Why prompting is the future of accounting

0:00 Um, welcome to another webinar by Accounting Yard. Now, this webinar was designed specifically around prompting. Um, and this was based on the fact that AI is like right now is like one of the most important thing in the current, in the world right now, right? Um, because it's basically improving the level of productivity of people. And if as an accountant, you're not leveraging the opportunity that AI brings on board, you're potentially not you're doing yourself a disfavor, right?

0:37 And if you're looking at even just from value delivery to your organizations, opportunity to help you get your next job, leveraging AI plays a very critical role. And one of the key concepts that talks about AI or what AI comes to do is prompting, right? On my screen, as you can see, is, um, the price of a badly briefed AI prompt. Um, most recently, um, there was this particular article I read around Amazon, where Amazon was trying to link authors with book titles on their platform.

1:10 And they allowed the, they allowed the, the prompt. Not to say prompt, right? Um, whatever they had designed to run for roughly about five months, and it cost them over a million dollars, right? That's not a case of prompting, but that's just a case of when they had not set up the parameter correctly, and it cost the business a lot. Now, I think, was it, um, Uber. I was reading about the fact that they give their, their engineers free access to use code to, to kind of, um, design their programs or update their, um, their apps and build new things, that what they spend in roughly about two months was the budget for AI for the full year.

1:55 Yeah. What am I saying in this sense? That AI comes with productivity gains and advantages, but you also need to know how to maximize the value, versus also managing the costs of leveraging those tools. Yeah. Now, so the title of the webinar today is Prompt like a Pro, right? Um, and this particular session, we talk about what makes a great prompt and what makes a bad prompt, right? And we'll loosely do a demo of two live cases where we're actually demonstrating it using the prompt.

2:31 So that will help us get a strong understanding of if, how good your prompt is. One thing I'll mention to you upfront is that prompting as a skill is already becoming obsolete. And you'll be surprised I'm saying that and I was still creating a train around prompting. But it's important to understand the evolution of AI to appreciate the value delivery of what AI is doing. There was a point in time where prompting was a very important thing.

3:06 But right now, what has happened over time is that you have now models, right, which have now evolved over time and are very intelligent, such that your prompt is not what they need that makes them give the kind of output you give. But rather, what helps them perform best is what we call a concept known as context. Yeah. Now, even context, the it's also gradually context will never die, right? Context is always very important in terms of AI prompting or working with AI tools.

3:45 But we're getting to a point where there's a next phase, and that phase is called looping. And that is basically where you have like almost like three AI agents carrying out a task and they are doing a critique of the output and validating it until the final output becomes solid enough. So basically, you have like three or four personalities, they are doing they have divided the work into various pieces, and there's like there's a preparer, there's a reviewer, and there's like the person that says, oh, it's fine, that passes quality checks, I send this back.

4:20 Now, very soon, you'll be hearing that concept called looping, right? A lot, very soon. But it's not very common yet. A lot of people are not really aware about it. And but we will start from prompting. Yeah. It's not a bad place to start from. Then in future time, you'll hear things around, okay, how can I make use of of context, right? We'll do context today, you'll see how we'll leverage context in the webinar today, just to give you more context.

4:52 Yeah. Now, for those joining us for the first time, um, they're potentially you're hearing about Accounting Yard for the first time. But Accounting Yard has been in existence for almost a decade now. And we basically training accountants, helping them bridge the gap between theoretical and practical accounting. And we've done that by doing almost close over 15 plus courses and over the years, it trained over 5,000 accountants. Now, and what if you follow us on various social media platforms, you see us on LinkedIn, on Facebook, on Instagram, and on YouTube.

5:30 And you see us with various free materials. One of the key things we've realized or we are doing this consciously this year, is include AI in all our courses. Such that participants are able to see in very clear detail. What AI can do and practically how they can put AI into what they do. Yeah, and this webinar is one of such where we're going to show you exactly how you could include AI in your work by prompting it correctly and getting the best result out of it.

6:08 Yeah. Our mantra is simple. Share, learn and help. And we've been doing that constantly over the years and brilliant, making impact on the lives of accountants over the years. Who am I? So my name is Oladili, and people call me Deli, some people call me Prof. I'm co-founder in Accounting Yard. Intelligence Consulting is the registered business name of, um, that owns Accounting Yard brand. So sometimes when people want to make transfers to us, where it is Intelligence Consulting Limited.

6:46 But Accounting Yard is a brand name that does mutually the training side of things. I have over 15 years work experience. Um, I've done over 8 years in financial modeling. Um, I basically lead, um, the lead the finance of an active business and also provide fractional CFO for SMEs and startups. Now, I believe that, um, you will see today, live demonstration of how you see the impact of a good, a good prompt, could impact the kind of result you would see when working, um, with an AI too.

7:28 So today, what you see is that you see me, two tabs open. You see me use Gemini, and you see me use Claude. So which for those who are not very familiar with the lot of AI tools. Um, so there's what we call your LLMs. And they are basically like three top LLMs in the world globally. You have, um, Chat GPT, Chat GPT accounts for roughly about 50% of the market share of AI, AI models. You have Gemini, Gemini accounts for slightly over 20%, I think roughly between 20 and 30%.

8:15 Then you have Claude, Claude accounts for roughly about 10%. Yeah. Now you have what we call your open models, the likes of Kim K and, um, Devseek, right, which are China related models that are also used for as LLMs. Um, at the end of this webinar, I will share with you a practice file, which you would also try and run at your own end, at your own free time, so that you don't just see me do what we're doing, but you also have the firsthand experience of playing along with some of these data, which you can try on your own to see how, um, the concept around prompting works.

8:55 If you have any questions, um, feel free to send a chat to the chat box. Um, so that you can, you can, you can basically be able to, um, yeah. Ask questions as the session goes. There will be dedicated sessions where people can actually, um, trial themselves, um, the session where they can ask questions. But before I proceed, I'll pause and give the room for the opportunity for people to ask questions at this point and or, um, what the expectation is, right?

9:30 So when they're joining this webinar, what were they expecting? You can either drop your expectation as a chat in the chat box, or if you have a reason of hands. And I'll mute you to just give what your expectation was when you joined this webinar. Okay, in this call. Yeah, thank you. Good afternoon, everyone. So my expectations specifically for this webinar is that most times I deal with clients and they send me back statements. So from this back statement, I have meant to generate a management account on a monthly basis.

10:27 So yes, I use Claude to try to to convert the statement and try to analyze them, but I'm still not getting to a point where I could just transition it from the statements into the management account. So that's really why I joined here and I really want to have more knowledge in that regard. Okay. All right. Thanks. Thanks. Anyone else? Okay. Hi Daniel. Yeah. Good evening. Yeah, good evening. So, my expectation for this webinar is that I want to understand the prompting aspect very well because I've been using um cloud Gemini and also GTP.

10:42 And I want to understand how to prompt like a pro, because sometimes you give the AI some instruction, but it gives you different something, what is not what you are expecting. So I want to know how to understand the AI in order that when I give it a prompt, it gives me what I'm requesting from it. Thank you. All right. Thanks, Daniel, for sharing. Um, let me just read some of the things being said in the chat box.

11:04 Um, so like I have some context. So someone said, this my first time joining accounting a virtual. My expectation is to learn more about accounting, but I learned how to prompt like a pro. I expect to learn a good AI prompting and to become conversant with the broad AI knowledge. My intention is to gain insight into how to create a quality prompt and quality outcome. Yeah. Good evening everyone. My expectation is for me to be able to critique the outcome of AI because I have not noted.

11:37 I've noticed that if you don't know the subject matter very well, you might miss it. I want to improve my knowledge on prompting to become a pro in using AI. Okay. Okay. Hi, Nkeem. Okay. Hi Prof. Um, thank you very much for this opportunity. It's a very, very, very good one. Uh, in fact, I might interest everyone to know that currently I'm in the office and I have just two days to meet up with the half-year report.

12:23 Uh, I'm not supposed to be working today, it's Saturday, but I had to come. Um, this training will be very, very helpful for me, uh, to gain knowledge on how to use cloud to, uh, deliver my task as quickly as possible. So thank you. It's a great opportunity. Thank you, Prof. All right. Thanks a lot, Nkeem. Okay. Hi Ali. Hi, good evening, Mr. Delay. Good evening, everybody. I'm expecting out of this, uh, session, uh, to learn how to do a prompt engineering.

12:53 I've learned that if you don't get the question, your prompting right, you will not be able to get what you are expected, expecting to come out as a result. So I look forward to getting that, adding more knowledge. Thank you. All right. That's fine. Yeah, okay. So I think to a large extent, I think a lot of us have similar expectations of what this particular webinar would be. Now, one thing I'm going to take note of that we are constrained by time, we only have one hour, 30 minutes.

13:26 So basically, I'll basically do like only like two demos, right? Um, so but in explanation, I'll explain some things that I pay close attention to when I'm prompting. Then two, give us the opportunity to ask questions and we can potentially see, or a particular problem statement. And we'll think about how best to prompt AI at that point, right? Based on the particular problem statement. Yeah. Um, so the idea is, this is what we're going to cover.

14:00 One is, um, setup, right? Understand what a framework is. So what typically happens in prompting AI, there are different frameworks, right? I think there's CRAFTs, there's BRIEF, so many frameworks. But the ultimate goal is to ensure that you're providing AI with sufficient context to be able to handle the scenario or situation you're giving to it for which you want to help you solve. Yeah. Then we'll do a demo, um, where we are dealing with the manual account.

14:36 Ask questions, then do another demo where we are doing like a bank statement analysis. Um, then we ask questions around that. Then we finalize. Yeah. Now, a very popular analogy that people use for AI is that think of AI as that first class graduate who has finished from OAU or has finished from UniLag or from Harvard University. Right? and it comes into your organization to now come and walk. And the reality is he's very intelligent. Nobody can take it from him.

15:12 But the problem is that he doesn't have context around the business that you are dealing with. There's a high probability that because no matter how smart he is, if you give him a particular task, there's a high probability that you will find a lot of errors in it because he doesn't understand the business. Copy, connect, authorize. Yeah. Where is it? Sorry. Okay, good. Yeah. So this is it. I'm not surprised why I'm telling about. Not my first time using it.

6:50 Yeah. Good. Hello, good evening. Yeah, thank you for the for the class. I want to ask. Did you actually attach that management, that management report to the cloud before before you prompt it? Yeah. So this is the adding in Excel. So it does already something similar to what, um, the one you attach to the web, right? Let's, let's get to the. I have a simple version of the prompt. Yeah. I have a simple version here, right?

5:44 Um, preparing incomplete records for an engineering retail business whose only reliable record is bank statements. Attach is the full bank statement export. Right? Profile the statement, period, opening and closing balance, prove completeness. Re run the running balances. Classify every transaction into GL ready categories with a confidence score per row. Decide expense versus capitalised. Asset versus revenue payments. Flag pre-payments, accrual and deferred interest. Build an exception. Session list, duplicate, unclear narrations, possible personal personal spend with other reasoning.

6:49 Right? So this is like a simple version of that prompt. Yeah. But this is a loaded one, right? And this is just telling you that in designing or getting the best result in a prompt. Yeah. Sorry, sir. This token where for free people. How can, or is there a prompt that one can give it to, uh, enable it to manage, to encourage it to manage the tokens so that one does not run out of the analysis mode?

5:56 That's my first question. Then the second question is that, uh, is there co-work for free version of chat GBT and, uh, cloud. So the co-work is now available on the paid version. For chat GPT, I'm not sure if it has to be paid for.