Using AI to Improve the Archiving Process – Digital CxO Podcast EP97

Hello, and welcome to the digital CXO podcast. I'm Amanda Ani, and with me today is Laura Stash. She is the executive VP of Solutions Architecture at iTech ag.

How are you doing today? I'm doing good. It's voting day and I voted.

Thanks for having me. Yes. Yeah, I voted early.

I wasn't sure how long the lines would be today. How long were the lines today? Oh, well, I took my 4-year-old and my two month old with me, so I got to cut the line.

Oh, nice. Wonderful. Well, to start off, can you share a little bit about iTech AG and what services are provided?

Yeah, absolutely. iTech AG is an IT consulting firm. Uh, we provide technology transformation mostly in low-code and no-code digital platforms, uh, including, you know, AI features, data management features, cybersecurity features, et cetera.

Wonderful. All right. Well, um, recently the US National Archives announced their new strategic framework, which will serve as a template to guide the development of a full strategic, and, um, it charts a course for the agency that emphasizes building digital capacity, scalability, and responsibility, embracing technological innovation.

Yeah. So we're gonna talk about that today, um, and, uh, talk about AI and what an impact AI has made and machine learning has made. So, um, what are you seeing out there?

What are you hearing from business leaders? Yeah, absolutely. So we've actually had a couple of AI contracts exactly doing what the archives does, which is, uh, document digitization, and we are supporting multiple agencies using machine learning tools, um, and platforms to digitize those previously paper files, um, and make those things that are digital files that, that became digital files or were, um, originally digital files easier to navigate, find, uh, and explore for different purposes across the government.

Wonderful. Uh, what are, can you share, um, a little bit about, uh, what are some of the things included in this new strategic framework that business leaders should be aware of? Mm, yeah, of course.

Uh, so I think the main thing is about, um, how much we're gonna trust that AI is the big, like, hot topic for everybody. Um, so the thing about document digitization and creating scalability and capacity is it costs a lot of money to digitize records, especially if you're digitizing manual records. Um, it costs a lot of money to index those digital records once you have them.

Um, some agencies have gotten quotes where it's like 15 cents per page, which is nuts, right? That's millions and billions of dollars. If you think about all the records that we have within the United States or within a given agency, even.

So, the benefit of machine learning and these large language models is that, um, those records can be digitized and index and searchable at a much lower cost, but the only way that we're gonna actually achieve that cost savings and be able to digitize those is if we, um, train those models. So there's a thing called human loop, um, and what's what's being talked about a lot is how do we train those models in a way that's responsible, um, and to, for government, um, leaders and government employees to feel confident and secure once they train those models, that they don't have to continue to touch every single document or every single metadata that they've trained it, it's accurate, we trust it, and now it can go and do its thing, um, at basically a free cost, right? Just the set up costs.

Um, so that's kind of, I think the hot topic is how much are we gonna trust that ai, um, how much do we trust that once we've trained it and then it's accurate, it's gonna stay that way. Um, and there won't be hallucinations, right? That's a, that's a key topic in ai.

Um, yeah, so I think those, those would be probably the two hot topics I would speak about first. Yeah, that's certainly, definitely, um, uh, a key issue of concern. We have seen hallucinations, um, how much can we trust the data?

And I don't feel that anyone really believes AI can just be left to run by itself at this time. It's, it's still just a good tool amongst other technologies. That's right.

And so it's like, what's that percentage of human in the loop that's needed at the start? You train it, um, and you, you feel confident in the, in the training and then, you know, is it 1%, is it 10% of the records that you wanna touch, um, to make sure that it's not hallucinating or to make sure that you're catching those outliers. Make sure that you're continuing to train that large language, large language model or train, you know, other elements of it.

Maybe there's new types of records that come in and needs to be trained, right? So what's that level of touch and, and what's the cost associated with it? Right.

That's, it's kind of the key topic going on in, in that industry today. Yeah, absolutely. And there, and this technology is advancing quite rapidly, so I'm sure we're gonna see, um, eventually down the road.

Do you think that it will be able to just be let loose and, and run on its own, or do you think we're pretty far from that? Um, I think it's, it can be let loose and run on its own if for standard forms, if that makes sense. So a lot of what happens in government is there's all these like standardized forms that don't change over time.

Like they have, some agencies have forms that have been the same since like the nineties or even before that, right? So, um, my opinion is that the large language models can be trained, um, on those types of things, and then we can kind of leave them alone for the unstructured data. I think it's more dangerous, and that's where you see a lot of the hallucinations happening, um, that requires a little bit more of a hands-on.

So, um, it, it just depends on the type of data and how much good, how good the training data is, and how much of that training data is available, whether you can actually leave it alone or not. Yeah. And the other question is, we get more and more data by the minute.

Um, so how do we keep up with all the new data and all the new records? Yeah, absolutely. And what's really interesting is, um, the NLP and the, and the AI personalization, right?

So, um, not just the data that the government has that people are trying to access, but the data about how people are trying to access it and what they're trying to access, and then using that data to make the response better in the future. So for sure, for that reason, you'd never leave the model alone because you always wanna see what are people accessing, how are they trying to access it? How can we make that customer service element or that personalization better?

Which is exactly what, um, iTech AG is trying to do with a lot of our AI products today is like, um, use AI for better customer service and personalization. So let's talk about that. I know that that's a key issue for business leaders, how to harness this data for that better customer experience.

Uh, what tips do you have for business leaders? Yeah, absolutely. So what we've seen in the market first is really just setting up, um, like a virtual agent chat bot and agent assist features so that when a customer is reaching out to your company or, or the government, whatever your model may be, um, that it can quickly summarize what they're asking for.

It can pull together all the history of any interaction with that customer, um, altogether really quickly. So that either the person or the system that's responding to that, um, customer has all that information and they're not starting from square one. What you see even now, um, maybe when you're not Amazon, but when you're in interacting with other tools in the industry that have, um, chatbots, uh, like my bank for instance, right?

I tried to ask it to Zell someone the other day. Um, it it, every time I go to try to zell someone, it doesn't remember that I zell this person last week or whatever, right? And it's treating me like I'm net new and that I, I have to go through all these 17 steps every single time, right?

So improved customer service would be like, great to see you again, Laura, do you wanna Zelle this person? You Zed last week, click this one button and it's done. Right?

So I think that's the personalization element that, um, we can start to leverage and harness and once we do, people will adopt AI better and faster. Absolutely. So from your experience, um, I know many people are still hesitant to embrace AI technology, uh, fear of the unknown.

So, uh, when it comes to implementing new technology of any sort, including ai, what advice do you have for getting everybody on board and on the same page? Yep. I think, and I've said this in a, a couple of interviews and, and different articles that I've been quoted in, is I think it's making it so that it's easier for them to use AI than to not use ai, making it seamless.

So a lot of times you'll see AI in the platforms we're already using, whether that's a Microsoft platform or maybe even in Zoom, right? Or whatever thing you're already currently doing. If you just add an AI element into it to say, Hey, you can do this faster or better using ai, click this button.

I think people will start to use it and they'll be less afraid and less scared of it, right? Versus having to go to a separate tool or a separate product to, um, do whatever AI steps that they're trying to do. Mm-Hmm.

Well, if there was one key takeaway you could leave our audience with today, what's the big takeaway you'd wanna leave them with? The big takeaway? Okay.

Well, the big takeaway I would say is, um, AI is happening either way, so we might as well embrace it, and if we do embrace it, then AI will be more useful or beneficial to whatever business model you're applying it to. Absolutely. It'll be interesting to see how it unfolds in the future and just what use cases it can be applied to.

Yeah. Thanks so much for having me today. Thank you.

And thanks to our audience. Stay tuned. There's more.