Mastering Financial Data Analysis with ChatGPT

Advanced use of ChatGPT with a particular focus on financial data analysis. But particularly what I want to talk about is I also want to share with you what is a huge pitfall you must be aware of, otherwise this could be a huge problem. So there is a big pitfall, a big problem with Chat GPT at the moment. And I will show you what that is and what I will also be doing today as well. I want to share with you a relatively new feature feature in ChatGPT. It's been out for a week or two now and I, I, in my opinion it is the most powerful thing they have ever done. It's just extraordinary. It's, it's making a huge difference to us. So we've got a lot to cover. So with that said, let's dive in to today, today's content, today's session. And so what we were we're talking about with advanced use of, of chatgpt we're looking at advanced reasoning which is something they've been developing over the, the last few months. They launched a few months back at the back end of last year their O1 model and more recently they've launched their O3 models. And these use reinforcement learning and essentially this means that when we are using Chat GPT there are now essentially two types of models that they have. And when you go to do a brand new prompt in ChatGPT depending upon the account that you've got. So for a lot of the stuff I'll show today, you will need to have the plus account or even the Pro account you will need to use have those. But as many of those you, as most of you who regular attendees of my sessions, you'll know that we can do some extraordinary stuff with the plus account or the teams account. So the $20 a month is extraordinary low for how much time it saves us and the results that we can get from it. So they have these, they've now got two different types of models, they've got their GPT series which is the one that we've are familiar with now and they launched GPT4O back in May of 2024 and that was a huge step jump in the ability of ChatGPT. And then just in the last few weeks they are now rolling out chat GPT 4.50 and you will definitely have that if you have a plus account. And I've been using, I've been using ChatGPT 4.5 and comparing it with 4 and just like the step jump last May, I'm finding 4.5 is a lot better than 4. So if you want to get the best models, yes, it's absolutely worth getting the $20 a month US dollars a month version because that also then unlocks other things as well. So we have our GPT models which they are continually improving and GPTs are GPT4O for example at the moment is the default. It's the one that you'll use most of the time, it's the one that you're familiar with. And if you regularly come to my sessions, when I talk about prompt engineering, it's the one where the way I teach prompt engineering, that applies to the GPT models where you give it a Persona, a goal, a task and you might upload templates, add templates, examples and so on. But then we also have these advanced reasoning models such as O1 and O3. There isn't an O2 by the way. You might be wondering why they went from 01 to 03, I believe, I don't know how true it is, but in the UK we have a telecoms company called O2. So someone said to me that they skipped O2 just to avoid any potential issues with O2, the mobile telecoms company. So Zero1 was the first one that came out back end of last year and then they've just brought out O3 and they have two versions of that, O3 mini and O3 high.

So what is the purpose of advanced reasoning? Well, the purpose of it is it has step by step reasoning, it thinks things through. And so it does mean that when you send off a prompt, it can take a little bit longer to get the answer because it's going through a process and it will also, before it gives the answer, it'll tell you the thought process it's going through. And so because it has this step by step reasoning, it comes up for certain things, it can give you better answers. What it's great at is breaking down multi step calculations. It also self reflects on errors and so if it realizes going down the wrong route, it'll correct itself. It's great for decision making processes. So the question then is okay, so when do we use O1 or O3 and when do we use. When do we use the. The one we're familiar with, GPT4O or even 4.5? Well, essentially I would suggest you would use the advanced reasoning if you want to do something a bit more complex where we want it to think things through and that might be something like complex financial forecasting and analysis. So what I want to do at this Stage is I think it's easy for us to show you and I know some people are listening to the podcast so I'll try to describe things properly. But you might want to go to the YouTube channel and if you want to, if you're a visual person and want to see it. So I'm going to show it because it does make life a bit easier to understand and I'm going to show a few things if I just go across to ChatGPT. So you'll see that when you start a brand new chat to the right of it, you've got the drop down box which where you can see the different models available to you. Remember, of course, this all depends on the account you are on. If you're on a free account, you won't see these options, you'll have less options. So I always recommend the plus or the teams account. I'm on a teams account. Our entire team has access to this. We all use it every single day because it saves so much time. The value to us in our business is so big that the US$20 a month is tiny in comparison. So we have the teams account. Everyone has a copy of it. So you would then choose, when you start a brand new chat, you would choose which model to use by default. It's probably going to normally be GPT4O, which is the thing that's the best for most users. But sometimes we might want to use 0103 and there's two versions of 03, Mini and Mini High at the moment. Now the Mini High version is supposedly better for coding and logic, particularly people doing code. So given that we're talking about financial analysis, I'm not going to look at the mini high version. We'll stick with the O3 mini, which is faster as well. It gives us faster results. So the first thing I'm going to show you the first prompt I've done. I've already cut and pasted it in. Let me just get rid of that. I've cut and pasted a prompt in here. Let's imagine that you've got a profit and loss account for a client and let me just. I think I can open it up just to show you it. I'm going to show you the profit and loss account. If I could just type in the right password to my iPad. Here we go. So I've just got a profit and loss account. It's a fictional business by the way. It's made up numbers. There's nothing confidential here. I have a, a business that's a fictional high street seafood restaurant in a and I know Jim's here. Jim kindly supplied me with this case study. So thank you to Jim Rayner who's here for this case study. I use it frequently and so I've got the profit and loss account and I have the balance sheet. And let's imagine that I've got this and I want to create a cash flow forecast for this business going forward. But perhaps I don't have a huge amount of experience in cash flow forecasts. I'm perhaps not sure how to build it, perhaps not sure what questions or what information I might need. I've just got the profit loss account, the balance sheet and I want some help with building this cash flow forecast. So I might create a prompt that's something like this. I have uploaded the profit loss and balance sheet for a high street restaurant and before I forget, I might send this off without doing it. Let me just drag in that profit and loss and balance sheet. So that's just uploading right now. As you can see the the pink, pinky orange button. I then got the goal. The goal I want to create a cash flow and profit forecast for the year. Now one of the things with these advanced reasoning models when you are constructing your prompts, one of the things that we are told by OpenAI is that it's more important with advanced reasoning to stress the goal of the goal that you're looking for. The end. Yes, when you create a prompt inside GPT4, it's a good idea to specify your goal, but it's even more important when you're using advanced reasoning. And the reason why that OpenAI tell us that is because of this step by step reasoning it goes through. If it knows what your end goal is, then it will in its thinking process it will make sure that the solution it comes up with will be working towards your specific goal. So that's my first tip with your prompt engineering. With advanced reasoning models, always specify what is the goal that you're trying to achieve. Then I have the task which is to analyze the information carefully and create a draft forecast. Think carefully by each line item and tell me what information you would need to more accurately model the forecast. And then I've given it a template or structure for my response. I wanted to do a draft profit and loss account, but I know it's going to make numbers up. But just bear in mind, imagine that I'm not familiar with doing forecasts and I wanted to start the process. But then what I wanted to do is I want to tell me about what is the key information that I would need and some tips for building a better forecast. And then I've given just a little bit of background about this particular business. Okay, so there's my prompt. So let's hit this button and see what happens when we use O3 mini. And as I scroll down you will see it says reasoning. It's now thinking about the goal. It's also giving me its thought process and it's thinking through the process it needs to go through. And one of those it needs to build creating a forecast draft. You can see it's constantly telling us the process that it's going through in the thinking, which I think is really interesting. And this is why we use it for step by step reasoning for things where we wanted to think through carefully. And OpenAI tell us that one of the things we don't need to do in our prompts is use chain of thought. This is the idea that when we do a normal prompt, we might say in GPT4, think this through step by step, we might build that into the prompt, but we don't need to with advanced reasoning because these models use chain of thought. They do think things through step by step as you saw. So it reasoned for 31 seconds. Now, whereas normally if you use GPT4 you, you will get a result in sometimes just a couple of seconds. It's pretty much got there. So let's have a look at what it's done and just see if this is any good or not. So it's done me a draft profit and loss account based on the data. And if we were to go back to the data, okay, where am I? If I go back to the data, the profit and loss account, then it looks like the numbers are in the right sort of ballpark. I've not double checked them, but they feel like if I look at flat food sales, it probably is about the right sort of numbers. But it's only just put an all it's done here is just filled in some draft numbers, which is what I expected and what I asked for. Of much more importance, it told us about some assumptions. So that's the profit loss, that's the cash flow. But then more important, it's then gone through and analyzed what information would I need to collect if I want to be more accurate with this forecast, what would I need to go back to the client and asked for? And it's going through each line item in detail. It's thought this through, it's analyzed all the line items in the P and L account. And it's suggesting that for sales, I would need to look at seasonality patterns, I would need to look at average spend per cover, would be a good thing to do, and planned menu or pricing changes. You see, these are things I might not have thought about when I was running my accounting firm. Is the client thinking of changing their menu, marketing plans or promotions that could drive volume changes, historical trends on booking reservation patterns? Perhaps we might want to get information for cost of sales around contract terms with suppliers and any anticipated price changes or expected wastage or spoilage rates for perishable items. We've got details about labor costs, the impact of seasonal variations on staffing needs. You can see it's going into a lot of detail about the things that we should think about collecting from the client that will help us build a more realistic forecast. And so if you're not familiar with forecasts, this would really, really help with identifying things to look for. And of course, because it's chatgpt, we can carry on the conversation. If, for example, in fact, I'll come back to it in a second, it's then got tips for building a better forecast. So we should document the assumptions regularly, update with actual stress test, include seasonality, do a detailed cash flow forecast, engage with key staff and monitor external factors. So that's what it's come back with, which I think is a pretty good response. And of course we can all always further question it. And so let me just share with you another little tip that many people aren't aware of with ChatGPT, which I think is really cool. Let's imagine, for example, I'm just going to pick something out here. Let's imagine if you're modeling sales, it's suggesting that what we do is we identify what is the average spend per cover and any planned menu or pricing changes. And I'm just going to actually do the average spend per cover. One of the things we can do is if we want further clarification, when you highlight something in the response, it gives you a little speech mark above. And if you hit the speech mark, what it does, it basically copies that in this case, average spend per cover down into above the chat at the bottom. I don't know if you can see that down here. So we could then just simply say.

Tell me exactly how I can get this information. So we can further drill down, we can get further assistance. That's what I love about Chat GPT, that if we need some further help, we can do exactly that. So now it's going and thinking through the process of how we might get average spend per cover. Now, of course I only highlight a three letter word, a three word sentence. I could have typed that in. But the point I'm making is that you could take for example an entire section like all of that, those three points there and say give me more detail on this without having to then type and list all those things again. So that's a little tip that I find really useful if you want to get some further insights on a part of the response that you get. Okay. And so it's gone through and it's now telling me a bit more detail how I might collect data around cover counts. Okay, that's the first thing I want to show you and that works pretty well. Let's move on to something different and something that I've noticed a pitfall and I've noticed this for a while now, for several months I've noticed chat GPT doesn't always do this really well. And so I'm going to. At this stage, let's imagine that we've got, I'm going to use the same.

Client, the same fictional client. But let's imagine now we've got the. Let's imagine something different we want to do with the data. Let's imagine that we've got the monthly profit and loss account for the last 12 months. Perhaps we've got it itemized per month and we want to analyze that and get some information about it. And so I'm just going to go in this case, I'm going to pick one prompt I did just a short while ago and.

Here I've got it. Let me just get it ready for you. Okay, so let's imagine that we've got, and I'm going to show it as well, the data.

So we've got this profit loss account for the same business, but it's now split monthly, the monthly profit and loss account and perhaps we want to do some analysis with that. We want to understand what's going on. Perhaps we want to get some analysis of some of the trends, some of the potential problems. Perhaps have a meeting with the client coming up and we want to get some insights from ChatGPT ready for the meeting. So what I've done is I've put in a prompt here which says you're a financial accountant skill. This is now by the way, this is in GPT4O. So I'm just going back to GPT 4.0 because I'm going to show you the difference between the normal 4o model and the advanced reasoning. So let's use 4.0. And my prompt is to say you're a financial accountant skilled in interpreting data. I've uploaded, uploaded the monthly profit and loss account which I did earlier and it's reproduced it here. And I did check it and it is correct. It did pull in the right numbers from the profit from the data. It's an Excel spreadsheet and I've asked it to do a number of things. I wanted to analyze the data and I wanted to summarize the result. I want to calculate the most important ratios for this business and, and describe the trends over the period. I also want to know what trends might give cause for concern, which I should bring to the attention of the business owner. And also five suggestions. And then I've given some context again. So that's the prompt that I've given in GPT4O. And what it's done. It's then it's put in the data here in a table, which by the way, we can expand if we want to. We can expand the table that will just. It's here. And one of the things I did earlier, I did double check, it has pulled in the data correct. It's taken the Excel spreadsheet, it's got the correct data and it's put it into a table. So that's a promising start. Okay. But it doesn't all go quite so well as you will soon see. So it's then told me what it's done, it's displayed the cleaned profit loss data for review and it's done. Explain what it's done. Remember, this is GPT4O. I'll show you the equivalent in O3 in a second. And then what it's done is. Here we go. This is what it wants to do. It summarized the results. Okay. And as I look at this, for example, the restaurant had a total revenue of 675241. And if I do check, if I go to check the data, it is correct. It's got that right. That's reassuring and good to know.

But. And that's good.

And then it goes through and does the key ratios. Then it's got trends that might cause concern. Now I've got. There's some issues here I'm going to come back to in a second. So hold that. Just bear in mind, there's a big problem here. You will see very, very soon. But let me just show you the prompt quickly. That's what it's done. So the question is, does Advanced reasoning do a better job. Well, let's have a look at advanced reasoning. So what I've done is I've just done the same prompt, exactly the same prompt, everything the same, but this time using O3 mini. And I'm going to do a side by side comparison of them. And let me just centralize that. Okay, so in O3 mini, again, it's pulled in the Excel spreadsheet, made it a table, that's the same prompt. And I can see here that if I compare the summary of results, it's done it in a different way. So that's the first thing you notice that whereas the GPT4O on the left was a little bit more the numbers, just stating the numbers factual on the right hand side, with O3 mini, it's done things with a little bit more analysis, which I quite like. So the restaurant generates an average monthly revenue of around 60,000 with some seasonal variation. Warmer months tend to see higher cover counts and increased spending. If I scroll down to key ratios just to see the difference in the output in this particular case, it's got a little bit more detail. It's done the calculation, it's shown what the calculation is and then it's discussing the trend with O3 mini. And if I then go to number three, which is the trends that might cause concern in this particular case. One of the things I like about the O3 response is that it's also added an action point and been a little bit more clear in what the client can do. There's more detail in the response with O3 mini. And then if we look at the suggestions for improvement.

And the five suggestions.

I don't know which is better. They are both actually very, very similar. They are different, different suggestions. So what do I conclude from this? What I conclude from this. Oh, and one more thing that O3 mini does, it also then asks some reflective questions. It comes back to us with some things that we should perhaps think about. So if I look at the response from the two, you would probably need to apply your own judgment of this. I think the O3 mini response is, is better, but not overwhelmingly better. Okay. And so the point here is that if you're on a free account of ChatGPT, we can still get a really good result with GPT4O with a carefully written prompt where we've put a template in there as to exactly what information we want to get. However, there is a problem I noticed when I and this has been a problem I've noticed for some time now, and perhaps You've noticed yourself. Let me go to the. What I'm going to do is I'm going to go to just one of them. Let me just go back to the O3 model response and I'm just going to open it up back like this. Okay, let me just close that. So when I looked at this and analyzed it, what I found when I looked at the calculations and is they weren't always correct, which is really worrying for us as accountants. It worried me sometimes it was rounding things too much. So, for example, the Trend is declined from 62 to 58. And if I was to go back to the data that's not quite right, it's close. But as I looked at it, I did a few calculations, whoops, wrong one. And it wasn't quite right. And when it calculated some of the other ratios, they weren't 100% right, which was concerning. And I also found it with the GPT4O. And I've noticed this several times that if you upload an Excel spreadsheet to analyze and ask it to do calculations, whether you use GPT4O or whether you use the advanced reasoning, I have found that sometimes it, it does either just simplify the numbers or you can just get them plain wrong. And I find that really concerning, particularly when it took the Excel spreadsheet, put the numbers into a table and the table was the same as the Excel spreadsheet. So that's the big pitfall you must be aware of as an accountant. And that really worries me. Now there is what I have found over the last three or four months is if instead, rather than uploading an Excel spreadsheet, and I'm just going to go back and pull a third one.

If I go back to ChatGPT, I've done the same prompt again, exactly same. But this time I uploaded the data as a PDF, which is the PDF that. Which is exactly what this is. I uploaded the PDF and ran the same prompt. And when I did that and it went and did all the various analysis and calculations, I double checked and the numbers were correct. So for example, the gross profit margin, it's 62% and it said it starts off higher, it's rounded the numbers a little bit to the nearest full percentage point, but I checked the data and the lowest point was 59.5%. And so what I'm concluding from this is that when you use PDFs of financial data and analyze them, ChatGPT seems to do a really good job of understanding the numbers and doing the calculations and giving some good answers. As soon as you upload an Excel spreadsheet.

Be careful, be careful. You might find that it's not giving you the right numbers. And the problem that then gives us is if we don't, if we don't have complete faith in the numbers, then we have to go and check it all. And then we think, well, why do we use ChatGPT in the first place? So that's my big warning. Just be very careful of that. I think ChatGPT is an amazing tool for so many things, but when it comes to analyzing financial data, I regularly find analyzing a PDF of the data, it does a much better job than an Excel spreadsheet. That's my word of warning. So I hope that's useful to you. Right? What I want to do is, is next is well let me give you some more use cases of these Advanced reasoning so we've looked at complex financial forecasting analysis and so on and it's good at that. But just be careful with this point about Excel spreadsheets. It's also great for any other thing, anything that needs thinking things through. So I find that advanced reasoning is good if we want to do for example strategic client advisory and business recommendations. So if you're doing business advisory then you might find you're getting more detailed analysis with O1 and O3. If you want to do audit assistance and data integrity, then again it should be better for that subject to the point about whether it's analyzing the Excel spreadsheets right or not. And also any kind of form of complex project planning you are doing where there are lots of steps you have to go through. So it's the complex stuff where advanced reasoning is usually better. However.

In all my testing I am not finding it.

So much better that I'm using it as my default. I usually for most of things I do, I find GPT4 is really, really good. And I think as long as you know how to craft great prompts, it works most the time really really well. But you need to test. What I would recommend you do is when you have an advance, something that's a bit more advanced, that needs some thinking through process and you have access to those advanced reasoning models, just run the prompt twice, once with GPT4, once with O1 or O3 and then see which one you like the best and you might find just like this side by side comparison I showed you. I like bits of both to be honest. You'll need to decide some people might love this style of the advanced reasoning. Some people might like the what comes out of chat GPT4. So with that said, let me move on though to something that's a little bit more optimistic and upbeat about Chat GPT's new stuff. Whilst I have my reservations about O1 and O3.

I think the single best thing they have ever, ever done and they've done some great things over the last since it was nearly two and a half years ago when it was first released to the public. The thing that they brought out very recently that I think is the single best thing is Deep Research. Now, Deep Research is.

An even more advanced version of advanced reasoning and initially when they brought it out about a month ago, it was only available to to their brand new Pro account. So back In December of 24 they launched ChatGPT Pro account, which is 200 bucks a month, 10 times the price. And you probably would not have a use case to justify spending that I did because I want to make sure I advise you on the latest stuff and let you know what's coming. And I'd already heard from a friend of mine that I'm seeing on tomorrow that he did the same thing and he said as well that he is just amazed at what Deep Research is doing. So I went to grab that and it has not disappointed. I've always found with ChatGPT. I love it, I love it, but very often I find things that disappoint me, like uploading Excel spreadsheets, for example, disappoints me. Most of what I use it for is amazing. But some things disappoint me. Deep Research has never disappointed me in the times I've been playing with it. So the bad news is initially it came out just on the Pro account, the $200 a month. But they did a couple of weeks ago they did release it to the plus and Teams account. They're rolling it out. You should have it if you have that account and you should have it. And if you do, you'll know that you've got it. Because when you do a new prompt in ChatGPT with let's say 4o, you will see underneath the prompt it will say Deep Research. Now, the one problem you've got with the plus and Teams account is you've got restricted use of it. So you can only use it 10 times in a month.

And I'm using it way more than that. So I'm glad I've got the Pro account. But you might find that 10 times in a month is more than sufficient. What it can do is absolutely extraordinary. I'm Going to give you a couple of examples. So let me just pull up the first example. I'm using it for all sorts of things, anything that you want to do research on and you want to get some really well thought through research done with all the citations. You want a longer report because normally one of the issues with ChatGPT is that you tend to find the output you get is relatively short because of context, windows and so on. Well, that's completely different with deep research. So to give you an example, this is nothing to do with business, but I use it for lots of personal use cases as well as business. I'm going to give you two or three ideas just to hopefully spark some interest. So I'm very much into health and one of the things that I came across a little while back was the health benefits of drinking hydrogen water and I wanted to find out more about that. I want to know what the health benefits were. I want to know if there are any downsides, I want to know how take it. Because there are two. You can either take it with a particular type of bottle or you can buy tablets, magnesium tablets. I just wanted to know as much as possible about hydrogen water so I could make a decision about whether that's something that I wanted to start drinking instead of normal water. So I went to ChatGPT and I asked it to research that and it came up with an answer. Let me show you very quickly the response. Response and you'll see this is very different to anything you've seen before from ChatGPT. So here's the response. It's very, very well written. This is written as if it's written by a human as a research report and throughout it's got references to scientific research reports and so on. If I just open up a summary of the pages, this is a seven page report and towards the end you can see, you've got. If you can see in your blue text, it gives you the clickable links to all the scientific studies and so on that's been done. But this is a amazing, highly detailed research report that's really well written. So if you want to learn something.

That'S important to you, for me, Health Hydrogen Water is one example, then this is really, really powerful. What else could you use it for? So I've used it for. My uncle is very much into researching our family history and he spent a lot of money and investing in platforms to research the family history, but was always frustrated because it would always go back only so far and there was a little bit more vagueness So I went in and crafted a prompt with deep research to go and research the history. I just gave it a bit more information about what he'd uncovered and I sent it back to him. And he was amazed that there were things that came out that he'd not found in a year of researching. A year of researching his family history. It came out with that report, which was really interesting. So that's another great use case. And the last use case I'll share with you, which is, for me, more of a business use case. And then I'll pose a question and to think about. So I'm just going to go back to one I did earlier today.

Let me just. I'm just opening up so. Because I want to make another point about.

Deep research. So I was. What I was researching is one of the things that we teach, as you know, myself and Emily, is, is AI. And things are moving so fast that at the moment we have a Facebook group with about 10, 11,000 people in it. But most people don't see all the things we post there because of the algorithms on Facebook. So as a community.

It'S not as successful a community as we would like. People don't see what we're sharing with you because of the Facebook algorithms. And so I was looking at and researching other code communities that we are likely to launch in the next couple of weeks. And we're going to be launching a free community for you. And if people want more stuff, we'll have a second community where we're going to be able to. Where you're going to get access to the Latest thinking in AI ChatGPT for accountants and bookkeepers. Anyway, the platform I had in mind, I wanted to do some research on how best to use it and how to set it up, how to get the best out of it. So I crafted a prompt. And what happens is when you do the prompt, you just use your familiar GPT4 or 4.5, but you hit the deep research button at the bottom. Okay. That's what you do. You hit deep research and you post your prompt. And what it does is before it comes up with the answer, it asks you clarifying questions, which I think is really useful. So even though I think I thought I'd given it lots of context, it came up with other questions that I thought were really useful. So I gave it the answers. And then what it does is it goes away and does the research. Now, this one was quite quick. It was six minutes. Normally it takes between 10 and 30 minutes to do the research. Which is very different to what we're used to where it can be just one second. And even with the advanced reasoning models as we saw it was 31 seconds. Deep research.

Typically is 10 to 30 minutes because it really does go deep. It goes and searches for all sorts of sources and uncovers a huge amount of stuff. This was six minutes, was the fastest one I've ever done. I was a bit surprised. And what's also fascinating with it is that as it's giving you as it's going through the process you can see on the right hand side you can see all the sources it's going to but you can also see its reasoning process which I actually found fascinating. Where it's telling you what it's thinking. It'll say I'm now going to research this website. And then it might say okay, this doesn't seem to be very useful. Let me try this website. No, I don't think that's giving you what I want. Let me try this. Ah, this looks promising and it's fascinating. You can see the thought processes going through really, really quickly. But this is why it takes 10 to 30 minutes because it's going highly, highly detailed. So what I'm finding is the results I'm getting back are extraordinary. In this particular case about this community you can see again this is a really in depth report. It's 26 pages by the way. I put it into word document to read is 26 pages with towards the end all the links and the citations, everything I need to know. Here you go, citations and where it's been to the. It's gone to 14 different sources. I find this absolutely extraordinary. And so how might you think about using deep research? Well, anytime you want to do research.

Whether it's tax research might be something whether or if you want to write a research paper for your clients, then we've done a few of these, we've got a few papers coming out very, very soon and yes we've done editing, we've changed them but we'll be sending it, we'll be producing some research guides and a lot of the work was done by Deep research and we've not had to edit it too much because the output's just so good. So there you go, Deep research. Definitely check that out, make sure you are on that's for me.

That is reason enough to go for a Plus or Teams account and pay the $20 a month. I'm blown away by it. I've not had any disappointments whatsoever. So go and check that Out. Okay, let me now see if, if we've got any questions or any comments that I've missed. I know we've got some people that are on LinkedIn and for some reason the LinkedIn comments don't come across. So Sarah's post them for me. So let me just acknowledge you on LinkedIn, we have Virgil is. Virgil's here on LinkedIn. Hi there, Virgil. Hi there, Gary. Good to see you. And I don't know if I pronounced this right, but Nidia, is that right? Good to see you. So everyone on LinkedIn, you are welcome as well. And if you do ask a question, Sarah will pass it on to me. Right, Randall, what does Randall say? He usually says really good stuff. The plus account is worth it. He says it's 29 to $30 in Canada with exchange being the way it is. Okay. Yeah, so he says it's worth it. I think it is. I know Randall's a power user and uses it all the time. I think it's worth it. Partly because of deep research, partly because you now will get access to GPT 4.5, which I use most of the time now as my default. I find it significantly better in terms of both the quality output but also the writing. So if you want to create, I don't know, you want help with writing a blog post, I'm finding that the, the response is more human, like, so 4.5 is great. And then there's lots of other great things, features of the plus account, like you can create GPTs, for example, which is really, really powerful. Right, who else have we got? Sunny is here for the first time from Florida. Welcome to you. And Tamar's here from the uk.

And then Jim has made a comment. And by the way, we will be finishing early, as I said earlier, if there are no questions. So please, now is your time to get your questions in. What do you want to know about ChatGPT, AI or anything else that I can help you with? Because if there's no questions, I will wrap up early. Jim said neither model has spotted the increase in utility costs from August onwards. That was because in the UK there was a big jump in energy costs. So, yes, Jim, it was Jim who created this, this case study, who is familiar with it. And yes, it can sometimes miss things and I guess that could be corrected by a better prompt. My prompt, by the way, was quite short and as you know, when I teach prompt engineering, if you do better prompts, you'll get better results.

So what I could have done in that first one is I could have specified like with that second one just like the forecasting one I could have said I want you to comment on every single line item. I didn't do that. If I had a done if that was built into my prompt I probably Jim would have got a better response and it would have looked at utility costs. Sarah Bridges also says the paid account is definitely worth the small payment amount. Okay. I don't know your name but somebody says as an accountant I feel going to be redundant soon. Okay so my thoughts on that. I know you won't but what I've also said and you might want to if you didn't never seen it Back in early February I did a live stream. It's on my YouTube channel on my predictions for the future of the accounting profession and what are the three things that we need to do to stay relevant. So if you missed that, go and check it out. I think it was my, it was my first live stream in February. What are the three things to stay relevant and or come along to our AI summit on 25th March. If you missed that Sarah will in fact at this point let me just share that link again just in case anybody missed it. Perhaps Sarah could share it. We are running a in on the 25th of March. It's going to be about three hours. Emily and I will be going much, much deeper on this on all the things you need to know about AI and ChatGPT. If you're an accountant or a bookkeeper and we are giving some free tickets away if you're, if you're in my value pricing academy, you don't need to register, you'll get the link anyway. So make sure you grab the link either with a QR code or Sarah will put the link in the chat.

And also if you've enjoyed this so far by the way, just give it the thumbs up. I'd really appreciate that. Feel free to share it and so on. The more accountants I can help with this stuff the better. I'm on a mission to make the to help the profession. So going back to what.

I don't know your name so I can't, I can't give you the but info said I feel going to be redundant soon. So back in that live stream in February, one of the things I said was that and having followed this space for two years and seeing the change and for example just in the last three months the changes in ChatGPT with all the new features they launched in December and all these the models I've shared So far, there's been more change in three months than the previous two years since it went live to the public in November 2022. And that reflects AI. Generally, the speed of learning is extraordinary. And so, yes, a lot of people are going to find that they will be redundant. And we are already seeing that. There are news items pretty much every day I see of large companies making mass redundancies because AI can do the job much, much quicker and much cheaper. And accountants and bookkeepers are near the top of the list from all the surveys that are done around this. So perhaps we should worry. But I'm an optimist. I see a huge opportunity. What I do predict is that by the end of this decade, and well before the end of this decade, by the way, but let's go with the end of the decade. Let me not depress you too much. But bookkeeping as we know it today, the data entry, the reconciling bank accounts, and much other compliance work, like preparing tax returns that will disappear, the AI will do that so well within a few years from now, we will not be needed to do the traditional compliance work that we've done. Now, some people challenge me on that and say, yes, but it makes so many mistakes right now. Yes, it does, right now, today. But you just have to look at.

The huge exponential speed that's taking place right now across all AI, whether it's large language models, whether it's image generation. Even now, we can create videos. It won't be long before with Netflix, we'll be designing our own videos. We'll just put a prompt in and say, I want a video featuring Harrison Ford based on this. And it will create the video for us. That's where this is going. If you look at robotics, it's extraordinary what's happening right now. So, yes, much of what we do today will disappear. But I'm also an optimist. As long as we stay ahead of this, I believe, and I've said this many times in, it's not AI that will replace you. What will replace you is how your competitors are using it. And I know there's people here like Randall, using it all the time.

When you are making use of this tool, and it's particularly good for doing things like strategy, we can add so much more value to clients through things like business advisory using AI, it's extraordinary. And so that's what we have to do. We have to move up the value chain, which is great, because that means we can become more valuable. We can deliver better solutions where we can charge higher prices and do it all for much less time. I think the profession is in a really good state right now. I think the next two or three years will be really exciting, but only for the firms that want to embrace technology and in particular AI, because many others to want will just disappear or just give up. So that's my predictions on the future. I hope that's useful in some way. Randall says A question do you think ChatGPT will get better at writing long form content such as chapters for a book? Yes, I do. And as I showed you Randall, as the Deep research can do long form content, it's all about the context window. They measure it in tokens and the more tokens that are spent.

That costs OpenAI or whichever company you use. And I don't know if you one of the things that you can do, you probably haven't played with this, but as well as the front end that you see, you can go into the back end of your account, into the playground and you can connect up your API and build automations, which is great fun, which I've been playing around with, where you can create your own little app that have ChatGPT or Claude in the background. And when you do that, when you hook up to an API, then every time you make an API call, you get charged on your credit card. It's not a lot. And so the reason they're having to pay, they charge the 20 bucks or the 200 bucks, particularly the 200 because of the amount of tokens used up by Deep Research. They have to it costs them money every time somebody runs a prompt. But there's no doubt that the cost will go down, the context windows will go up and we will see that we can get longer responses. Randall and what I'm already seeing is that I'm getting better and more thorough and longer responses with GPT 4.5 compared to 4. So it's absolutely going that way. However, when it comes to writing a book or an ebook, I still think that.

Even if we could write an entire book, personally I feel that's cheating a little bit too much. As some of you know, when I teach content marketing like writing ebooks, I suggest you use the model to come up with the ideas and then to come up with the structure of the ebook. And then you'll use subsequent prompts to help you write each section by section. But you'll do it helping you rather than just doing it for you. Because I do think there'll come a time very soon when, when it does, it all for you. When you can have ChatGPT writing books for you, then everybody will do that. So what's the value in a book? And it's going to be the same, as I said earlier, with Netflix and films, it'll get to the point soon where we can just do a. We can just type in a prompt and then we'll get the exact film that we want and everybody will just watch unique film based on their prompts and then that will probably then devalue films. This is the way the world's going.

Right? Clive Stewart says, can you use deep research with 0103? If so, do you think it will produce better results, for example, better financial analysis? No, you can't, Clive. So at the moment, Deep Research is done inside ChatGPT. So effectively what happens is, and. And if you were to start a new chat and go to, let's say, 01. Oh, I'm just lying. Sorry. You can with O1 now. They keep changing this, by the way. You can do it with O1. Let me just check with O3 mini. Oh, you can with that now. Okay. They've changed it. They change things all the time. You have to be careful because I was finding, when I got O3 mini, for example, the very first time, I couldn't attach upload documents, but now they can, they've added those tools. So, sorry, you are correct, Clive. Yes, you can use deep research with 0103. Will it do a better result or not? I don't know because I've not tested it. I'd not even noticed that that's changed so recently. I don't know whether it's going to help or not, to be honest. I don't know because.

When you do it inside ChatGPT4, then you have your normal GPT4 conversations and then when you're ready to do the research, you hit that button. It then does the Deep research and then unless you hit it again, you can then have subsequent conversations and get it to explain something. Like I showed earlier, you could highlight a bit of it and say, explain this a little bit more. And rather than spending another 30 minutes, it will then do the usual really quick response. So you are inside the GPT environment for conversations that take place afterwards until you want to do Deep research. So I'm not totally sure what using O1 and O3 will bring to the party. Test it. Test it. I haven't yet. Hi there, Rebecca. You're on LinkedIn. Dipin says anyone use digits AI in UK? The UK. I can't comment on that? I haven't. But if anyone got any response for Dipn then please let him know we have four minutes. Okay? Love these AI sessions. Thank you for the feedback. Okay, I've got a question From Rebecca on LinkedIn. If you are just starting out, what plan of action would you take with GPT in the accounting field if you're just starting out? Firstly, Rebecca, I would suggest make sure you book on our event on the 25th of March because we'll it's going to be I think, I think it's three hours. We'll have a lot more time for answering questions on that and I can show you show you stuff and answer specific questions so I'm certainly come along to that. Other than that.

I assume I'm not sure whether you mean starting out starting at your accounting firm or starting out using Chat GPT. If it's starting out using using Chat GPT, my advice is always the best way to learn it and figure out how to get the most out of it is firstly go and grab some courses. I have lots of training, we have these live streams and we have courses that you can buy as well if you want to do learning but also just play with it. And if you're not sure where to start, I always find that the best place to start is this. Number one, start playing around with some personal fun stuff. So I remember about a year or two ago I did a little bit for a bit of fun. I asked it to create a quiz for me for a family Christmas quiz. Count the questions. I asked it to analyze some song lyrics. I love music so I asked it to analyze lyrics. So do some fun stuff just to start to see what it can and can't do. And then you might start doing it for some more serious important use cases personally. And then as you get more familiar with it and what you can do with it, then start using it for business purposes for your firm. If you I mentioned earlier you've got the link if you come if you if you get on our VIP list Every Monday we send out our newsletter and I said earlier one of the things people love the most about our month Monday Newsletter Our weekly newsletter is the prompt of the week. So if you're looking for some ideas for Chat GPT and how to use it in your accounting firm, every week we give you a complete detailed prompt specifically for the accounting profession. So just make sure you're on the list and look out for that every Monday. And also we've got some more stuff coming your way. I Mentioned earlier, I hinted we have a new community coming soon, a free one and one that's a bit more detailed if you want more training where we are here to help you navigate this space. I want to help as many people as possible through the community because it's moving fast. Okay, somebody else uses plus for all social media planning. It's been great. Virgil on LinkedIn, I have two minutes left, says, how does the Deep Research functionality compare to deepseek? Oh, great question. So Deep Seq hit the news a few months back because it's the Chinese company that suddenly was a sensation because, well, firstly they created something absolutely extraordinary for a fraction of the price of OpenAI and there's a lot of question marks about how did they do it? And they. Some people said they cheated. Anyway, shortly afterwards what happened was the OpenAI released deep research and in the testing that's been done, Deep Research performs better than Deep Seq. The advantage of Deep Seq is it's free. I believe it's free. I've never used it because I then prefer to use Deep Research. It's free. I like ChatGPT. I like the fact that because I've been using it for over two years now, I feel very confident that my data is secure. I trust them. I wouldn't feel quite so confident with a model that was done in another country like Deep Seq, where there's some question marks about how they managed to build it and where they got. Where they got the chips from to do that. Where do they get the data from? I don't know. So that's just me. I'm skeptical. But the feedback is amazing about Deep Seq and it's free. But as I said, research does suggest that Deep Research is now a better model and that.

Hi there, Rick Paine's here. Hi there, Rick. Long time no see. The only way you'll become redundant is to stop learning. Absolutely. You have to keep learning and you're all here learning, which is awesome. Okay, any other questions? We've pretty much out of time. Thank you, Tracey for that feedback. I can't see any other questions, but lots of comments. So I did say at the start, these are exactly 60 minutes. What I will do is close out with some music, gives me a chance to read any final comments. If you think of any more questions, make sure you register your free place for our online virtual summit on March 25 and do check out the other couple of links I shared with you earlier. Thanks so much for coming along. I will see you all soon. Bye. For now.