AI in Logistics: Beyond the Buzzword With Ibrahim Ashqar From Lumi AI - Ep. 30

Lori Boyer 00:00 

Welcome back to Unboxing Logistics. I'm your host, Lori Boyer, here with you as always to talk about the latest issues in the logistics industry. And I am really excited today for our guest, Ibrahim Ashqar from Lumi AI is here to talk to us all about the big buzzword in the industry that you can't get away from no matter where you are. Every conference, every webinar I'm on you, we really can't escape the topic of AI.

Fortunately, I have brought in an expert and specifically we're titling this episode Beyond the Buzzword. I want to hear from Ibrahim what in the world AI really means to us. You know, beyond all the cool stuff, how can you actually use it? How is it going to be important? So Ibrahim, why don't you go ahead and introduce yourself to our audience out there?

Ibrahim Ashqar 00:56 

Yeah, it's awesome. Lori, thanks for having me and excited to be here. AI is definitely a buzzword and hopefully we'll demystify some of the myths from what's actually happening in industry with AI. And you know, dive deeper there and talk about how it's actually being used in industry. Give a quick intro about myself and we can kind of take it from there.

So yeah, Ibrahim Ashqar, CEO and co-founder of Lumi AI. We're basically trying to redefine how companies with supply chains interact with their operational data and make that interaction as easy as having a conversation. Prior to founding Lumi, I was the director of data science at Stord. People online may have heard of them supply chain, 3PL, warehouse provider, and fulfillment.

And before that I spent about four years within Deloitte's supply chain AI practice, building enterprise-grade data products for a variety of large clients. And really it was all about finding ways to leverage data analytics and machine learning to visualize, optimize, automate supply chain functions across a few industries, CPG, retail, 3PL.

So think image optimization through better demand forecasting, replenishment you know, recommendation, purchase recommendation engines, and kind of dashboards for executives and middle management. So, in a nutshell, data and supply chain, that's what I'm all about. 

Lori Boyer 02:16 

We were introduced from our co-friend Nicholas Daniel Richards. Shout out to you, Nicholas. But he introduced you as the smartest guy he knows. So I love it. That's exactly what we want here. Smart people who can take really complex stuff. Like when you said that you want it, it should be easy. Easy and logistics hardly seem to go in the same sentence. It is so complex and so hard.

So we are relying on you here to help us make AI easy to understand. The smartest people out there are those who can make it easy for us. So thank you, Nicholas, for introducing us. Ibrahim, we are excited to hear all from you today. 

Ibrahim Ashqar 03:02

That's awesome. Yeah, Nicholas is awesome. I think he's overselling my capabilities, but I'll do my best here.

Lori Boyer 03:08 

Oh good, perfect. I'm sure you're so humble. Okay, so we're gonna start out by me asking you just a couple of little background questions to get to know you a little bit better. And then I'm gonna ask for some hot takes from you. So my first question I have for you is do you have, and what is, your go-to comfort food?

What food do you want that just brings you back home, makes you feel comfortable when you're feeling like you need a little something. 

Ibrahim Ashqar 03:38

You know, this is gonna be odd because I feel like you know people expect an answer like burger or something of the sort. It's really funny. I actually have a very basic salad that I love to eat of just cucumber, tomatoes, peppers. And I put salt and olive oil with it and salt and some mac powder.

 And it reminds me of home and I, I, I eat it all the time. I think I might be one of the biggest consumers of vegetables cucumbers in general. That's my comfort food for sure. 

Lori Boyer 04:11 

That is incredible. That is incredible. I need to put away the ice cream and the pizza and, and learn a little bit from you. You mentioned it reminds you of home. So tell, tell our audience where, where is home, where are you from? 

Ibrahim Ashqar 04:21 

Yeah you know, I was born in Jordan, you know, it's, I still go there a lot and have family there, friends there and everything, but grew up in Dubai went to school there, and then for university moved to Toronto, to University of Toronto, did engineering there.

Started off with, you know, I wanted to do robotics over there. I was really inspired by Iron Man. Then I realized hardware is really tough. So I just stuck to the software side of things. I'll just build artificial robots and things of the sort. Yeah. 

Lori Boyer 04:54 

You wanted to go the easy route. So you went for some artificial robots.

Ibrahim Ashqar 04:58 

Software is much easier than hardware. I'll tell you that. People who love some dynamics are, you know, just some next-level stuff building a robot. 

Lori Boyer 05:07

Okay, I love it. I'm gonna have to tell my husband. My husband is a software engineer, and my brother-in-law is a hardware engineer. And so, I don't even think it's called hardware, but, so, I'm gonna tell him.

He's lost out to my brother-in-law now. Okay, question number two. If you had no time or money constraints, nothing keeping you back time or money wise. What hobby would you like to start or what'd you like to be involved in? Dream hobby. 

Ibrahim Ashqar 05:33 

I, I, so I think it's an easy one. I always, I love, you know, I'm an adrenaline junkie in general. I love sports and things like that, but I really love diving. I'm so curious by just marine biology. And I actually, at some point thought I, in between third year and fourth year of university, they have a thing called professional experience here where you take a year off and you go work somewhere.

I kept trying to apply to Discovery Channel because I want it to be in, you know, one of these like expeditions exploring the sea and all the creatures and things like that. I love diving. I have a license, the PADI license, and you know, dive, dove in, in some, some really cool spots. And honestly, I would, I would love to do that. 

Just spend more time diving and enjoying and learning more about the ocean. That's just kind of what I would do. 

Lori Boyer 06:22 

Oh, that's amazing. It is like our big frontier that we don't know very much about still. So love that. Love that. Okay. So in every episode it's really important to me that, you know, if people only have a few minutes if they haven't been able to listen to the entire thing that they come away having learned something. So I ask right up front.

What are your you know, couple of takeaways that you've got that you want people to come away remembering. That's something they could implement, actionable, just something in their life that they can take away and remember from today's conversation? What do you got for us? 

Ibrahim Ashqar 06:58 

On the topic of AI, so I'll keep the, the key takeaways, you know, more in line with that, but really we're, we're entering a new era with AI and large language models.

And I think capitalizing on their ability to do anything is really a good application. The benefits are truly incredible. Don't be scared to embrace the change.  Those who adopt this technology will be able to really elevate and kind of leap forward.  Compared to others who don't want to adopt it. So definitely embrace the change.

We're entering a new world. It's almost as revolutionary as, in my opinion, as you know, the, the, when the internet came out. And then for, for, for kind of those in kind of more in the corporate world. You know, how do we implement AI? My key takeaway there is be very strategic and deliberate with your implementation of AI. Pick easy use cases with high ROI. Start small, prove the value, and then use the excitement and momentum to scale across the organization.

That's really the best way to embrace and adopt these technologies. Don't start with a big bang. Start small. Learn, iterate, and scale across the organization. 

Lori Boyer 08:07 

Okay, I love both of those. So number one, don't be scared to embrace the change. You're going to get left behind if you're not doing something. And number two easy use cases with a high ROI.

That sounds amazing, and I'm going to definitely ask you, hold you to that. What are easy use cases that also have a high ROI? And then, you know, scale from there. Love that. But first, let's, before we even talk about that, give me a little background on what you think. You know, how is AI, what does it mean here in the logistics and kind of supply chain space? What, what is, and how is it different, I guess, than outside of it? 

Ibrahim Ashqar 08:48 

Yeah. So I think there's three levels here and I think they're all kind of meshed under this one term AI. And I kind of want to break up, break it up into three parts. And I think it'll help kind of set the framework for how to pick use cases, how to think of what, you know, what, what can, what can be done.

I feel people always confuse just basic analytics and statistical models as AI. You know, you synonymously use the word machine learning with AI. And people think of large language models like ChatGPT as AI. And in own terms there's, you know, there's, I get why the terms kind of got muddy there.

 But let me, let me kind of break it down. I think in logistics, just being able to have basic analytics and run statistical models adds a huge huge benefit, right? Like you have statistical models that can forecast demand. They're not, there's nothing crazy. There's no AI in it. It's just a really smart algorithm.

Do you want to consider AI as a really smart algorithm? Some people do. But so there's like your statistical models, just your advanced analytics that add a tremendous amount of value just for things in, in  in, in, in analytics, like sort inventory in your warehouse  according to kind of strategic pick locations.

Right, like placing the inventory, you know, high-velocity inventory closer to the pick station and the ones that are slow moving towards the end of the warehouse, you can minimize the distance traveled, right? It's not really AI. That's just a simple minimization problem, right? But it's advanced analytics, it's statistical algorithms, it's, you know, and it's very valuable. So that's layer one. 

Lori Boyer 10:25 

Okay, so layer one. And I want to back up there just a minute. I love that. Because, honestly, AI is a little bit. I work in marketing. I know all about, it is a term that people love to plug into like pretty much everything. I read some study once that said, oh if you just put AI onto your description then it's going to improve, you know your sales and whatnot. So for our audience, what I'm hearing from you, and I think it's so critical, don't be confused about what AI is and what it isn't.

Analytics are a really cool, awesome thing that absolutely, we talk about that here. I've spoken with multiple guests on the importance of analytics. And in this industry, they're critical. That doesn't mean it's AI. That's not what AI is. Is that what you're saying? Is that what I'm hearing? 

Ibrahim Ashqar 11:10 

A hundred percent. I think that's very true. But you know, a lot of people try to package things like, oh, these statistical algorithms are now AI, right? Just so like, it sounds better, right? Cause otherwise.

Lori Boyer 11:20 

It's a marketing technique. 

Ibrahim Ashqar 11:22 

It's a cool model. You're gonna be like, I don't want your statistical model. I want AI, but hey, that model works really well.

If not, you know, you, you can't ask ChatGPT, hey, go forecast demand for me, right? It just doesn't, it doesn't work like that. ChatGPT or, you know, large language models you know, they're good at comprehension. They're good at generating texts, generating images. Good at coding to a certain extent but it's not going to kind of forecast demand for you. 

You know, it's, an algorithm is going to forecast demand. Now that's like, you know, I want to talk about generative AI, large language models. Yeah. That's like, yeah, right. And, and those are, you know, I think those hit the market in November of 2022. Yeah, November 2022 when it revolutionized the world.

 You know, ChatGPT is powered by GPT 3. The underlying large language model was revolutionary, right? Like understood what you're saying. Responds to anything. 

Lori Boyer 12:20 

Blew our minds, Ibrahim. We're all like, what is happening? 

Ibrahim Ashqar 12:24 

Blew all our minds and sparked a lot of interest and and just like out of the possible, right? It's okay. And then everyone's like, okay, what can I use? What can I use? 

Lori Boyer 12:32 

Yes. 

Ibrahim Ashqar 12:32 

You know generative AI for. That became the new term for a while. And then, and then it was like reinventing the wheel. Use generative AI for demand forecasting. Use generative AI inventory optimization for route, route optimized, for predictive maintenance. All the use cases you can think of, they just slap generative AI in front of them, and it just doesn't work that way, right?

That's not what this technology does.  We can get into how generative AI is really impactful in the world of supply chain and logistics. And yeah, you know, it's, it's, I think some of the, the, the, the uses there are like in summarizing contracts, you know, like your, you know, your, your, your, from your supplier if there's any SLA constraints or something like that.

Lori Boyer 13:13

So do you consider that AI then? The generative AI, the large language models, the, do those fall under what you're describing as AI? 

Ibrahim Ashqar 13:22

For sure. I would describe that as artificial intelligence. Yeah. And I think OpenAI is in general building towards this concept of the artificial general intelligence. And that's when the machines take over and, you know, we're done. I'm kidding. 

Lori Boyer 13:37 

We're like, we've completed. So, with generative AI then, what were you saying some of those use cases are for our audience? What can they do? You were saying contract stuff. 

Ibrahim Ashqar 13:46 

Well, you know, think of what generative AI can do. It can create content, it can summarize content, it can create images.  Can  you know, can, you can feed it images and it can tell you what's going on. And they can code, right. And capitalizing any one of those capabilities is good. So, you know, in, in, let's say you get long contracts from your suppliers and you just want to know, hey, what's the SLA for this vendor.

Right. Instead of having someone kind of dig in to read it, let me go to page 72 of this and, oh, this is the SLA day. I can kind of summarize that document and give you the insight very quickly. Also things like image detection, right? If there's like, you know, you can give a picture of something to AI and say, classify it for me.

Right. So it can, it can very easily kind of just tell you like, hey, is there something missing in this order or not compared to what's, what's there? I think there's like a really cool vendor, Rabo where they're, you know, they're essentially computer vision AI technology for pick stations.

Right. They just kind of give you incredible visibility of what's happening on the pick station to improve order accuracy. Like, hey, here's, here's all these items that are there. And you missed an item, right. And you know, flag it. And it gives you full transparency that wasn't, wasn't really there. And yeah, so, you know, that's, you know, I would definitely say that's, that's in the realm of AI and in the realm of generative AI.

Another aspect is its ability to code. I think there's a really cool use case there. So think of, you know, obviously logistics companies, they're not software companies like, you know, whereas we're, we don't code or anything like that. But you think of the process needed to extract data insights from your database or data warehouse or sorts.

You have a data team and they tend to be writing in SQL or Python to extract insights, right? And that's something AI is very well suited to do. So now you can, you know, empower business users who usually are overly dependent on data teams to extract insights from these systems with the ability to talk to the data themselves.

Right? Like, what are the shipments that are five days past SLA date? Or hey, what are my slow moving stock? Which of my carriers has been giving me, you know, for this particular lane better prices than others. Right. And that AI can, can translate your question into code, run the code in the database and give you the answer back instantly.

Basically bypassing the data analyst in this case. Not replacing the data analyst. I think the data analyst still needs to manage this system. It's important. It's not gonna replace jobs, it's gonna just, it's the future of work. It's augmenting. 

Lori Boyer 16:27

Yes. 

Ibrahim Ashqar 16:27 

It's culminating this manual back office function, which is extracting insights from your WMS or your ERP or your OMS systems.

They used to kind of run your logistics operations. And, you know, just empower you truly to get self-service analytics, demarketizing access to the people who know what question to ask, know why they want it, but don't know how to get it because they're not technical. 

Lori Boyer 16:51 

Right, right. So you see AI being kind of the best-case scenario, a way for regular mortals like myself who don't code to be able to communicate with complex systems in natural language and just asking the questions, you know, which is the best option for me? And then it will run that code. That's, that's where you're seeing some of the best use cases of AI. 

Ibrahim Ashqar 17:18 

Correct. For generative AI, that's, that's for sure. For generative AI. I want to go back to the, you know, you said there's analytics, which tends to be confused as AI. Then there's generative AI that I just described.

And then I want to say the third bucket, or, you know, bucket two, and you know, how I have in my evolution. Is what we probably call narrow AI or machine learning. Predictive models, right? Like you're trying to build a forecast demand forecasting engine. There's all sorts of algorithms out there. Neural Networks, XG Boost Random Forest that can be used for demand forecasting and all of those, you can call them as a kind of machine learning or, you know, the umbrella of AI, but they don't, they're not answering your question. You can't ask it a question in plain language, give you an answer.

That's not what they do. What they do is they take  they've been trained on a large corpus of data.  And they do a very specific thing. Like I'm going to predict how much demand we're going to see next week. Right. And you can take this demand prediction to help inform labor planning decisions downstream.

Lori Boyer 18:16 

How is that different than the analytics piece? 

Ibrahim Ashqar 18:18 

So analytics is, I would say you can, yeah, it's true. It's, it's a, there's a blurred line there. I would say it's in the sophistication of the algorithms. It really  is kind of how you would separate it. There's things that are very clearly under the realm of machine learning that aren't analytics.

Like for example, for demand forecasting. And a statistical approach would be using the 52 week moving average. That's just basic, average. You look at average demand over the past 52 weeks and say, hey, that's a proxy for future demand, right? And you know, for the longest time that was industry standard, by the way. But there's, you know, all sorts of you know, assumptions and limitations of just using a simple moving average that can detect seasonality or sudden, you know, spikes caused by promotions or anything like that. Because just saying, hey, the average of last week is where you're going to get, or average of the last 52 weeks is what you're going to expect next week.

What if next week's a time period, you know, Black Friday, or there's some promotion running or some seasonality or some, some event you, you know, you. The moving average is not going to detect that. These other more sophisticated algorithms, XG Boost. You know, under the world term of machine learning they're able to kind of pick up on these peaks and troughs.

Lori Boyer 19:31 

So, and you feel like all of these layers, as you're putting them, have a place in the logistics and supply chain. They don't necessarily all fall under AI, technically, but they do all have kind of a role to play on their own. Is that what I'm hearing? 

Ibrahim Ashqar 19:47

In, in the analytics, the first layer analytics, that's like your business analysts, business intelligence folks, they're usually embedded in the business function.

There's a requirement for them to be danger enough with SQL. Understand, but have to be really, they really understand the ins and outs of the business, right? Because they're surfacing metrics for the business user. And you have your data scientists who are, you know, they're a bit removed from business, but they're there just building predictive models, right?

They're the ones creating the, the, the predictions that, the, the business analysts would then use to, you know, convey the story like, oh, okay. Based on the forecast we're getting from the data scientist I'm going to use this figure to tell you how much labor we need in our warehouse next week.

Right. They're not the ones doing the predictive modeling. They're the ones converting that into a way that business users can use. That's like the difference there. 

Lori Boyer 20:33 

Yeah. That makes sense. That makes a lot of sense. And then the large language models falls into, that's the third layer, right? 

Ibrahim Ashqar 20:40 

Yeah, that's the third layer. I think it's like a new area altogether. Data teams are, you know, large language models. People are still experimenting, learning, like how can we use large language models? Every company is trying to explore and it's like, okay, how can we, you know, this is transformative technology. It's awesome. A lot of, a lot of the questions I get from executives, like, hey, listen, like, we know we want to invest in this space, but where do I start? Right? Like what, how do I answer this? 

Lori Boyer 21:03 

Yes, completely, completely. That's exactly what everyone, that's why everyone's talking about AI, why everyone's talking about large language models. We can see, it's very easy to see the potential, and to see the fact that there's gotta be amazing, you know, use cases. But it's a lot harder to figure out where to start and how to implement and, and move forward with that.

So are people super excited about, I'm just curious, like within the, the programmers, developers, are people excited? Are people scared? What, what is just their general, what do you feel like is the vibe there? 

Ibrahim Ashqar 21:37

Yeah, I mean, I think in general, people, if you solve a problem for someone, a pain point in their work workflows, they get super excited about it. Right. 

Lori Boyer 21:46 

Okay. 

Ibrahim Ashqar 21:46 

So back when I was leading data teams at Stord had about 12 12 people, you know, the peak of it. And we were, we keep cranking out dashboard after dashboard. And, you know, we're trying to hit this utopia of self-service analytics, like a one-stop shop where all the business users can go and, you know, get the insights they need. Right. 

Lori Boyer 22:09 

Ibrahim, I'm the one asking for all those dashboards. 

Ibrahim Ashqar 22:12

Yeah, exactly. 

Lori Boyer 22:13 

I know you're smart. Make me a dashboard. 

Ibrahim Ashqar 22:16 

And you know, it's like, and it was like, it's like almost like, it's like a utopia that you can never actually chase or, or actually get to. It's almost like a cartoon where, you know, you have like a stick and they dangle a carrot in front of them and you're running out, the stick stuck to you.

So you're never going to get there. So that's kind of, you know, how it is. And it's, it's because, you know, despite the number of dashboards we, we produced people always kind of just bypass the processing. Hey, I want to go to the data analyst, my go to data analyst, give me this metric, give me this answer you're going to do it for me.

I, you know, I, I want, I either don't know how to use a dashboard. Two, I don't know if the dashboard exists. Three, what I want doesn't actually exist on a dashboard. 

Lori Boyer 22:56 

Yes. 

Ibrahim Ashqar 22:57 

Right. It's like a variation. And so the default reaction is I'm just going to go to my go to data person. And data folks are, you know, I think they're pretty nice and they like to be helpful.

And it's almost like it's bad in this instance, because like the second you say yes, the other person's like, oh, I'm going to keep coming back and coming back. And, and, you know, this gets, it becomes a vicious cycle where, you know, everyone just kind of goes to their go to data person. All of a sudden data teams are swamped with these ad hoc data requests, very simple, low level tasks of like basic things, right. That, you know, it's, it's, it's, it's not a good use of what expensive data resources time, right. Right. 

Lori Boyer 23:39 

So it excites them if they can simplify that somehow, hand it off, is that what you're saying? 

Ibrahim Ashqar 23:44 

They're super excited if AI can take care of that for them, right? So you're telling me, listen, I can tell the AI about how my data is structured, and you're telling me it's going to answer all the questions that I get bogged down 50 times a day for. Yes, sign me up. 

Lori Boyer 23:57 

Yes, okay, cool. 

Ibrahim Ashqar 24:00 

People always thought like, oh, aren't you replacing data analysts? No, we're empowering them to actually work on the things that are actually driving value  and, and you know, strategic initiatives of the sort. Not have to give you a report on, you know, give me a report of all the shipments that were five days past late and I needed yesterday kind of a thing.

Lori Boyer 24:18 

Yes. Right. 

Ibrahim Ashqar 24:18 

You know, the loudest, loudest person in the room. And there's a lot of teams, a lot of, I think all data teams resonate with this. Like. There's always going to be ad hoc requests. That's like a part of the, the job, right? But everyone can work on, you know, some, some cool things. But if your cost is just working on ad hoc requests, it's demoralizing. And it's not a good use of time, really, given the cost of data resources. 

Lori Boyer 24:37 

How much time do you feel like is spent? Or was spent, I guess, even before AI, on those kinds of requests, the ad hoc requests, the Lori Boyers coming in and saying, I'd really like to get a dashboard on how these things performed and which one was best. And totally guilty, sorry, apologizing to my own data scientist team, but yeah. How much time is being wasted, would you guess? 

Ibrahim Ashqar 25:00 

I actually did a survey and I have a definitive on this. 

Lori Boyer 25:03 

There we go. We got it. 

Ibrahim Ashqar 25:05 

I, I, I'll tell you like earlier today as well, like I was just talking to someone who said they, they saw, you know, they, they, they saw data teams of 50 people strong and all they did was ad hoc requests, like a hundred percent of the times, like it was nuts. It was crazy. 

Lori Boyer 25:18 

Stop it. 

Ibrahim Ashqar 25:19 

Yeah. And that's, that's unhealthy. I think when I joined Stord, there were a few analysts there to begin with. And they were also kind of like just swamped with requests or just actually just doing work from the loudest person in the room. So it was completely ad hoc.

Towards the end, it was like, we couldn't eliminate it, right. It was still about every sprint. And we had like weekly sprints. Took 25 percent of time was like dedicated to ad hoc requests, at least. But you know, I've, I've seen people say it's 50 percent and I've seen like best-in-class people best-in-class companies say like, oh, we've actually got it down to like 15 percent or something, but it's still.

Well, you know, it's pretty impressive, but I would say most people are still kind of hovering in that  25 to 40 percent range, and that's kind of what the survey showed. 

Lori Boyer 26:03 

Well, and that's, it's, it's not a low-paying job when you, you know, when you're thinking about all the hours, and that adds up to what you're paying the data scientists. Aren't, you know, a 10 bucks an hour kind of job here, and so that gets very expensive. And it is, even though you can try to put boundaries around it, I know here at EasyPost are, again, shout out to our engineers. I love you. They work hard to try to get some boundaries to protect their time a little bit because they do get so many requests.

 But even with boundaries, it's hard to, I think it's hard to just fully protect. So, I'm really excited then to hear, so you're saying AI then is a great use for all of those kind of time-consuming requests that actually aren't that technical and don't need the depth and and and breadth of knowledge of these data scientists that they have, but it's taking up their ability on kind of a shallow end. 

Ibrahim Ashqar 26:58 

Agreed. Yeah, it's a great use of of AI for that sense. Yeah, just the ability also kind of just to pull information through large documents. You know, let's talk, you know, we talked a lot about data, structured data, but let's talk about unstructured data as well. All the documents that you have, the large manuals on how to repair a certain item.

That's, you know you know, maybe more so in, in the world of manufacturing, how do you repair this machine center or something of the sorts. You know, if you kind of connect all those unstructured manuals on data sets and things like that like content  I can kind of skim through them It can read very quickly and it can summarize very quickly. 

Lori Boyer 27:37 

Okay, Ibrahim. I'm gonna totally tell you this secret story about me. Okay, I'm gonna hope that … I had a product person here and they sent me this huge list of documentation on a new product rollout. And they said, we're looking for some marketing, we're needing to. And I ran it through AI and said, can you just give me a summary. And you know, what are the most critical elements?

What are this? I put it together, I'd, I'd whipped something back and he was like, holy cow. How are you so smart? I didn't tell him! I didn't tell him! I was like, I know, I'm just a genius. I, you know, he's like, I can't believe you like succinctly understood that so quickly and got it down. And so yes, that's my embarrassing story of me taking credit because it was really AI. That is exactly the kind of use case you're talking about.

Ibrahim Ashqar 28:35 

Well, you know, you're, I mean, people are going to find ways to incorporate this into their workflows. They want to be more productive. And you saw an opportunity, you acted on it, right? 

Lori Boyer 28:43

I didn't want to read the 40 pages of documentation and technicalities. And there was code in there. And I was like, okay, just tell me what this is.

Ibrahim Ashqar 28:50 

Exactly, exactly. And then, you know, think of large enterprises where that becomes like, oh, well, you can't feed in proprietary information to ChatGPT which is going to use the information to model. 

Lori Boyer 29:00 

Great point. That was a question I was going to ask. Security. 

Ibrahim Ashqar 29:03 

Exactly. So, I mean, like, that, that's where, like, the enterprise grade permissions and security comes into place, right? And the need to have, kind of, solutions in house, localized, that, you know, ChatGPT is quite generic. It's awesome. I love it. Yeah. But it's anything by your business. 

Lori Boyer 29:18 

Right. Exactly. 

Ibrahim Ashqar 29:19 

You could customize AI to, kind of, learn your lingo, right? You know, like, when we say, you know,  UAM, we're referring to merchants with at least one paid order. Or a new customer is this. Whatever custom definition you have or whatever lingo you guys use. Like we had a customer who you know, they collectively referred to raw materials as this very specific subset of products in their catalog, right?

But, you know, there's nothing, there's no item called raw material. It's just a very specific, you know, where the item category code is, you know, X, Y, Z, whatever. 

Lori Boyer 29:51 

And no one else in the world calls it that. And every time you have a new employee who comes in or so, you're, it's like, you're learning a whole new language. And yeah, I, yeah, the internal lingo. 

Ibrahim Ashqar 30:01 

Right, the internal lingo. And I think companies have an opportunity now with AI to like, you know, it's almost like, you know, you onboarded a new person into your organization. What do you want this new person to know about? Right? What data sources do you want to expose to it?

And what do you want it to know about your business? Right? And you can teach it all of these things. And you only need to teach it once because it's smart enough like that. You add it to its knowledge base or brain or whatever. And then the AI now can, you know, speak to you in your terms. Right? In EasyPost language.

Lori Boyer 30:33 

Or sometimes when I talk to AI and I'll say, Talk to me like I'm 10 years old. Right? Because I just don't understand this. Okay, maybe like I'm eight years old now. You know, so that it's fantastic at explaining and teaching. 

So let's talk though, like security. So let's pretend we're onboarding a new team member. It's an AI. It's maybe a platform like Lumi or whatever it is, whatever you're using out there. How do you make sure it's secure? What, I mean, what are you looking for? That's a big fear people have, you know, and I know you're, you're item number one, don't be scared to embrace change. But there's good reasons people are scared, right?

 There's security risks, and we hear about those things all the time. So. How do you, what should they be looking for. Let's say that you're decided to use a new AI platform or something, how can you know it's secure, how can you know you're safe? 

Ibrahim Ashqar 31:28 

For sure, so I think, I think I want to address this in multiple ways, there's secure from a data security perspective, or from like quality of output perspective, right? You don't want the AI giving you wrong answers. 

Lori Boyer 31:38 

Yes, yes, yes. Like that lawyer who put something into ChatGPT and it came up with a fake story and he went to court with it. You know, as a fake court case, you know, something like that. We don't want to be embarrassed that way. 

Ibrahim Ashqar 31:49 

Side tangent, there's a funny story here. There's, actually I can't remember. I think one of the car companies, automotive company, they added the chatbot to their website, or it might be a dealer, something like that. And the person, the person started talking to the chatbot and realized, oh, you know, this is pretty smart, pretty cool. And then somehow tricked the AI into giving him an offer to buy a car for, for, for basically, you know, well below what it's actually supposed to be. And he's like, hey, this is what the AI told me. This is what your bot told me. 

Lori Boyer 32:19 

We definitely want security in terms of business practice and the, you know, as well as risks. 

Ibrahim Ashqar 32:26 

Creating quality of output and security and data security, especially in enterprises. I know that. I realized in more in the mid, small to medium size, people don't care as much about their data.

I think it's stored in the cloud already. So like, yeah, you know, everyone has my data. It's not a problem. But with big enterprises, they're so hesitant and reluctant to share data with anything. And, and, and, you know, it's, it's funny, cause like, I think, They're all like, oh, you know, you're going to be using our data to train your models.

And, you know, we, we, it's not fair and things like that. I explicitly remember talking to an executive at one of the largest food and beverage companies out there. And I was like, you know, if you use OpenAI via their APIs or the enterprise package, they say that they don't, they actually delete all the information. And they don't retain any of that. He's like, yeah, but I don't believe them. 

Lori Boyer 33:14 

Right. There's a trust issue. 

Ibrahim Ashqar 33:16 

Yeah. Even to that extent, there's like a trust issue. People want to make sure that data stays within their environment as my enterprises, at the enterprise level. And that's, that's right. You need to make sure that the, you know, you got, you got to play nice with the, the, the IT team, so to speak.

So if you're, if you're, if you're a software, AI software provider and you're telling them in order for you to leverage our system or our AI software, you have to give us your data, it's probably not going to fly. And so you've got to kind of go around them and, you know, make sure that you're, you're addressing all their key concerns.

Data stays within the environment. And I think that's kind of how you address the data security aspect of it in, in terms of just data security. In terms of the hallucination aspect that the problem is wrong responses. I mean, it, I think this is something that you know, it's very prevalent in just large language models in general, they sometimes give you wrong answers, right?

It's, it's, it's fact of truth. And I think you have to be very careful and add enough guardrails. So that you minimize that risk as much as possible. But I don't think you were there that where technology is not mature enough yet, that it's fully eliminated. I think that risk is always going to be there to a certain extent.

Lori Boyer 34:26 

And that's why humans are still needed too, right? I mean, we have to be keeping an eye on this as well. We're not ever, well, I can't say ever. Maybe someday all the robots will rule the world and we'll lay around and drink martinis, but right now, definitely not. 

Ibrahim Ashqar 34:43 

And that's why I always say I was, I'm pleasant and respectful and I chat, chat to ChatGPT. Please and thank you. Just in case it remembers me in the future. I want to make sure. 

Lori Boyer 34:55 

That's right. That's right. Someday you're like, I was the nice one. Remember? 

34:58 Ibrahim Ashqar 

I was, nice one. I wasn't, I wasn't mean to you. 

Lori Boyer 35:00

I gave you gracious feedback. Didn't yell at you. Yeah, exactly. So, so I, I think that's an important thing. That is one of those risks. A business is at risk at a time. I, you know what I mean? That's how you succeed, all the biggest companies out there. You have to be willing to, as you said in your number one, don't be scared to embrace change. That doesn't mean you embrace change without putting up guardrails and without having, you know, little things on your chatbot that if it offers you an amazing deal on your car that this is still going to be reviewed by a human or so, you know disclaimers things like that. 

Ibrahim Ashqar 35:35 

Yeah. And another way is like just being transparent in how it came up with the answer. Right. And so like, you know, I think building trust through transparency is huge. A lot of the problem with, you know, going back to narrow AI, you know, these machine learning models is that sometimes there are black boxes. As in, hey, the forecast told me that we're going to sell 322 units next week.

Why? I have no idea. That's what it's really, you know, really the data science will tell you because they're accounting for so many features, they're like, yeah, you know, it's 22 percent cause of this. 14 percent cause of that. 5 percent cause, you know, this reason. And like, that's like, it's actually very hard to give you a straight answer as to why the AI predicted that.

 With generative AI, I think you have an opportunity to be very transparent with, with the user, with the question, like, hey, here's the answer. And here is a supporting facts or here's where I pulled, you know, here's the references. Here's where I pull this information from. Here's how I thought through it. Here, let me state my assumptions very clearly. Let me walk you through step by step, how I went through it. And I think that's very useful because, you know, from the user's perspective, they have all the information they need to be like, is this the, is this insight actually correct? Or did the AI misunderstand what I meant slightly?

 And if it did, you can just like reprompt it, hey, actually, you know, you use the ship date. I want to use the delivery date actually for this analysis. And you, you prompt it accordingly. 

Lori Boyer 36:55 

Yes. 

Ibrahim Ashqar 36:56 

Yeah. Because one thing I've, we've learned from, and you just from, from seeing how people interact with, with AI is people tend to be very vague in the way they communicate.  Yeah. 

Where AI is very literate.  Very, very explicit and will take your word to a tee.  And so there is a bit of a communication gap  at the moment. And I think it's only going to get better with time, but you have to be cognizant of that. 

Lori Boyer 37:19 

And it is a skill to prompt people correctly. I know my husband and like, and I will joke because we'll play around on ChatGPT or something like that. Trying to find answers to things and he's terrible at it. I don't know why. He can't, and I think that's what you're saying. He'll ask vague questions, or, and of course I'm, I have a long history of research and, and reading and, and, but it is kind of a skill to learn how to ask the right questions, how to get the right follow up questions, how to know how to move it through it.

And it, again, kind of that place where, so I guess for you, I'm going to say that place where humans are still necessary. But it brought the question for me, for you, if we are talking to a company or something and we're wanting to make it simple, what recommendations do you have for helping your staff or. So let's say, explain a little bit about what Lumi AI does first. And then explain how you would teach people to train their staff or something to use it. 

Ibrahim Ashqar 38:13 

How do we bridge that communication gap, basically? 

Lori Boyer 38:15 

Yeah, right. 

Ibrahim Ashqar 38:16 

Yeah. So, a bit about Lumi. We use generative AI. Our use case of generative AI is essentially so that it can go into the database, retrieve the insights, and get it back to you. So making extracting analytics an easier conversation. 

Lori Boyer 38:29 

Okay, so kind of replacing that data scientist piece of all those ad hoc requests, all that. 

Ibrahim Ashqar 38:34 

Exactly, bypassing, bypassing the data resource that you would have to go to. Just ask your question in plain language and the AI will do the hard work for you in a few seconds and give you the insight. As opposed to you having to wait three days or something. Now we gave, we gave people this tool and we realized, oh, people actually don't know what questions to ask and, you know, how to ask them. So how do we, how do we, you know, we, we, in our attempt to minimize hallucinations, when the AI detects the question is vague, it actually asks for clarifying questions. Right? Right, and it rephrases it for you. And it's like, oh, okay, yeah, that's actually what I meant. Let me try running the suggested prompt. So that's one way you could do it. The other aspect was through prompt playbooks of sorts. So giving you, hey, try out these prompts. We know you have a, we know you have a WMS.

We know you have an ERP.  And you know, the data sets are. We have the inventory transactions. We have the shipments table. We have the orders table. We have the procurement tables and things like that. Like all, all part of a normal supply chain organization, right. And so it's like, okay, well, we know what kind of questions are interesting to these users, because that's the space we've built in.

So we give people this playbook that's embedded, like, hey, try these questions and just, you run them and you're likely to find some interesting tidbit, and we always try to make these insights very tight to like, actionable actionable results. So like their insights to either reduce costs, free up working capital, or in some instances actually help you increase revenues to a certain extent.

One, what we're trying to do there is offer ROI right up front to the users and to teach them how to prompt, inspire them, say like, oh, here's how I structured my prompt and here's how the AI responded to it. And so people, you know, people learn. People like, oh, okay. So this is how I need to communicate with AI, or here's, here's how AI likes to be communicated to. 

And so you, you, you start picking up on these prompting best practices and habits, and eventually you're going to get better. We're also kind of hedging on the bet that people inherently, as they interact with AI in their daily lives, in all other platforms, we're going to get better at communicating with AI. So it's just a natural wave that's going to happen. 

Lori Boyer 40:49 

I say, and I've spoken to other people before when we've talked about AI and analytics, and one of the recommendations that I think is absolutely critical here. If you, even if you are not yet adopting AI or using large language models or doing, please get on things like ChatGPT in your own personal life and start practicing using it.

Like my husband and I trying around with it to get things. Because it is kind of a skill you'll just learn, right? And so, even if you haven't adopted something, it is so critical. Critical for our kids to learn. It's critical for everybody to figure out how to work with it. So anyway, I just, I wanted to back that up. That is something I've heard from multiple people. You've got to get in and just start using it. 

Ibrahim Ashqar 41:31 

I agree. I mean, it's like as transformative, as I mentioned earlier, it's as transformative as the internet. 

Lori Boyer 41:36 

Yeah. 

Ibrahim Ashqar 41:37 

It's almost like someone in, you know, 19, 1999 saying, I actually don't want to use the internet. And I don't believe it's going to be a thing. Well, you're going to be left behind. 

Lori Boyer 41:44 

My husband always says 12 o'clock flashers, which then just ages him because that comes back from when you'd have a VCR and the 12 o'clock would flash because you couldn't figure out how to turn on the time. And that would mean you're old and not up with technology, right?

And so that is kind of, we can't get stuck in that mindset of, oh, this isn't important. I don't need to learn it. It is really critical. We are totally out of time. Ibrahim, I could speak to you for about 20 hours. And maybe we'll have to have you back sometime because I would love to talk about all kinds of AI uses within the warehouse, within business, within inventory management. You know, where are we and how. But I love, love, love your insights today were right on. Don't be afraid to embrace change. 

Okay, the bigger your organization is, the change might be small, but still try to do something embracing change. And your second point is to get those use cases that are easy to put into place, but are a high ROI. And that's going to be the kind of uses that we're just talking about in terms of you know, spare saving time for your data scientists and whatnot like that. Are there any other use cases that you'd like to throw out really quickly at the end that are a great high ROI?

Ibrahim Ashqar 42:58 

Yeah. I really think there's, you know, within the world of supply chain, again, across three verticals of ai. The simple analytics, the narrow analytic, the, the, the narrow AI and generative AI.

There's use cases across the board. I mean, we're, you know, the one I talked about with Rebo is awesome. I think where you're leveraging, you know, AI's ability to see to improve picking visibility, right? And improving order accuracy. And you know. The benefits, you know, 'cause you don't actually realize the cost of not fulfilling an order properly.

Right? It's, you know, one, customers dissatisfied, and two, you have to deal with the reshipping and, and all of that stuff. So, you know, if you can use AI to eliminate these like missteps that happen in from pick pack and, you know, ship to the customer, you, you should embrace these technologies. And so that's another, another good one I think to consider. 

Lori Boyer 43:49 

If people want to connect with you, are you on LinkedIn? What's a great way for them to connect with you and ask for more use cases or suggestions, or even learn about Lumi AI? Can they, you know, connect with you on LinkedIn? What's the best way to get in contact with you?

Ibrahim Ashqar 44:03 

Yeah, for sure. I'm definitely on LinkedIn. I try to post educational content, or I think at least of AI applied to operations in particular. So definitely feel free to connect with me on LinkedIn and happy to chat. Also kind of visit our website. It's lumi-ai.com. You know, and our emails are there and you can kind of reach out to us.

So yeah, happy, happy to chat to anyone who's interested to learn more about AI and how to integrate it into their operations and reap the transformative benefits it has to offer. 

Lori Boyer 44:34 

I love it so much. Thank you so much. This has been an incredible conversation. Seriously. I love AI so much. I'm a nerd at heart as well. And so I kind of geek out over it. And loved having you here. Loved hearing about actual use cases. So please everyone, if you have questions, reach out. Again, appreciate you being here. Any final words of advice or just a sign off goodbye to our community here? 

Ibrahim Ashqar 45:02 

Honestly, thanks for having me, Lori. This was awesome. All the listeners, you know, thanks for tuning in and I think embrace the technology. You know, we're still in such early stages. There's so much experimentation. I think a lot of the innovation comes from just playing around and finding use cases of yourself and saying, Oh yeah, that's a good one. So, you know, don't be shy, play around and explore. AI is here to stay, I think, and  it's going to make all our lives easier, I hope. 

Lori Boyer 45:28 

I love it. Thanks everyone for being here. Take care. Connect with us on social media. Let us know what you're doing with AI. Give us, share your ideas. We're going to have some sort of big geek conference together where we'll all talk about what we're doing. So we'll see you guys next time. Bye bye.