannot rely on AI to do knowledge dependent transformational tasks (e.g. tax calculations) that we aren't already highly proficient in, because it gets it wrong all the time.
@benjaminfabi which tools are you using for transforming dictations. Are these your own dictations or someone else's given to you?
Paraplanner. F1, Apple, Nutella, ice cream. No trite motivational quotes. Turning a bit northern.
Different options depending on where I am and what it's for...
The former is good for all kinds of on the fly technical queries, where I have the answer in my head but need to reduce the time it takes to get it into a reply.
The latter is good for sending alongside complex advice reports as file notes, or run-throughs for the adviser, or just as a brainstorm of all the things I've considered when arriving at the recommendation.
Copilot in MS365 (the ~$300 pa one) is a really good productivity tool across the entire product suite.
I'm always looking at other options.
The bespoke tools in the FS market mentioned are all decent enough, but I am finding that for a monthly spend on AI of about $100 (inc 2 chatgpt licenses) I can get excellent results from building my own workflows across combinations of off-the-shelf options.
Thanks. I've been using Cleft for getting stuff out of my head and it produces a really good, well structured transcript.
Paraplanner. F1, Apple, Nutella, ice cream. No trite motivational quotes. Turning a bit northern.
10/10 for the Trust section on that service. Very clear for due dil.
But no Android app means it doesn't pass my friction test 😢