LPRC - CrimeScience – The Weekly Review – Episode 241 Ft. Elissa DiPierro

Episode Date: July 2, 2026

In this episode of the CrimeScience Podcast, host Tiffany Frison sits down with Elissa DiPierro to explore best practices for AI prompting and practical ways to leverage artificial intelligence in everyday work. Elissa shares strategies for crafting more effective prompts, improving AI-generated results, and identifying high-impact use cases that can enhance productivity, decision-making, and innovation. Whether you’re new to AI or looking to refine your approach, this episode offers actionable insights to help you get more value from today’s AI tools.

Hi everyone and welcome to crime science. In this podcast, we explore the science of crime and the practical application of this science for loss prevention and asset protection practitioners, as well as other professionals. Hello everyone and welcome to this week's LPRC crime science podcast. I am your host, Tiffany Friesen, research administrator here at the LPRC. And this week we have another edition of the Team Spotlight series. And our special guest this week is one of our interns, actually, Alyssa DiPiero. And we'll get into it, but we'll be talking about the effective use of AI and prompt engineering today.

But first, Alyssa, do you want to tell us a little bit about yourself, kind of maybe how long you've been here at LPRC and some of the different things that you've been working on since you've been. been here. Sure. So as Tiffany explained, my name is Alyssa. I am one of the interns here at the LPRC. I've been here for a little over a year now. And in that time, I've worked on a variety of projects. I kind of started off more on the open source intelligence side, really working on our digital Overwatch initiatives. In recent times, I've kind of migrated a little bit more towards workflow automation and coding product, project, sorry. And what that entails is, really just finding ways to create products, software projects, that will make everyone's jobs

a little bit easier, a little bit more efficient, which is kind of what I'm here to talk to you all about today. Awesome. Thank you, Alyssa. Yeah, I honestly, I feel like it's been more than a year that you've been. You feel like such a core part of the LPRC team. But, yeah, no, we've loved having you here and we thank you for all the work that you've done in the year that you've been here. Okay, but, all right, we're talking about AI and specifically effective use of AI. So, first of all, can you give us like a brief overview of some terms that may seem obvious? But just in case anyone is not familiar, just a brief overview of some of the terms I will be talking about today. Of course, of course. So a lot of these terms you might have heard before, but maybe they sound a little bit similar. You don't really know the difference.

So to break it down, to start off, we have artificial intelligence, AI. This is going to be a broad category of systems that perform tasks that typically would require human intelligence, but now obviously we're using computers to do that. Some examples would be, reasoning, pattern recognition, language generation, prediction. Building off of that, we have large language models. These are LLMs. This is a type of AI, and this is trained on a massive amount of text. So these help power tools like chat GPT, Claw, Gemini, copilot. It predicts and generates language based on patterns that it observes within the training data. Then what's going to be really important for our discussion here today is prompting.

engineering. This is the process of designing instructions that help AI produce better outputs. So it's basically like learning how to communicate effectively with an AI system. It's not necessarily program, but it is another type of skill. And the last thing we should probably establish first is something called hallucination. And this is when AI confidently provides incorrect or fabricated information, but it's saying, I think this is true and it really creates kind of an issue. This is why it's an important reason to verify any outputs, especially in something really important, like research or investigations or any form of decision making. Awesome. Great. Yeah. Those are very good for us to keep in mind today.

All right. So today we're talking all about prompt engineering. Can you briefly explain the difference between a good prompt and a bad prompt and what that might look like. Sure. So mainly the main difference is in how specific you are. A bad prompt is going to be vague. It might be missing context. It's missing the goal that you're trying to get out of the prompt and any constraints. So for example, you might just say, write me an email.

Well, what type of email? What do you want in this email? Who is this email going to? The AI has to guess who it's for, who you're sending it to, the tone. and that really gives you a very generic output that you don't necessarily want, and you're probably going to have to spend a lot of time refining. A good prompt, meanwhile, provides context, it might define the audience, it includes any requirements or constraints,

and specifies the desired output format. So, for instance, using our email example, this might be write a professional email to retail operations managers, requesting participation in a research study, keep it under 150 words and with a clear call to action. The difference isn't usually in the AI's capability in and of itself. It's more the quality of instructions that you're providing. And this is a key point that I want to emphasize.

The AI is a very powerful tool, but it's only as powerful as you're able to instruct it to be. Yeah. So when creating a prompt, I know there's like 100,000. of thousands of ways that this could be applied. But thinking of our members here, the LPRC, what are some applications that you could see people using prompt engineering in, this is very specific as well, but in like retail, in asset protection or loss prevention, are there anything that like come to mind in that area? Yeah, so there's a ton of things that it could be applied to in these specific

industries. Specifically, you might have summaries, whether this is incident reports, ORC intelligence summaries. You could also have drafting for things like case narratives. You could do a lot of analysis of large scales of data. Typically, these might require more of your paid plans, but it can process large quantities and scales of data and take out those key points. So maybe any trend analyses, it could also review a large volume of things like CCTV, observations and make notes on that, all of these things that might be a little bit tedious for a human to do, but the AI is able to do it a lot faster. Awesome.

Yeah, and I know, like, there's also, like, countless applications and just, like, general workflows as well, and just streamlining, like, tedious or, like, time-consuming tasks as well. Okay. Yeah. So now that we've talked a little bit about like different use cases, I want to get into the nitty gritty. So what are the elements of a successful prompt? So again, a successful prompt should be as specific as possible. So what that's going to entail is things like requirements and constraints. Requirements are things you want. Constraints are things you don't want. You want to provide context. This might include the audience and background. So, for instance, who you're sending it to, why you're sending an email, or just in general, what that summary is going to be used for. It really helps shape the overall tone that you want. Maybe you want it in a specific output. Maybe you want it directly in the LLM's output format. Or maybe you wanted an award document. You want to specify that. You might provide examples of previous successful outputs so that it has something to model it based off of.

And at the same time, you might provide an example and say, this is what I don't want from you. You could also provide it a role, ask it to take on a certain job assignment. So for instance, if you're creating a summary, you might have it take on the role of researcher or an analyst, somebody who would have this kind of background knowledge and industry knowledge that you want to see represented in whatever output you're doing. And as you keep prompting it, as you keep going through it, you obviously want to provide any feedback. The goal of prompt engineering is to kind of reduce the need to keep doing this back

and forth to get the desired output. You want to be able to be specific the first time to get that desired output. But again, you probably will still have some need for feedback. So to kind of summarize it, the more guidance you provide, the less guessing the AI has to. do. Gotcha. Yeah, because I think I've heard before and I could be making this up, but like AI doesn't know what it doesn't know. Exactly. Yeah. No, that was perfect. As I was saying, I was like, wait, does this make sense? But yes. Awesome. So just to do like a really brief review,

So kind of those elements in a list format, so it's easy to digest. Go ahead, what would those be? So that's going to be your requirements and constraints, your context, including background and audience, the output formats, examples of successful outputs, and maybe some bad examples as well, roles and job assignments, and then any feedback and iteration you might need to do. Awesome. Thank you so much. So obviously you've created a lot of prompts here at the Lercy and I'm sure in countless other areas of life. But is there anything else that you think is important that we should touch on today about creating effective prompts?

Yeah. So prompts do not need to be perfect. That's something that a lot of people get wrong. Whereas in the past when you were doing any search engine searches, you had to make it a little bit concise. And AI isn't like that. It could take in massive amounts of data. So you could honestly just ramble on and on and on. And it will still most of the time be able to make sense. So you don't have to worry about making the most perfect prompt. Like I've been saying, I've been saying a lot of things, you don't need to include everything if you really can't. Just try to be as specific as possible. and even if that requires paragraphs of explanation, the AI can still summarize those main points.

That is a main capability and feature of AI. It understands natural language surprisingly well. Another thing is that you don't have to rely on the first answer. You can really keep prompting it over and over again to get what you want, and each time it's going to learn a little bit from you, you're going to learn a little bit from it, and hopefully you'll get a great answer out in the output. And lastly, we kind of touched on this in the beginning, especially in very sensitive things.

AI isn't perfect. It's not going to give you a perfect output. It might hallucinate a little bit. So always, always, always double-check the data. You can pass it through, honestly, another AI. But for super sensitive stuff, always have a human reviewing it. So in the end, AI isn't going to completely replace the work for humans. We're always still going to need that human element to check it.

Gotcha. Yeah, so you brought up a point about, like, okay, so if you write a prompt and the output isn't exactly what you wanted, like, obviously, then you can go back in and then change your prompt and, like, modifying things. And depending on what platform or whatever you're using, then that AI agent can hopefully kind of re-evaluate and give you the output that you are looking for. And I know this is very, like, basic. Probably anyone that uses AI knows these things.

But I think it is important to point out, like you were saying, like, you don't have to be perfect in your prompt. And, like, the agent will pick up what you're trying to put down, essentially. And if not, you can go back and try again until you get that output that you're looking for. Yeah, so that really is a key feature that separates AI from, say, a search engine because you are able to go back and really refine it and create something very personalized that will give you exactly what you want. You should really be treating AI almost kind of like a junior employee that you want to provide direction, but also review their work, provide feedback. and then you could keep going through these rounds of iteration, again, trying to be as specific as possible each time

to really craft that desired output that you want. Awesome. I like that idea of thinking, like, this is something that I do have to check, but in the idea that it is giving the best of its ability. Yeah. Awesome. Okay, so we're rounding things up, but I do want to ask one last question. So, Alyssa, personally, which AI tools do you find that you use commonly? And then what would you say would be like an easier one to use or like is more user-friendly or something like that?

Yeah, so I honestly think the best way to choose which tool to use is not to focus on one, but to use a common. of multiple ones, just because they each have their own strengths and weaknesses. Personally, I use chat GPT mostly for more general stuff. I use Claude when it comes to coding. Microsoft copilot I use, just because, you know, using the Microsoft suite, I use that pretty frequently, but I also do tend to prefer it for things like language generation, like drafting emails. And then another one that's a little more less known, but still pretty well known, is

perplexity. It's great when you have to do any research because it cites Google. So you're able to find a lot of sources that are cited. Awesome. Wonderful. Well, this has been enlightening and hopefully very useful for everyone listening. So thank you so much, Alyssa, for being on today. And we'll see.

Maybe we'll have you back another time on the podcast. But thank you so much for being here. today. Of course. Thank you for having me. I'd love to come back. I'll make a note real quick. Awesome. Well, thank you, Alyssa, again, and thank you to our listeners. And as always, you guys can find all of our resources. There's a lot on AI and LLMs on our Knowledge Center. If you need any assistance getting plugged in there, feel free to reach out to myself or any LPRC team member. And then I do want to also mention on the data analytics working group, one of the focuses

that they are kind of churning to these weeks, these months, is AI. So if you're interested in more discussion and kind of getting plugged in into that area, feel free to reach out and sign up for the data analytics working group. I think that's all for today. Thank you guys for listening and we will see you on the next crime science podcast. Have a good one. Thanks for listening to the Crime Science podcast presented by the Loss Prevention Research Council. If you enjoyed today's episode, you can find more crime science episodes and valuable information at LPRsearch. The content provided in the crime science podcast is for information. informational purposes only and is not a substitute for legal, financial, or other advice.

Views expressed by guests of the Crime Science podcast are those of the authors and do not reflect the opinions or positions of the Lost Prevention Research Council.