Why You Can’t Analyse Client Data (Unless You Do THIS)

Thing I thought I'd do is just give you a quick update on what is new in the world of ChatGPT. And it's changing all the time right now and I'm focusing on ChatGPT. It may be that you use different large language models, but I know that most of my audience use ChatGPT. It is the most common large language model. And so whilst I could talk about updates to Gemini, updates to Claude, if I covered the updates on all the AI tools right now, that will be the full session. I want to get into today's content, so I'm going to keep it just to the OpenAI stuff. And there is actually an interesting, just in the last few hours, a new update there. So let me just open up their page, which I think is here. This is their release notes and I'm not going to go back too far. You can see that there's a lot of stuff happening at the moment. Last time I went live, I think what I mentioned that GPT5 is being rolled out. You should have GPT5 now and some people love it, some people hate it. I've just done a video which I'll be releasing soon on how to get more out of GPT5. One of the things that's interesting is OpenAI released some guidance notes on how to use better prompting for GPT5. So watch this space that's coming very soon. I filmed it earlier today, it will be out very soon. Image generation is available in ChatGPT for free. Also, they've been building in lots of connectors and they are rolling those out quite regularly. So initially, if you're on the plus plan, that would be box, canva, Dropbox, HubSpot, notion, etc. And then they announced that they would roll out Gmail, Google Calendar, Google Contacts initially to the Pro plan. But as you will soon see, that's being rolled out to other people. So you've probably got the Google connectors as well, which is interesting. So we can connect up our ChatGPT to other sources for context and also as part of its move towards an agentic model. Then we had updates on GPT5 because there were so many complaints about GPT5, people were struggling with it. So firstly what they did was they broke down GPT5 to allow you to choose whether you wanted the default, which was automatic, or whether you wanted to choose and specify whether you wanted a quick, fast response, or whether you wanted to wanted to do its thinking. So you could now specify that and then what they also did a few about A week ago is they also allowed you, if you, if you didn't like GPT5, you could go into settings and you could toggle a switch which would then allow you to get access to all the legacy models. And I know some people in the AI Academy have done exactly that. They've switched back to 4.0 and which.

I personally love ChatGPT5, but we have to learn how to prompt it better. But one of the things that's interesting right now is back in, I think it was April, there was a big update in the image generation and OpenAI moved ChatGPT away from Dall E3 to their new image generation, which was awesome. With text, it would get spelling correct on text and that's been amazing for the last four or five months. And then ChatGPT5 comes along and if you generate images with text, with ChatGPT5, we're finding spelling mistakes on our images. So if you are find out if you like to create images, perhaps for social media posts, etc. Using the image generation and you want and you're finding that it's making spelling mistakes with the text on there, then it's really simple. All you need to do is make sure you unlock the legacy models and if you're creating images, do it with four O and the. The text is fine. Okay, so a little tip there. If you want to use images, I would use 4.0 for that. It's one of the areas where it's been disappointing. I'm sure they'll fix it. Okay, what else is important? Yes, more. I don't really. This one doesn't excite me at all. But they decided to make GPT5's default personality warmer and more familiar. And so it would say things like, good question, great start. Personally, I hate that. I mean, it did say people have complained it's too sycophantic and they're trying to reduce that, but they want it more friendly. Personally, I just want the answer. I don't want it to say, oh, that's a great question mark. Question mark. My name, not a question mark symbol.

And then this is just. If you are in India, then they've just announced in the last few hours they're introducing ChatGPT go only at the moment in India. So it's kind of.

An advanced version of the free plan, but I think you have to pay for it. So I'm not quite sure I understand that. I only saw it just a few minutes ago before it went live. And given that most people aren't in India. It's not relevant to us at the moment. Okay, I did forget to say, as part of the announcements, one more thing very, very quickly. If you're new, I'm here to answer your questions. The way that works is if you have a question, put it in the comments. If you are listening on LinkedIn. Unfortunately it's not pulling in. My system doesn't pull in the LinkedIn comments. But nevertheless, Sarah will do an amazing job at grabbing your comment, putting it into YouTube on my behalf and I can see your questions. Then if you have a question, please put two capital Q's in front and that way I'll see it. Because sometimes the chat does go crazy and I encourage you, please do chat. If I don't read out your chat in the session, I read them all afterwards, my team reads the chat. So please, please keep the chats coming. I appreciate that. If you want to ask a question though, put two capital Q's in front and I'm going to get to the questions towards the end. Usually we have about 10, 20 minutes for questions, really, depending on how many questions come in, I think there might be few questions on today's topic, but we will see. So if you don't ask questions, I can't help you. I can't answer your specific issues. So please, any questions about anything to do with AI ChatGPT Today's topic, please do ask. I am here to help you. Okay.

So our topic today is really all about what are some of the really powerful things that we can do with with ChatGPT, but why privacy? And security is such a big issue that potentially it's stopping us doing the things that are most useful to us. Particularly if you are an accounting professional, the owner of an accounting firm, owner of a bookkeeping business, then there are some severe limitations because of privacy, security, client confidentiality. So what I'm going to do is in this session today is we're going to look at, we're going to start off, I'm going to show you, we're going to work through a prompt together. I'm going to show you a really, really powerful use case of ChatGPT and then I'm going to talk about what are the pitfalls. Why can't we use this use case unless we do certain things? And I'm going to show you what you need to do to be able to to do the thing I'm about to show you. Okay? So I know some people are worried about privacy, security, which you are right to be. We will look at the workarounds and the best strategies for enabling us to do the things that are most powerful for us and particularly for our clients. So with that said.

Let'S get started with the example first. So that if you're an accountant or a business owner, one of the things that's really powerful is we can use it to analyze financial statements. And that can be really, really powerful. If you own a bookkeeping business, an accounting business, and you have, let's say, a meeting with a client once a year to go through their financial statements, then one of the things that we can do is we can create prompts in ChatGPT that will analyze that data and give us really good outputs. With the right prompts, we can get some amazing outputs. And there are so many things that we would want to do that for. We'd want to use clients financial data and we just have to be very careful about privacy and security. But let's start off with an example of how powerful ChatGPT is in this area. And what I'm going to do is I'm going to run a prompt that will analyze financial statements and produce a report. So let's imagine for example, that you are, if you're the owner of an accounting firm or bookkeeping firm and you have a meeting later today or tomorrow with a client and you're going to talk them through their end of year results. And what you would like is you would like a report that you can use to help you spot any particular issues you might want to bring to the attention of the client, or even a report that you might give to the client. So really powerful and a very common thing an accounting business would want to do. So I've created a standard prompt for doing that, a prompt that we could use frequently. And so let me first of all go and find the prompt. So what I'm doing right now is I'm going to my prompt library. This is the software that I've created for managing prompts. And I know many people here have got these prompts and I'm going to just filter to, there are hundreds of prompts here. I'm going to filter to my financial data analysis prompts for accounting firm owners. And there's one called Financial Year end Report. So I'm going to get that prompt, which is here, and I'm going to just simply copy that from my prompt library. I'm going to copy that to the clipboard, I'm going to jump across to ChatGPT and I'm going to paste it in. So there's my prompt. Now I'm going to use chat GPT5 on automatic.

But as I. But one of the things we can now do is we can also specify whether we wanted to use fast or thinking or pro or even one of the legacy models. I'm going to leave it on automatic for now. In fact, no, I'm going to leave it on fast. You might want to use more thinking because this will be a really important report and if we're analyzing financial data, I would definitely suggest the thinking model. It just takes a little bit longer. And seeing as this is a, this is live, I don't want you to sit there waiting, twiddling thumbs while ChatGPT does some thinking. Okay, we need some data, of course. And so I'm going to use a, I'm going to use a case study and I'm going to pick a business in the US in Texas. This is a real business. I have no idea who this is, but it's a pool maintenance company in Texas. This is their website. And what I'm going to do is grab a link to that website. Because if we are, if we want to give some advice to a client, let's say perhaps we're in this case analyzing financial statements, then whilst we can get a lot of information from the financial statements, from the raw numbers, if we have more context, more background information, if we give to ChatGPT more information about the business, it will come up with better research, better responses. So if you have been, if you're a regular to my regular to my sessions on AI, I talk a lot about context is really important when it comes to using AI models. Give it background information. And if we were, if we're getting, if we, if we're helping a client, then one of the sources of background information is their website and we could absolutely grab the website. It's not the only source of information, but it has a lot of information on there. It will tell ChatGPT what business they're in. Perhaps it might have details about their products, their services, their pricing and so much more. So I've grabbed that. And what I'm going to do is I'm just going to go to the end of my prompt where it just asked me to give it some further information. It's called background information. And I'm going to just replace this placeholder text with.

I'm just going to say here.

Analyze the business website here. Okay, so that's the, that's the context. Now ChatGPT will know something about the business. Now of course, there's one more thing that we do need to give it, and that is the financial statements. So if I just go over to here, I've created some financial statements for this business. I don't know, Paul E. So these are actually made up numbers. But I created an income statement for this business.

I created a balance sheet for this business. I think it's here, here's the balance sheet. And I created a. When my. When it'll let me do it. There should be a cash. There should be a cash flow as well. My ability to use my mouse is. My trackpad is failing me. But anyway, there's the financial statements. So all we need to do with ChatGPT is, is I've got those financial statements on my desktop and I'm just going to drag those in, which I'm doing right now. So if you're not familiar with ChatGPT.

You can give it documents like PDF files, Excel files, word documents, images and so on. So this is a PDF document and there's a plus button at the bottom for allowing us to add more documents. But I find it easy just to drag them in. That's what I tend to do. So I've now given, I've got this standard prompt which I cut and pasted from my prompt library and I've given it two pieces of information. I've given it the financial statements and I've also given it some background information in the form of its website. Before I run this, let me just very quickly tell you about some of the things in this prompt. It's a long prompt. If you're new to my sessions, One of the things I teach is prompt engineering. And if we want better outputs, better results from our prompting, then we need to create better prompts. The better prompts, the better and more consistent the results. And I talk a lot about why anything you do recurring, like analyzing financial data, let's create some prompt templates that are well written and save them somewhere like a prompt library, so we can just grab them whenever we need them. So this particular one, this prompt, it's telling ChatGPT to act as a financial accountant with expertise in interpreting and analyzing financial data. That's the role I want it to play. We know from prompt engineering that when we ask it to play a role, we get a better result, better answers.

And then its task is to carry out a detailed examination of the profit, loss, account and balance sheets for the current year. And then what I've done is I've given it more specific details that I want it to look for discrepancies. I want it to do some financial metrics analysis, calculate some ratios. I also want it to look at cash flow and efficiency. I want it to look at a cost structure, review growth and risk assessment, and then provide some recommendations that I could present to the client. And I've got some output, so outcome expectations. And then I fed it an example of the sort of report that I'm looking for. So.

If you're not familiar with prompt engineering, one of the things that's really powerful is to put examples inside your prompt, particularly if your prompt is going to create some sort of output that you might present as a report to a client. For example, if you give it an example, because AI systems, large language models are basically looking for patterns. When you give an example, that will help it to make sure its output is laid out in a similar kind of way. So this is, I put in here an example report setting out the way that I would like it to create my report based on the financial data then. And there's the recommendations and then we have.

The background information. So as you'll see, this is a long prompt, but don't be put off by that, don't be scared by that, because when I create these prompts, very often a large part of the prompt is an example or examples or templates. Okay, let's run this then. I'm using GPT. I'm forcing it to do it quickly, the fast model, rather than thinking, just to get a quick result, just so I can show you the sorts of things that it will do. But if you were doing this, I would suggest you would probably want to use thinking. So it's reading the documents and now it's creating my report for me based upon this business, it's a real business, but the financial statements, this demonstration several times now. So we've got, it's done some calculations of metrics of return on assets, return on equity, which might be really useful in our meeting with the client to explain some of the trends. And it's given a few interpretations around those. Then it's given some analysis of cash flow and operational efficiency with some observations. We've got a cost structure review. It's picked out that what it thinks is the most important costs on the profit loss account. And it's given us what is the, this year's, last year's, the absolute change and the percentage change, which is really useful with some insights. Then we've got a growth and risk assessment and then we've got our recommendations, some recommendations we might give to the client in the meeting. So I don't know about you, I think that's a really powerful report. And if you're an accountant or a bookkeeper and you were having a meeting with every single client every single year, the ability to create a report like this in seconds, when you have the correct. The right prompt at your fingertips, I think is really, really powerful. This, I think, is one of those really powerful use cases of ChatGPT and all other large language models. But there is a huge problem, there is a huge problem with doing something like this and you must be aware of it. This is the privacy and security issues. And so what do we have to consider? Well, we have to consider, firstly, we have to consider professional ethics. So.

I don't know whereabouts you are in the world and what institutes you're governed by, but generally speaking, I'm a Chartered accountant, so I'm a member of the icaew, which is the Institute of Chartered Accountants in England and Wales, which I guess is the closest equivalent to a cpa, for example, in the us. And I'm guided by the professional ethics, which very often tell us we have to maintain client confidentiality. So you are probably affected by that. That's a big thing. Not only that is we are then affected by data protection laws. And whilst they are perhaps different around the world, there are certain commonalities. And what I'm going to focus on in today's session is gdpr, partly because GDPR is really important for anybody in the UK and the eu, but actually, I think GDPR is a good set of guideline for anywhere, wherever you are around the world. But if you're not in the EU or the uk, certainly look to your own data protection laws. But if we think about gdpr, we have to be aware of things like it talks about a lawful basis and consent from our clients. That's a big thing. Also, data transfers, that is huge, as you will soon find out. There's also this right to be forgotten, which is potentially a big problem for us. Again, we'll get into that and then we have to be concerned about security and breaches and liability. So, in other words, we have to make sure client data is protected. And so we have GDPR and data protection laws, we have professional ethics, and then we also have confidentiality agreements and client consent. What I mean by that is, if you're an accounting professional, you probably have, for example, engagement letters in place, and very often those engagement letters that we have in place, there are contractual agreements and probably those Signed engagement letters specify already that we will keep client data confidential. So just the documentation we already have probably restricts us in that area. In the us, for example, I understand that tax preparers, they're governed by IRS regulations around disclosing tax return information to third parties without permission, and tools like ChatGPT is a third party. So wherever you are in the world, this is a big issue. I'm going to focus on GDPR from now on just because I think that's one of the, if not the most comprehensive set of regulations. It certainly affects people in the, in Europe and UK particularly, but I think it affects everybody around the world. So we're going to look at gdpr, but you will also need to think about your own local data protection laws and any professional institute guidance. So when we think about gdpr, there's a number of articles they have and there's many articles that govern AI. Probably the two biggest ones to think about is Article 28. Article 28, it says very clearly that we can only use AI tools that offer a data protection data processing agreement, a dpa, a data processing agreement. Now this could have far reaching implications across all of your other AI tools. I'm focusing on ChatGPT today because there is a solution, a neat solution. So we should have a data processing agreement in place that's article 28 of GDPR and then article 17 is basically the right to be forgotten. Erasure rights. The erasure, the fact that clients should be allowed to say we want to have our data back, we want it being deleted, that's potentially a really big issue.

Now if you are in the EU or you have clients in the eu, there is actually something else that came into effect on 2nd August 2025 and that is the EU AI Act. And this applies to you if either you are in the EU itself. So if you're in Ireland, for example, you have to be aware of the additional rules in the EU AI act, or if you have clients in the eu. So you could be a US accountant and you have some clients in France. Well, if that's the case, the EU AI act is something you need to be aware of. And there are a few things in there that go above and beyond gdpr. But most of the things that I think are relevant that we need to consider are gdpr. Right. So that's why I'm going to focus on that. Let's talk about what you cannot do, what you cannot do, and then let's talk about what you can do. I want to, I want to make this positive and not get you running scared and thinking you can't touch AI models. Yes you can. We just have to understand what we can and can't do. So what you can't do if we think about ChatGPT. If you are on a free plan of ChatGPT, you cannot, cannot put in any client data. You cannot do what I've just shown you. You cannot upload a set of financial statements if you are on a free plan. Why? Because it violates Article 28amongst many other things. What about the plus plan that you might pay for if you're on the plus plan? Same problem. Paying 20 bucks a month to chatgpt to OpenAI does not solve that problem. So if you are on the free plan or the plus plan, you cannot enter client data. It's really clear. Okay, what about though I have seen a few people saying, ah, but if I go to ChatGPT.

Then there is. And they've moved it where they moved it to. There is this temporary mode. I was going to show it to you and I can't see it right now. Let me start on it. Oh, perhaps I have to start a brand new chat. Yes, in the top corner you can turn on temporary chat. And what temporary chat does, if you turn that on, it means that your chat does not go into your chat history. It doesn't go into the left hand menu. And some people have said that if you turn on temporary chat mode because it doesn't get remembered that we're okay with that. No, no, that does not solve our problem because even with temporary chat, your chats are stored by OpenAI for 30 days. And they're doing that because they want to check people aren't violating their various terms, which does make sense. So temporary chat doesn't work either. So what can we do? Well, what we can do is we can still have our free accounts or our chatgpt+ account if we are using AI with anonymized data or synthetic dummy data. And for many things that might be sufficient, but not with analyzing financial statements because it's, what use is it to give it made up numbers? Now we could give it financial statements if they're anonymized. In other words, we take away the client's information. If they're just numbers, then yes, that is okay. But that takes quite a bit of work then because we have to figure out, well, how are we going to make sure we remove reference to the client from the financial statements? That's going to take time in itself, we're going to have to take the financial statements and we're going to have to redact out all mention of the client, of where they're based, their address, any details that mean that we can then link those numbers to a business. So it is doable. We could, for example, have the financial statements as an Excel spreadsheet and just make sure we delete the cells so it's just the numbers. If that's the case, then we're okay. But the problem with that, the problem with anonymized financial data is I said to you earlier that we get a better result, a better analysis if we give context, background information about the business. And so if in the same chat we also give the website of the business, well then ChatGPT can put two and two together. It will know that this anonymous data based on the context of the chat relates to this website. Ah, it's this client. So that's where I think we then have a problem when we start to give other context. So our chats are limited to anonymized data, which means that we're going to have less context. Now we can use public domain information. Anything in the public domain is fine with a free account, with a plus account. So that means yes, we can feed ChatGPT with a client's website, I believe, but not the financial data. Even if we analyze it, it'll know from the chat which client we're talking about. So feeding a website doesn't solve our problem. Now we can do it in other circumstances if we want to simply give the website because we want to create a prompt that's going to help us give the client some advice. As long as we're not also putting in private data like numbers, then that can be fine. So the public domain information is okay as long as we're not using client data. Now then, next level of overcoming these problems, next level of overcoming is we can use the ChatGPT team or enterprise plan. Now the enterprise plan is for super big businesses. The all the big accounting firms use the enterprise plan. But for most small businesses, probably everybody listening to this, the enterprise plan is just way too expensive. It's not practical. So our workaround is to use the ChatGPT team plan. And the team plan does comply with Article 28, which is great.

Also when your data is transmitted. So one of the issues is, particularly if you're not in the US your data is stored on OpenAI servers, then that data is encrypted in transit. Also with a team plan, your inputs are not used for training the model. So when you're on the team plan, you will notice at the very bottom it says OpenAI doesn't use. Our plan is Value Pricing Academy. OpenAI doesn't use value Pricing Academy Workspace data to train its models, which is great. Now, interestingly, you can turn that off in the plus plan, but you still can't use it because of GDPR with the plus plan because it doesn't comply with Article 28. And don't think the data is encrypted in the same way that team plan is.

And also the team plan is covered by OpenAI's data processing agreement, which is why it then complies with Article 28. So on that note, I think let me just pull up a really useful website for you that you might want to check out, which. And we'll put the link in the description later. So there's a website where. Which was updated back on June 4, 2025. It's called Enterprise Privacy at OpenAI. This is a really useful website to have a look at webpage because they have these FAQs around data and how they use it. It talks about the enterprise plan and the teams plan. So there's some questions there that you might check out, particularly how does OpenAI ensure data security?

So that's really useful. Then there's a section on the education plan. Probably we're not using the education plan, but there is the team plan. So there's an FAQ around the team plan. And so what compliance standards does ChatGPT team meet? And so one of the things is it complies with SOC 2 type 2. They had this back in July, sorry, June of 2025, they passed this SOC 2 type 2 audit. So that's a good place to get some information about privacy and security relating to ChatGPT and the team's plan. Another really useful page which you can find this link to that previous one is they have their trust portal and what's reassuring about the trust portal, and it gets a bit technical in places, but the key thing is you can see on the compliance section that they've passed SoC2, SoC2 type 2, SoC3. And if GDPR is important to you, which absolutely should be, I believe wherever you are in the world, we should be looking at gdpr. Then if you click on the GDPR badge, you've got some interesting information here because what they've done is they've got a data processing agreement in place and if we click on the link, my understanding is that what you need to do is complete this. If you complete this for your business, that'll send it off to OpenAI, there'll be an online digital signing process and you will have a signed data processing agreement in place in your firm. And that then satisfies Article 28 of GDPR. And that means you can happily use the team plan of ChatGPT. Phew, that's a relief. But it's not quite, doesn't quite solve all the problems because and here's the issue, and this is a really, really complicated issue and you may possibly remember in the early days back in 2023, for a few weeks, Italy banned the use of ChatGPT. I remember that because I was on hold in Venice at the time and I couldn't access ChatGPT. And the reason for their issue or one of the many issues they had, and it's still an issue, is this Article 17. Because Article 17 says that when you're holding client data, they have the right to get you to delete it.

So if the data is sat on an AI model, how on earth do you delete the data? And that's a huge complexity. I don't think anyone's got the answer to that one yet. This space is evolving all the time. Perhaps we'll come up with something. When GDPR was written, it was originally written in a non AI world and so this is evolving. We don't have a final solution yet. But my point here is that I think the team's plan is fairly safe to use.

But who knows, it may, maybe we find out at some point. We find articles on GDPR getting strict on this, saying no, you can't use it, but the moment it looks like it's passed, it's got all the ticks, everything on the site I've showed you. It looks like there's approval between OpenAI and GDPR to say if you're on a teams plan that you can use it. Now what if you are a solo practitioner? Very often bookkeepers are solo practitioners. Well, unfortunately you still can't give client data if you're using the plus plan at 20 bucks a month. I would strongly recommend that any accountant in business needs to use the team plan. And yes, that does mean that you're paying for two seats. Unfortunately, that's the price of being able to do financial data analysis.

Now it doesn't end there because there's some more things that we still need to do. As I perhaps touched on earlier, we must, must inform clients if AI is being used. It's again part of GDPR So we need to make sure that we've got an agreement in place, perhaps through your engagement letters. You need to be very clear on getting their agreement. Also you should have an AI policy in place, your firm as well.

So that then that I think is pretty much everything that you need to do. Use a teams account, make sure the clients agree, have an AI policy and then you can do this amazing stuff that we've just Talked about with ChatGPT we can do. Also you should have an AI policy in place in your firm as well. We can do financial analysis and I've shown you just one version of a prompt that's really powerful, creates a great report. But we can go even further than that I created for members of the AI academy I recently created. I think it was a request from somebody, we could create a full financial reporting pack. And I created a series of prompts that does exactly.

So that then that I think is pretty much everything that you really build on that first one and we just run in the space of about two minutes to use a teams account, make sure the clients agree, have an AI policy and then you can do this amazing stuff that we've just Talked about with ChatGPT. We can do a series of five more prompts and it builds out this amazing reporting pack that we can that goes beyond just looking at the numbers, gives the client advice on so many areas around their business that we could use in our year end meeting with the client and then we could then leave them with the report to add huge amounts of value. There is so much stuff that we can do with AI with ChatGPT when it comes to advising clients based upon financial data, it's really powerful. We just have to make sure that we are aware of the pitfalls. And of course one of the pitfalls I mentioned, even with the teams account, we have to be aware of Article 17, which it doesn't really solve that conundrum of the right to have your data deleted. Now there is another solution by the way, and this is something I think will become more commonplace in the future. So at the moment most people are using tools like ChatGPT and Claude and Gemini through their browsers, for example.

And we can do that. But an alternative is to use open we can do financial analysis and I've shown you just one version of a prompt that source models on our own computer. And so that's really powerful, creates a great report. But we can go even further than that I created for.

That'S something that is now possible to do, I think it's the ultimate solution because if you can run a large language model locally then.

You don't have to worry about any of the things we've talked about. So let's just talk very quickly about that. How is that possible? Well, if you run members of the AI academy I recently created, I think it was a request from somebody, we could create a full financial reporting pack. And I created a series of prompts that does exist. A large language model on your own computer, then everything stays on your computer, none of the data goes anywhere else. And so GDPR is no longer an issue. Privacy and security is no longer an issue. All your stuff is on your computer. You are fully compliant with GDPR because there is no third party data processing if you're running a local LLM. And this is something that is becoming increasingly something that is practical and we can do because there are various apps that enable us to run local LLMs, a number of different apps that we could use like LM Studio, Ollama and many, many more. So all we do is we, we find the appropriate app for our use and we download it. That's really easy. And then once we've done that we can download.

Show that it really builds on that first one and we just run in, in the space of about 2 minutes models and there are a lot, there's been lot, there's dozens of open source models met a series of five more prompts and it builds out this amazing reporting pack that we can, that goes beyond jita for example, they have their LLAMA models that have been arguably in the past the best of the open source models. That's Llama by Meta. Google have their Gemma models. There are many, many open source models that you can download onto a computer. Yes, you need to have some spare hard drive space. Looking at the numbers gives the client advice on so many areas around their business that we could use in our year end meeting with the client and then we could then leave them with the report to add huge amounts of value. There is so much stuff that we can do but it doesn't require a lot of. But the big issue is the memory, the RAM and most importantly the virtual, sorry, the vram. VRAM is really important. You need to have a fairly powerful computer to run LLMs but that's becoming more affordable, particularly if you're like me and an Apple user. A lot of The M series, M1, M2, M4 are really good computers where you can run local LLMs. Now there are some problems, one of which is you don't have access to the Internet which you don't want because you're keeping them local to solve privacy and security.

One of the big problems I found is that even some of the better models like Llama when it comes to complex financial data analysis, it spat out rubbish, it didn't calculate ratios correct and so on. So I was getting a little bit frustrated with that with AI with ChatGPT when it comes to advising clients based upon finance. But then the biggest announcement of August 2025 so far, everyone thinks it's ChatGPT data. It's really powerful. We just have to make sure that we are aware of the pitfalls. And of course one GPT5 of GPT5 no way bigger than GPT5 is OpenAI released GPT OSS O S S and that's their first open source model and we can the pitfall as I mentioned, even with the teams account we have to be aware of Article 17 which it doesn't really solve that conundrum of the right to have your data deleted. Now there is another solution by run that on a computer. There are two versions. There's the 20 billion version and the 120 billion version. The big ones just two, you need a huge computer. But the GPT OS, I tested that out and I've run my financial reporting prompts through that and. And it gives if anything, better results than the one I showed you. GPT OS is an extraordinary model and we can do things. If we can have a system set up where we can run an LLM like that locally, then we can analyze as much data as we want because everything is sat on our computer. Okay, so if you want to find out more about that, if there's only so much we can talk about today, this is still a big issue. But if either way and this is something that I think will become more commonplace in the future. So at the moment most people are using tools like ChatGPT and Claude and Gemini through their browsers, for example, you want to go deeper on privacy and security. If you want to know if you want to know how to create an AI policy, what to say to get agreements from your clients and even more important, if you want to know step by step how to run a large language model locally. It's one of the many things I cover in the AI academy. So if you want to go deeper on this, if you want to learn more about privacy, security, client confidentiality, how to solve it.

And we can do that. But an alternative is to use Open then I have a complete course in the AI Academy. I'm going to ask my team to put a link to it BIM source models on our own computer. And so hello, I know there's some people are here already in the AI Academy and I have a very special announcement. The AI Academy is just a small monthly payment. It's just 59 bucks a month and you get access to a community of hundreds of accounting firms that you can ask questions of me and them and learn more. There are in the academy there are dozens of courses that I've created including the privacy and security course which includes a recent addendum on how to run a large language model locally.

And also members of the of the AI Academy get a completely complimentary. That's something that is now possible to do. I think it's the Ultra copy, the software that I created, the prompt library software with ultimate solution. Because if you can run a large language model locally, then over 100 prompts that you can copy and paste specifically for business owners, most important, specifically for the owners of accounting firms. There are prompts for financial data analysis, for business advisory and strategy, for marketing, for content creation, for copywriting, for creating SOPS systems. You don't have to worry about any of the things we've talked about. And processors, there are prompts for creating your value, pricing and so much more that I've written over the last couple of years specifically for my community and the accountants that I work with. And you get that. So let's just talk very quickly about that. How is that possible? Well, if you run as my gift when you join the AI Academy, it's only $59 a month. The price will go up in the future, by the way. But one of the things people love about our academies is that the price never changes for you. When you join today, you get the special, special price which will never ever change. And I keep creating more content. But if you're not sure, if you are not sure yet, then what we've also done for the first time today is we have decided to give 7 day free trial and that will be available for the next few hours. A large language model on your own computer. Then everything stays on your computer. None of the data goes anywhere else. And so GDPR is no longer an issue. Privacy security is no longer an issue. All your stuff is on your computer until first thing tomorrow. So if you want to try it out, Sarah's going to put a link, the.

Link in the chat which I think she's done to go to that link and you can find out much more about it and then get your free trial and you can then check out the various courses for the next. You are fully compliant with GDPR because there is no third party data processing if you're running 11 days and then anytime before seven days. You can always cancel if you want. We never done that before. You will have to do it in the next few hours. By tomorrow morning we will turn local LLM. And this is something that is becoming increasingly back off again partly because we're testing to see how many people are interested in a free seven day trial. So Sarah's put the link into the chat. Go and grab that. Right.

Let'S answer some questions. But before I do, just a few final things. If there's anything you've learned today that's been valuable, if you could let me know by just hitting that like button, I'd really appreciate that. If you're not yet, if you're not yet subscribed to my YouTube channel, then do that and then you'll get notified of new videos. Also another thing to tell you about is next week we have a big event, it's Thursday of next week. That is practical and we can do because there are various apps that enable us to run local LLMs, a number of different apps that we could use like LM Studio, Ollama and many, many more. So all we do is we find the appropriate app for our use and we download it, tweak. And that's the AI Summit. We run that a few times a year. It's a three hour webinar where we go much deeper on all things ChatGPT and AI and use cases. And the great news is you can get a completely free ticket for the AI Summit. So if you want to go deeper on this, make sure you. That's really easy. And then once we've done that we can download.

Show, grab a ticket. Sarah's going to put a link in the chat. If you are already a member of the value pricing academy, please do not sign up for it because you will get a link anyway but in usual channels. So I think that link that Sarah's going to give me will appear fairly soon. Yes, and we'll. It's in the. There you go. That's the link to use to come and join me next week for three hours And Emily on the AI Summit, as I said, completely free Amazon models and there are a lot, there's been lot, there's dozens of open source models. We're going deeper on all of this. What else? My next live stream is next week, August 26th what's my next live stream? I only looking at one of my friends asked me. He said Mark. So for example they have their Llama models that have been arguably in the past. I don't really use ChatGPT yet and what I want to know is I really want a. It's the best of the open source models that's Llama by meta. Google have their Gemma models, the beginner's guide to ChatGPT and also I want to know how I can use it from a practical point of view, from a personal point of view. So that's what I'm going to do next week on the August 26th my live stream is going to be really a beginner's guide but also I'm going to share my favorite use cases for personal There are many, many open source models that you can download onto a computer. Yes, you need to have some spare hard drive. Spare use not business things but personal things. How can we use it in our personal life? And also if you're interested in some ideas on that, some use cases. I will hopefully see you on August 26th. Okay, I think, I think it's time for some questions and I'm just going to go back. Remember I said put two capital Q's in front and I think ah, my good friend Marty was the first. Hi Marty, hope you are doing well. And you say if we move from plus to team, does our chat history, projects, etc. Come over to the team version? Yes. Yes is the answer. I forget exactly the process because.

When plus came out, which was I think in March or April, I think April of 2023 the plus plan came out. I got the plus plan as soon as it came out in 23. And then I forget when the teams plan came out but when the team's plan came out we decided we wanted it for our entire team. Everybody here we are, we're a team of six. We all have. We will have our plan, the team's plan and the reason we do it. At least it doesn't require a lot. But the big issue is the memory time was because one of the nice things about team plan is that you have centralized billing before that the RAM and most importantly the virtual sorry, the vram. VRAM is really important. You need all of my team, we're submitting monthly expense claims to us for their personal subscription to the plus plan. And when we did that one of those questions was well does it. Can you pull across all of your information? And the answer is yes, it pulls across your chat history. You'll have a fairly powerful computer to run LLMs, but that's becoming more affordable projects. But it doesn't, it's not shared by your team. So think of the team plan as being a wrapper that sits around each person's individual plan. So each member of the team, your plans stay, your chats stay, yours and personal to you. The team plan itself doesn't have a huge amount in terms of sharing. You can share GPTs in BULL, particularly if you, if you're like me and an Apple user. A lot of the, The M series, M1 team plan, that's one of the benefits of it. Of course the big benefit is the privacy and security.

So Marty, I know you're in the AI academy, so if you are moving from plus to team and if you have any issues with that, any questions, whatever, just post them in the AI academy if you can't find the answer and I'll see if I can find the answer myself.

Okay, I know there's some people here on LinkedIn so let me give you a quick shout out. Hi Robert, I know you're on LinkedIn. I don't get the comments don't come across. And I didn't say good hi to Michael earlier either from Atlanta in Georgia. Right, we have 10 more minutes. Let's see who else have we've got. Let me say hello to Sunil who is on LinkedIn as well. Welcome to the session. In fact we have quite a few on LinkedIn who are commenting. Carrie is there as well and I think Carrie has a question on LinkedIn. Carrie says so thank you Sarah for sharing this. Is there a template for this to put into AI ChatGPT so that we can create these reports for our firm? Yes, I mentioned it earlier. Carrie, I've created many, many standard prompts for accounting firm owners like that M2, M4 are really good computers where you can run local LLM and those prompts are part of the prompt library software which you can have as now there are some problems, one of which is you don't have access to the Internet which you don't want a free gift as part of the AI academy. So if you join the AI academy and I know from feedback from members, one of the things I love the most about it is the prompt library software because not only do you get because you're keeping them local to solve privacy and security.

One of the big problems I found dozens and dozens and dozens of prompts that I've written specifically for accounting firm owners. You can also create your own prompts and you've got one centralized place where you can manage those prompts. There's all sorts of facilities whereby you can just filter to the prompts that you've written and also you can favorite prompts and many, many other things to make it really easy to manage your prompts. Manage the prompts that we've created for you and manage the prompts that you create yourself. And if you create them yourself, they're not shared with us. Okay. They're just for your use. Is that even some of the better models like Llama when it comes to complex financial data analysis, it spat out Ross. Right? Is there. And yes, there's more questions, I think. Let me just say, get on LinkedIn. Shahzad's here as well on LinkedIn. Great to see you. So we have a question from Organized Books, so I'm not sure who that is. Perhaps I should know. I feel like I should. Anyway, I apologize. The question is, do you use the same ChatGPT account for both business and personal use? Sort of, kind of. So I, I did have A plus account, but we moved it to teams account as soon as it came out, which I think was probably a couple of years ago. So I've been on the teams account for a while, but I also use that teams account for my personal use as well because as I said.

When you move your account to a team's account, it's not your, your history and your prompts are not shared with the rest of your team. So you can still carry on using it essentially as your personal chat GPT. It's just the, the, it's just part of the team umbrella. Rubbish. It didn't calculate ratios correct and so on. So I was getting a little bit frustrated with that. You benefit from all the things we talked about with privacy and security and the centralized billing. You just get one invoice. But then the biggest announcement of August 2025 so far, everyone thinks it's ChatGPT, which by the way, with the teams plan, I believe it's still you pay for it annually in advance, whereas plus you pay every month. So they're both approximately 20 bucks per month. But one, you do have to pay annually. That's the teams plan. I don't think that's changed. So, yes, GPT5. GPT5. No way bigger than GPT5. Is OpenAI released GPT I do use it for. Oh, it's Karen. Hi there, Karen. Great to see you. I thought I should know it. Yes. So I use it for. I use it for both.

But I also have, I have a pro plan as well. Now Pro plan is the 200 bucks a month version and the only reason I have the Pro plan is because very often they release things first to the Pro plan before filtering down. And given I teach all of this stuff I need to know what OSS O S S and that's their first open source model and we can coming along down the pipeline and I still get access to things that you don't get access to. So back in January they launched Operator and that's never been available to anybody else because Operator was really flawed. Operator was their agentic, their first agentic model and it didn't work very well. And then of course in middle of July they released Chat GPT Agent which is essentially a combination of Deep Research and Operator. So effectively you do have Operator now, but it's inside Chat GPT the agent model. Anyway, sorry, that's me waffling now. That's just me saying I have a pro plan because sometimes I want to check out what else is coming so I can then create training for when it gets released to the teams. Plan. Run that on a computer. There are two versions. There's the 20 billion version and the 120 billion version. The big ones just two. You need a huge computer but the GPT OS tested that out plus plan and free plan. Okay, we have five minutes left and I'm looking for more questions. I can't see any questions at the moment.

I'm sure there should be questions but I will give it just a few seconds. We never normally finish these early. We have five minutes left but I'm happy to finish early if there are no more questions and I've run my financial reporting prompts through that and it gives if anything better we can finish early. I did mention earlier that I have just done another video which I will be posting uploading sometime soon. I think it's about 25 minutes and it'll be uploaded on social media somewhere this week. I will tell people about it in the various channels. So the video I did earlier was talking about results than the one I showed you. GPT OS is an extraordinary how you can get better results from ChatGPT5 which is interesting. Recently OpenAI released some guidance notes on how to use GPT5 because it turns out, which is interesting. It turns out that prompt engineering is more important with GPT5 than with with 4.0, which is really interesting. I've been teaching prompt engineering for over two years now. Two and a half years, which is basically how do we craft prompts? How do we get better results? You saw from the example earlier, how do we get better results? Will we create better prompts? And when GPT5 came out, my first thought was is it more smarter now? Does that mean we don't have to be as detailed in our prompts? Can we cut some corners? Well, interestingly, and I think this is one of the reasons why people have been reporting it doesn't always give us good results. Interestingly, In a recent OpenAI blog, their research team discovered that and have reported that with GPT5 we have to give it better structured prompts than with GPT4O. And as long as we know how to do that, then we will get extraordinary results. And that's what I'm finding. I'm finding I did something earlier today. I just improved one of my old prompts in the prompt library for doing, for transcribing things, tidying up transcriptions, which had gone downhill when GPT5 came out. So I built in some of the things they're now recommending and wow, the output is really, really good. So if you're one of those people that's been frustrated with ChatGPT5, please do not give up on it. We've just got to learn, learn or relearn some skills and just learn how to use it. Because it's a fundamentally different model. It's a big jump, probably the biggest jump, but when we move from 3.5 to 4, it was an improvement in the model, but we didn't have to, we didn't have to relearn anything when we went in re model and we can do, if we can get, have a system set up where we can run an LLM like that locally, then we can analyze any set or as much data as we made. 2024, it went from ChatGPT 4 to 4.0. Again, it was a huge G want because everything is sat on our computer. Okay, so if you jump in capability, but we didn't have to learn anything, now we've gone from 4.0 to 5. You want to find out more about that? If there's only so much we can talk about today, this is still a big issue. But if there's a bit of a learning curve and we're going to discover these things, it's been out as I'm speaking, it's been out for what, two, two weeks or so. And if you've been following the news, it's also done some strange things. So chatgpt OpenAI keep changing it as well iterating it. I think it's going to be an extraordinary model. We just have to figure out how to get the best out of it. Okay, I can't see any more questions which is fine. So I will, I think we'll wrap up here and, and if you've got more questions for me, then join the AI Academy. Remember there's a five day, sorry a seven day free trial that will keep that turned on for the next few hours. So make sure you grab that chance. If you're not in it, go in there and it's free for the next five, six days to you and then you can make a decision whether cancel or not. And if you do, it's not cost you a penny but you'll find some great information and great courses in there. Okay, I will see you perhaps on my next live stream, which will be next week I think. Okay, I'm going to play with some music so I can read your comments. If you've enjoyed today's session, give it a thumbs up and I will see you soon. Goodbye for now. You want to go deeper on privacy and security if you want to know if you want to know how to get an AI policy, what to say to get agreements from your clients, and even more important, if you want to know step by step how to run a large language model locally. It's one of the many things I cover in the AI Academy. So if you want to go deeper on this, if you want to learn more about privacy, security, client confidentiality, how to solve it, then I have a complete course in the AI Academy. I'm going to ask my team to put a link to it below. I know there's that some people are here already in the AI Academy and I have a very special announcement. The AI Academy is just a small monthly payment. It's just 59 bucks a month and you get access to a community of hundreds of accounting firms that you can ask questions of me and them and learn more. There are in the Academy there are dozens of courses that I've created, including the Privacy and security course, which includes a recent addendum on how to run a large language model locally. And also members of the AI Academy get a completely complimentary copy of the software that I created, the prompt library Software with over 100 prompts that you can copy and paste specifically for business owners. Most important, specifically for the owners of accounting firms. There are prompts for financial data analysis, for business advisory and strategy for marketing, for content creation for copywriting, for creating SOPs systems and processes. There are prompts for creating your value, pricing and so much more that I've written over the last couple of years specifically for my community and the accountants that I work with. And you get that as my gift. When you join the AI Academy, it's only $59 a month. The price will go up in the future, by the way. But one of the things people love about our academies is that the price never changes for you. When you join today, you get the special.

Price, which will never, ever change. And I keep creating more content. But if you're not sure, if you are not sure yet, then what we've also done for the first time today is we have decided to give a seven day free trial and that will be available for the next few hours until first thing tomorrow. So if you want to trial it out, Sarah's going to put a link in the.

Link in the chat, which I think she's done to go to that link and you can find out much more about it and then get your free trial and you can then check out the various courses for the next seven days. Then anytime before seven days, you can always cancel if you want. We've never done that before. You will have to do it in the next few hours. By tomorrow morning we will turn that feature back off again, partly because we're testing to see how many people are interested in a free seven day trial. So Sarah's put the link into the chat. Go and grab that. Right.

Let'S answer some questions. But before I do, just a few final things. If you've got, if there's anything you've learned today that's been valuable, if you could let me know by just hitting that like button, I'd really appreciate that. If you're not yet, if you're not yet subscribed to our YouTube channel, then do that and then you'll get notified of new videos. Also another thing to tell you about is next week we have a big event, it's Thursday of next week and that's the AI Summit. We run that a few times a year. It's a three hour webinar where we go much deeper on all things ChatGPT and, and AI and use cases. And the great news is you can get a completely free ticket for the AI Summit. So if you want to go deeper on this, make sure you get.