How Two Students Taught 52 Auditors to Build AI Agents

00:00:02 The IIA

The Institute of Internal Auditors presents all things internal audit tech.

00:00:06 The IIA

What happens when 52 internal auditors sit down to learn how to build AI agents?

00:00:11 The IIA

And the experts at the front of the room are two college students who figured it out on the fly.

00:00:16 Mohamed Maiga

I just screamed, yes, yeah, I can do that.

00:00:19 Mohamed Maiga

not knowing.

00:00:20 Mohamed Maiga

And I had to go on YouTube for about 10 hours to learn about AI agents and how they work.

00:00:25 Ivan Diaz

Yeah, I had no idea what I was doing.

00:00:27 Ivan Diaz

I'll be honest.

00:00:28 Ivan Diaz

In terms of building an agent, I had no foundational understanding.

00:00:31 The IIA

Here's host Alex Rosati, starting with accounting student Ivan Diaz.

00:00:36 Speaker 4

So Ivan, can you help set the scene for us?

00:00:39 Speaker 4

You were in a room full of auditors at UVU.

00:00:43 Speaker 4

Can you give us an understanding of the event that you did and how did you end up in front of everyone presenting on this?

00:00:50 Ivan Diaz

So there's an annual conference for Utah State higher education for the internal auditors.

00:00:57 Ivan Diaz

And so what this event was is it gathered the internal audit departments from every higher education institution in the state.

00:01:04 Ivan Diaz

And my institution where I'm working at, they were the ones in charge of presenting, delivering.

00:01:11 Ivan Diaz

recommendations or ideas on certain topics.

00:01:14 Ivan Diaz

And one of the topics this year was AI.

00:01:17 Ivan Diaz

And prior to this, I wasn't on the list of presenters.

00:01:22 Ivan Diaz

I simply had created this agent beforehand.

00:01:24 Ivan Diaz

I expressed it to my managers.

00:01:25 Ivan Diaz

They saw value in it.

00:01:27 Ivan Diaz

And from there, they just asked if I would simply present what I built out, if I could give a run through.

00:01:32 Ivan Diaz

And so that's really where this started.

00:01:34 Speaker 4

Can you give me some background as to how you came about this idea and how you actually executed on it?

00:01:40 Ivan Diaz

I built it out of curiosity, really.

00:01:43 Ivan Diaz

I think as a student, you hear a lot about AI this, AI that.

00:01:47 Ivan Diaz

I'm someone who does listen and read to a lot of what's going on in the business world.

00:01:52 Ivan Diaz

And I found that there was a need to adapt to this new technology.

00:01:57 Ivan Diaz

And so in my current position, I'm an internal audit intern at my university.

00:02:01 Ivan Diaz

And so in that position, I see a lot of obviously the things going on around the university.

00:02:06 Ivan Diaz

And there was a process that works into our revenue work paper that takes departmental spending using purchase cards, travel.

00:02:17 Ivan Diaz

And we test that, obviously ensuring that what departments are purchasing is up to policy.

00:02:23 Ivan Diaz

And so I saw that.

00:02:25 Ivan Diaz

And the current process for me was a lot manual work,

00:02:30 Ivan Diaz

a lot of pulling from the accounting system, formatting in Excel.

00:02:34 Ivan Diaz

And with what I knew, I kind of just dove into it.

00:02:38 Ivan Diaz

I opened the AI and said, how can we automate this?

00:02:42 Ivan Diaz

And from there I ran with it and built an agent.

00:02:45 Speaker 4

So it's really interesting what you said around you prompted

00:02:50 Speaker 4

the LLM to be able to help you to build the agent.

00:02:54 Speaker 4

It's not like you had to have a playbook that you knew right off the bat.

00:02:58 Speaker 4

Is that correct?

00:02:59 Speaker 4

Because a lot of people that you asked that haven't gone through this process before think it's too daunting because they don't know how to go about it.

00:03:06 Speaker 4

Can you give a little bit of background into that?

00:03:08 Ivan Diaz

Yeah, I had no idea what I was doing.

00:03:10 Ivan Diaz

I'll be honest.

00:03:11 Ivan Diaz

Again, a little bit of background.

00:03:13 Ivan Diaz

I mean, I use the chat bots for school, get some ideas.

00:03:17 Ivan Diaz

But in terms of building the agent, I had no foundational understanding.

00:03:21 Ivan Diaz

So I mean, what I did is I explained to it what my issue was.

00:03:24 Ivan Diaz

I explained to it what the current process was.

00:03:27 Ivan Diaz

I even fed it some of the documents or the template that we use that helps process the whole transactions and whatnot.

00:03:36 Ivan Diaz

And so from there, I was able to get an idea of what could be done.

00:03:40 Ivan Diaz

And again, like I mentioned, just continued iteration.

00:03:44 Ivan Diaz

I think it's very important to understand the process that you're trying to automate.

00:03:48 Ivan Diaz

And so because I had done the manual process multiple times now, I understood what the final output should have looked like.

00:03:55 Ivan Diaz

And I think that really helped the development of the agent.

00:03:58 Ivan Diaz

But again, I would say to people who are scared, who don't know how to start, just ask the question.

00:04:03 Ivan Diaz

I think AI is smart enough to really understand and just being able to get your feet wet, ask the question and see what you can do.

00:04:10 Ivan Diaz

I think a lot of people would be mind blown at the results that they would get.

00:04:14 Speaker 4

So I've seen a lot of people that try to use a form of Gen.

00:04:18 Speaker 4

AI and the first iteration doesn't come out what they want and they just put their hands up and say it doesn't work.

00:04:24 Speaker 4

What would be your recommendation for that?

00:04:26 Ivan Diaz

Yeah, I don't want to sound like a broken record, but just keep iterating.

00:04:30 Ivan Diaz

I think once you really see a glimpse of what it can produce, I think you are open to a lot more opportunities.

00:04:38 Ivan Diaz

And so there were many times where I just wanted to give up, I would get frustrated because I knew what I wanted it to look like, but it wasn't giving me the output.

00:04:47 Ivan Diaz

And sometimes it's changing your whole framework, sometimes it's using other AI tools, and that's what I dove into a little bit.

00:04:53 Ivan Diaz

I still have the AI tool that I was using, Copilot.

00:04:56 Ivan Diaz

It wasn't doing exactly what I wanted to, and so I went to other AI LLMs, I went to Cloud, I went to Gemini, and I changed my perspective, and I went to them.

00:05:06 Ivan Diaz

I got their feedback, which then I fed into Copilot.

00:05:09 Ivan Diaz

So I would say to use all the tools around you.

00:05:11 Speaker 4

That's great.

00:05:12 Speaker 4

Yeah, I've seen it so many times where one thing's broken, people give up, but if you start working through and saying this specific portion isn't working.

00:05:21 Speaker 4

it can be able to help you troubleshoot through that and fix it.

00:05:24 Speaker 4

So those iterations really help to be able to try to figure out the why.

00:05:28 Speaker 4

And the best thing is talk to it like it's a coworker or another student that you're working with.

00:05:33 Speaker 4

Because if you had a problem in real life, you have to work through those iterations and have that dialogue to be able to resolve it.

00:05:39 Speaker 4

So that really helps.

00:05:40 Speaker 4

And a lot of people don't.

00:05:42 Speaker 4

they treat it more like a calculator than it is a person.

00:05:45 Speaker 4

And treating it like a person really helps to get the best results.

00:05:48 Speaker 4

Some of the audience might be varying who's listening, and some of them might only have the chatbot LLM experience versus any kind of agent experience.

00:05:58 Speaker 4

Can you talk to that a little bit about the difference with that and why you chose an agent to be able to help out this group of internal auditors versus the conventional chatbot production?

00:06:09 Ivan Diaz

So the chatbot, like a lot of us have used

00:06:12 Ivan Diaz

it's more so I ask a question, it gives me a result.

00:06:16 Ivan Diaz

I ask how many states there are, they will turn 50, right?

00:06:19 Ivan Diaz

So it's a simple back and forth, whereas an agent is more of an augmentation aspect.

00:06:24 Ivan Diaz

And so you are working alongside AI.

00:06:26 Ivan Diaz

You are telling it what it should do, it then produces output, you then reinvent its output.

00:06:32 Ivan Diaz

And so it's a little different and it works differently.

00:06:35 Ivan Diaz

Agents are tasked and they're meant for

00:06:39 Ivan Diaz

manual things that can then be automated, right?

00:06:42 Ivan Diaz

So the process I did, me feeding that into a chatbot won't produce an output, right?

00:06:47 Ivan Diaz

It can't.

00:06:47 Ivan Diaz

That's not how it's created.

00:06:49 Ivan Diaz

However, with an agent, because I prompted it in a certain way and gave it instructions, it was able to produce an output that I can then use.

00:06:57 Ivan Diaz

So I would simply explain that the chatbot is more of a Google search almost.

00:07:02 Ivan Diaz

whereas an agent is more of a co-worker, a partner.

00:07:06 Speaker 4

How about areas where you had speed bumps, lessons learned that you could be able to share with the audience so they don't encounter the same kind of problems?

00:07:14 Ivan Diaz

The biggest thing, and this is what I actually shared when I did present this to those 52 auditors, was in Copilot, you can change the model.

00:07:24 Ivan Diaz

And so ordinary models, the default version is to keep it on the auto version.

00:07:30 Ivan Diaz

And the description for that quite literally is, will think how long is I want to, right?

00:07:34 Ivan Diaz

And so for more complex tasks, I would definitely change it to a thinking model, something that can actually comprehend what you are doing.

00:07:43 Ivan Diaz

And once I changed it to a better model, I found that my results were more consistent and were actually better.

00:07:51 Ivan Diaz

And

00:07:52 Ivan Diaz

For those who are starting to build out, again, back to the initial question, just keep iterating with AI.

00:07:59 Ivan Diaz

I'd say when you run into an issue, ask the model why it's running into that issue, ask what the obstacle is, and ask for the way to reiterate and fix the issue.

00:08:10 Ivan Diaz

I think that for me was the biggest help during this process was to simply ask the AI why it was making that issue.

00:08:17 Ivan Diaz

It would tell me, and from there we would find a way to change the instructions.

00:08:21 Speaker 4

As A follow-up to that, any good co-worker still needs to have their work reviewed and everything.

00:08:26 Speaker 4

What recommendations would you have if you're building out an agent, not only on the front-end design, but on an ongoing monitoring perspective to make sure the outcomes that you're getting are what you're intending from a design perspective?

00:08:39 Ivan Diaz

This is something I discuss a lot with my managers and the director of internal audit, because that's, I think that's the most important thing for us who

00:08:49 Ivan Diaz

create recommendations based on the output that we are giving, it's very important that we automate our work and we are able to validate it as well.

00:08:58 Ivan Diaz

And so the thing I did personally was I built a verification tab within the output.

00:09:05 Ivan Diaz

And so I told AI to validate its own work and to make sure that it's including all the values, all the transactions from what I'm pulling.

00:09:13 Ivan Diaz

And so, just be able to, again, going back to your earlier point, communicate with AI that it's important that you want to validate this and create a way that you find more efficient for your company and for the whole process to automate, to create a way to validate the work.

00:09:31 Speaker 4

I've seen a lot from a front-end design perspective, validating, like you said, reconciling back to original reports, having a use case example where you know the items that should pop, if you had five that should pop, did five actually pop, and doing those use case testings, and then on a go-forward basis, the human in the loop, or I know some people have coined it, the auditor in the loop in internal auditing, making sure you have that ongoing monitoring to make sure has something shifted

00:10:00 Speaker 4

it might have worked when you initially designed it, but now it doesn't work anymore.

00:10:05 Speaker 4

Yeah, and it's always.

00:10:06 Ivan Diaz

Good to validate it with the core data, the main data.

00:10:10 Ivan Diaz

And that's what I do.

00:10:11 Ivan Diaz

If I find a transaction that I think should be looked at, I say, all right, let's make sure that this was in the initial output from the accounting system.

00:10:18 Ivan Diaz

So that's another way you can go back and forth from what AI produces.

00:10:23 Ivan Diaz

to the data you actually know is correct.

00:10:25 Speaker 4

Yeah, as you're going through, it's definitely trying to build out the comprehensiveness of what is a red flag, what would constitute something that should be triggering is a big part of the design there because you can say for X, Y, and Z, flag these as regular activity, but what about A, B, and C that might also be flagged?

00:10:47 Speaker 4

So it's super helpful as you're building that out to have a comprehensive list.

00:10:52 Speaker 4

And me personally, I'll try to come up with a list together on my own, but then I'll work with Claude or ChatGPT to be able to say, here's what I have.

00:11:02 Speaker 4

Am I missing anything here?

00:11:04 Speaker 4

Are there risk factors I haven't considered?

00:11:06 Speaker 4

I really stress test it through that and a couple different channels before finalizing that list.

00:11:12 Speaker 4

So that really helps to be able to ensure that you're getting

00:11:15 Speaker 4

a good end product because sometimes it's only as good as the instructions as provided from a perspective.

00:11:21 Ivan Diaz

And I think that goes back to really understanding what the output should be.

00:11:26 Ivan Diaz

I think a lot of people have a misconception of, AI will do it for me, but if I don't know what the output should be, then how can I validate the work?

00:11:33 Ivan Diaz

And so by understanding what the red flags are and what would cause an issue, I think that in and of itself will

00:11:41 Ivan Diaz

increase the effectiveness of the AI agent.

00:11:44 Ivan Diaz

I mean, I use the enterprise version of Copilot, so obviously that is given out by the organization and all the data that I feed it is protected.

00:11:54 Ivan Diaz

To those who have their own and that isn't protected with the enterprise, obviously be careful.

00:12:00 Ivan Diaz

Obviously, I think the most important is talk to those who may be more experienced, whether it be a manager, whether it be IT.

00:12:06 Ivan Diaz

Understand the environment, understand that there are

00:12:09 Ivan Diaz

ways that giving AI can leak confidential information.

00:12:13 Ivan Diaz

And obviously there are security risks.

00:12:15 Ivan Diaz

But personally, I did use my local to help me with this agent I was building.

00:12:19 Ivan Diaz

I never fed it the actual data.

00:12:21 Ivan Diaz

But what I did feed it was my questions.

00:12:23 Ivan Diaz

What I did feed it was the instructions I gave it.

00:12:26 Ivan Diaz

Because the instructions don't lay out any confidential data.

00:12:28 Ivan Diaz

It's just telling it what the agent should do.

00:12:31 Ivan Diaz

And so I think it was a good way for me to, again, just get a better understanding and more ideas

00:12:37 Ivan Diaz

how I can better my instructions.

00:12:38 Ivan Diaz

So yeah, I would just say communicate with those around you.

00:12:42 Ivan Diaz

Understand what's confidential, what's not.

00:12:45 Ivan Diaz

And for the tools you do have available, definitely use them.

00:12:49 Ivan Diaz

But definitely.

00:12:50 Ivan Diaz

definitely be careful with what you're feeding it and obviously never want to feed AI stuff that is confidential to the company.

00:12:57 Speaker 4

So one of the recommendations, always look at acceptable use policies within your organization and talk to your information security team before you put anything in there, even if you think it's okay.

00:13:07 Speaker 4

Just always check to make sure that lines up with the organization's risk appetite.

00:13:13 Speaker 4

So any lessons learned that you guys had as far as from the start of this project to the end, where did you grow and learn and understand better from the needs of internal auditors as well as the technical onuses of creating an agent?

00:13:30 Ivan Diaz

Getting various perspectives, I think I learned a lot from.

00:13:33 Ivan Diaz

When I initiated the project, I only saw it from my eyes and from

00:13:39 Ivan Diaz

the person doing the manual work.

00:13:41 Ivan Diaz

But when I got the perspective from my manager, for example, I was able to alter the agent in ways that he saw to be more effective.

00:13:50 Ivan Diaz

And so that was implementing A validation sheet inside the Excel file that it was given.

00:13:56 Ivan Diaz

And so I think the best thing I learned from start to finish was just being able to change perspective and being able to kind of see the vulnerabilities.

00:14:06 Ivan Diaz

Because I think when you talk to someone who's more experienced,

00:14:09 Ivan Diaz

they can give you a situation that you may have not thought of.

00:14:12 Ivan Diaz

And so I would say from this agent to the next agent, I definitely saw in that aspect a little bit.

00:14:18 Ivan Diaz

I realized that there was a lot of interest in the topic.

00:14:23 Ivan Diaz

I found that a lot of people were trying to get into it and were asking questions similar to what we're answering now.

00:14:30 Ivan Diaz

And I enjoyed that.

00:14:32 Ivan Diaz

I really do think that a lot of people are starting to realize that there is a need to adapt to this new technology.

00:14:38 Ivan Diaz

and I believe people are finding ways in their own work to automate.

00:14:42 Ivan Diaz

And so although I haven't been able to communicate with any of the auditors since the presentation, I really felt that throughout the presentation they were engaged in.

00:14:51 Ivan Diaz

They're really eager to be able to apply what they were learning into their own situation.

00:14:55 The IIA

While Ivan was solving work paper headaches as an audit intern, recent cybersecurity graduate Mohamed Maiga took the stage from a totally different angle.

00:15:04 The IIA

Here's Mohamed.

00:15:05 Speaker 4

So Mohamed, appreciate the opportunity to be able to connect.

00:15:08 Speaker 4

Ivan walked us through some components of what his element of the project was, but could you be able to outline what you did on your side for the project?

00:15:17 Mohamed Maiga

I kind of stepped in that project a bit.

00:15:21 Mohamed Maiga

Abruptly, it was very much out of my own curiosity.

00:15:25 Mohamed Maiga

I met the senior director of the College Applied AI Institute at one event where students were essentially presenting about how AI can be applied in the classroom, what bothers them about AI.

00:15:38 Mohamed Maiga

And that's how I got to meet the senior director, Tyler Small.

00:15:42 Mohamed Maiga

And at the end of that meeting, he asked me a question, can you build an agent?

00:15:46 Mohamed Maiga

And

00:15:47 Mohamed Maiga

At the time, I had no idea what an agent was.

00:15:50 Mohamed Maiga

I've built it in the past, not knowing that I built an agent before.

00:15:54 Mohamed Maiga

So with my very much excited, willing to learn college student heart, I just screamed, yes, yeah, I can do that, not knowing.

00:16:06 Mohamed Maiga

And I had to go on YouTube for about 10 hours to learn about AI agents and how they work.

00:16:12 Mohamed Maiga

So for the specific example of the 52 internal auditors from the UC, Tyler Small, the director of the AI Institute, reached out to me and he said, hey, Mohammed, do you want to present at this event?

00:16:25 Mohamed Maiga

I was like, yes.

00:16:27 Mohamed Maiga

Again, with my college student curiosity, he gave me this project that he had presented in the past.

00:16:34 Mohamed Maiga

So I was just basically using his work to teach other people.

00:16:39 Mohamed Maiga

Here's how it works.

00:16:41 Mohamed Maiga

People had problems at UVU with pro cards, the procurement cards that people use to make purchases on campus.

00:16:48 Mohamed Maiga

And the reason is there's a lot of rules.

00:16:51 Mohamed Maiga

Like the document probably has 99 pages and people are doing hundreds of purchases every day, right?

00:16:58 Mohamed Maiga

So for every single one of those purchases, you have to ask your question, is this right?

00:17:04 Mohamed Maiga

Does this align with our policies?

00:17:06 Mohamed Maiga

Am I gonna get in trouble?

00:17:07 Mohamed Maiga

Am I getting get deducted points for this purchase?

00:17:10 Mohamed Maiga

So what Tyler Small did to tackle that problem, he created an agent that essentially has all the knowledge of the different policies that Utah Valley University has to be able to explain people whether or not their purchases are legit in just one simple chat instead of reading through 100 pages, having to do control F and looking keywords, maybe they're not even there.

00:17:34 Mohamed Maiga

So that's the project that I presented during that presentation.

00:17:37 Mohamed Maiga

And

00:17:39 Mohamed Maiga

essentially how to build an agent from scratch, meaning from the first step to the last step.

00:17:46 Mohamed Maiga

And I focused on telling them about the importance of AI, what it can do for us and what it cannot do and definitely should not do, which is the reviewing part, which is what we need, where we need the expertise of those internal auditors, right?

00:18:04 Mohamed Maiga

You have to be able to tell the difference between

00:18:06 Mohamed Maiga

AI slop and where it's doing well in your industry.

00:18:10 Mohamed Maiga

But I quoted President Tumines, president at the time of Utah Valley University, who usually says, we are human beings, not human doings.

00:18:19 Mohamed Maiga

And I think with AI, we're getting the ability to spend a lot more time on being instead of doing.

00:18:27 Mohamed Maiga

Because to be honest, in every single job that is on a computer, we've created a lot of

00:18:34 Mohamed Maiga

meaningless meaningful task, meaning the output is very important.

00:18:40 Speaker 4

Huge thing about being an internal auditor is having that level of professional curiosity to be able to be open to learn because there's no expectation that someone's gonna start as an internal auditor and know every aspect of the business, doing the repetitive tasks that you've already built a test of design, review over, and then you gotta do 24 more samples of it.

00:19:01 Speaker 4

have an agent be able to do that while you're focusing on higher risk items that add more value to the organization.

00:19:08 Mohamed Maiga

Perfectly summed.

00:19:09 Speaker 4

During your presentation to these internal auditors, did you go through any kind of specific examples, a case study?

00:19:15 Speaker 4

Can you expand on that a little bit?

00:19:17 Mohamed Maiga

I also had some personal examples in my own learning process.

00:19:23 Mohamed Maiga

I just recently took the Security Plus exam, which is a very important professional certification in the cybersecurity space.

00:19:30 Mohamed Maiga

I had a lot of the knowledge about it already through my courses and stuff, but I had a hard time coming to pick an exam date with graduation happening and so many things just crumbling together at the same time.

00:19:44 Mohamed Maiga

So I was like, how can I make my learning process faster with AI, right?

00:19:51 Mohamed Maiga

So I had a lot of course materials from previous classes that I took.

00:19:55 Mohamed Maiga

There was YouTube content from Professor Messer.

00:19:59 Mohamed Maiga

He is a quite renowned professor in the cybersecurity space.

00:20:03 Mohamed Maiga

I had all those information combined in one simple agent that would create me a website where I would be able to learn all the content without going outside of that context, right?

00:20:15 Mohamed Maiga

So what agents allow us to do is

00:20:18 Mohamed Maiga

take the knowledge that you already have, put it inside of a knowledge base that is just focused on your own, and then use that to learn or to teach or to perform tasks. In this case, it helped me to learn. And just within five hours of studying that course, I was able to pass the exam. So that is just one of the many use cases where I use AI on a daily basis to learn and

00:20:43 Mohamed Maiga

to perform tasks.

00:20:44 Speaker 4

When you're working on this project, what were the different validation procedures, what you guys did for this project?

00:20:50 Mohamed Maiga

So we told the AI in our prompt to give us where he got that information from. So it wasn't like just rambling up, okay, yeah, you can do that purchase for sure.

00:21:01 Mohamed Maiga

Yes, it can do that purchase on page 97, article, dot, dot, dot. Okay, that's where I found the information. You click right there, you see the information. Okay, it's validated. It's in the document. It didn't make that up. So hallucination is a big concern. So I think that's where the expertise is needed as well. Sometimes AI can make up information, but if you have someone to review the output and validate it, that is where

00:21:30 Mohamed Maiga

is going to give you the right outcome, right?

00:21:32 Speaker 4

So if you give a general loose prompt, an open-ended prompt, the risk of hallucination is much higher. If you upload specific policies and procedures and criteria and give very specific prompts to it, the hallucination risk gets mitigated significantly. So really, really important if you're working off of criteria as an auditor, that you have in hand policies, procedures, a risk and controls matrix

00:22:00 Speaker 4

Make sure you're uploading that as long as it's a secure way. You mentioned having enterprise license of copilot or something like that. If you're confident that the information that you're uploading is being secured through that measure, upload that information because it gets a much, much better result from there. It's one of those you get as much as you put in type things. AI is definitely that case there.

00:22:24 Mohamed Maiga

Exactly.

00:22:24 Mohamed Maiga

Your input is your biggest output. Yeah.

00:22:27 Speaker 4

So if you were a director of internal audit that's listening to this and you haven't rolled out AI agents within your internal audit function, what would be the biggest takeaway coming out of this for you?

00:22:40 Mohamed Maiga

If you're a director of an internal auditing team or any kind of leader really, have a meeting with your people and ask them, where are you spending most of your time? If those

00:22:52 Mohamed Maiga

times are being spent on repetitive tasks that can be automated by AI. Now, the problem is people want to jump too much to conclusions. People want to implement AI too fast. I think you can start with something narrow, something very easy, something as simple as replying to emails in my tone, right? You train the AI with your own writing style, with your own voice, and you fit it like

00:23:22 Mohamed Maiga

emails that would take you tons of time to reply to. I think a lot of people spend a lot of time on those tasks as we go because there is so much communication in companies now. And something as simple as that, you see the result and you're like, okay, this can do that, then where else can we apply? And that's where it expands. You don't have to do something big from the beginning. And 2, make sure that you are not inputting any kind of personal

00:23:51 Mohamed Maiga

identifiable information, any kind of social security, whether the AI is being trained by the data or not, that is just not something advised from a security perspective. So start small, incremental steps towards a bigger goal and talk to your employees. Maybe they're already using AI in a shape that is not being controlled. So you want to avoid that.

00:24:14 Mohamed Maiga

Right. So if you as a leader don't take the decision that's going to enable your team to be more efficient, then you're doing everyone a big disservice.

00:24:25 Speaker 4

You see most of failed AI implementations at a company is because they try to go from zero to 100 immediately. They try to do a full scale rollout. It's really important, like you said, to be able to start with those small wins

00:24:41 Speaker 4

do a pilot program, small incremental buildup so people build the awareness and understanding of how to use AI first and then be able to do practical applications of the areas that take up a lot of their time. You don't want to do AI just for the sake of doing AI. You want to be able to take off the time, free up your space and use it to become a force multiplier.

00:25:03 Mohamed Maiga

Yeah, I'd even go as far as saying like, don't apply AI everywhere. You don't need it everywhere.

00:25:08 Speaker 4

Well, thank you so much, Mohamed. Really appreciate

00:25:10 Speaker 4

It's the time to be able to talk and very exciting work that you're doing.

00:25:14 Mohamed Maiga

Hey Alex, it's been a pleasure. Thank you for having me.

00:25:18 The IIA

If you like this podcast, please subscribe and rate us. You can subscribe wherever you get your podcasts. You can also catch other episodes on YouTube or at the IIA.org. That's T-H-E-I-A.org.