Why Do So Many AI Projects Fail?

John Stroud and Jen Shellinck – AI Guides

About this episode

In this episode, we bring you two special guests that have a lot of experience with AI to break down the common reasons why many AI projects fail and prevent yours from falling into those patterns .🚀

We will cover:

– reasons why AI projects fail
– how to frame a newly designed AI project
-what to keep in mind when launching an AI project

Listen to the full episode or read the transcription below. 💪

Read more
This video is hosted by YouTube. By watching this video, you are accepting the terms and conditions of YouTube.

Interview Transcript

Jordi Torras:
Hello, everybody, and welcome again to the Future of Customer Service podcast, where we analyze the latest in artificial intelligence, how to deal with your customers, progress, and the future. And today I’m really happy to have here with us John Stroud and Jennifer Schellinck. Both are founders of a company called AI Guides. So my first question here for you guys is, who are you and what is AI Guides?

Jen Schellinck:
Well, as you know, my name is Jen. Thanks for that introduction. And, yeah, I’m a data scientist. I have a focus on artificial intelligence machine learning. I also have a background in cognitive science, and I have a data science company in Ottawa, Canada, that’s focused on design and implementation of these systems, but I was very excited to come together with John to push forward AI Guides, because we both share this really strong conviction that it’s not just about the technology. It’s about the human piece plus the tech piece, and I’m coming in from the tech side, and John, I’ll turn it over to you to talk about the human side.

John Stroud:
Sure. So I’m in Ottawa, Canada, as well, and I am not a data scientist. My background is in the humanities. I spent about 20 years in public service, in the government of Canada. So my last job there was as an executive where I was responsible for a large team in an organization of 8,000. And my responsibilities were around strategy, HR, corporate services. And while I was there, I got really intrigued about the possibility of how technology and people could work together, and I got so excited about it that I decided to leave the job and set up shop on my own to help other companies in the same area. And then in terms of what AI Guides is, well, it’s Jen and I coming together and we are consultants, and we are who entrepreneurs are calling if they are worried that they’re falling behind in AI and they don’t know what to do about it.

John Stroud:
The origin of the company came about a year ago. We were at a coffee shop and we were just comparing notes on our respective projects, and we noticed that often these AI projects get off track because decision-makers get matched to the wrong solutions. So we thought it would be so much better if we could connect with decision-makers earlier in the process to give them independent advice, vendor-agnostic advice on what was the best solution for their environment. And so that’s how AI Guides was born: us serving as matchmakers between clients, decision-makers, and vendors out there so that they got the best fit for their environment.

Jordi Torras:
Wow. That’s fantastic. That’s fascinating, right? The combination of humanities and science to create a company that, the way I understand, is to prevent these kinds of fiascos before they happen.

John Stroud:
Exactly.

Jen Schellinck:
Yes.

John Stroud:
Precisely.

Jordi Torras:
That’s amazing. And by the way, let me tell the audience, there is a website. It’s aiguides.co, right? You have a corporate website that’s amazing. There’s a lot of material, and I would ask the audience to go there and take a look. And one of the things that I saw there is a sentence that starts with your pitch, right? It says “Many AI projects fail.” That’s kind of the point there. So, of course, I’m really intrigued by that because we sell AI projects as well. So the question is, why do you think is that? Why do you think many or some AI projects fail? What do you think about that?

Jen Schellinck:
Yeah. I’ll take this one on, because as I sometimes say to John, this is kind of the not-so-secret secret in the industry right now, that this is a thing. It’s an issue that people are tackling. And I think a big reason for that is that, I mean, we know already that even just… Well, I want to say plain-Jane IT or software development projects are really, really hard to get off the ground and make a success. Then when you add the AI machine learning part into the mix, that just really multiplies the challenges in some quite specific ways. In particular, there’s a really strong interaction between the technology itself, like the algorithms and the data, and then the infrastructure required for that. If you do not get those aligned really, really well, then you’re going to have a lot of issues. So, to me, that’s a big piece. That’s actually why I’m excited about the AI as a service option, offerings like yours because I think it can help to mitigate some of those risk factors.

Jordi Torras:
Got it. Got it. Absolutely. So there the role is for you guys to be like a consultancy, right? An advisory role where you say, “Okay, before engaging in this and that technology, or technology stack, let’s take a look at what you’re trying to solve.” So, how do you do that? What is the most important thing when it comes to having an advisor, and what are the role and the way that you deploy your consultancy, let’s say.

Jen Schellinck:
Sure. Well, so absolutely. So, first of all, we do think having an advisor is important because, like I was saying, there’s a lot of moving pieces. So I mentioned the technology pieces, infrastructure data, but then you also need to have a very good custom fit with your particular organization, your particular business goals. And so you really need someone to come in and say, “Look, here’s the landscape. We need to understand exactly where you are on that landscape and exactly what other pieces that are out there can help you. And we think that you need someone who isn’t necessarily attached to one particular part of that landscape to be that guide.”

Jen Schellinck:
And so that’s kind of what we do. We start out at a high level, getting to know your situation, what your use cases are. Then we move on from that into more of a deep dive. Let’s really look at where you sit on this landscape, get into the weeds a little bit. And it’s only at that point, after we’ve gone through all of those pieces, that we would say, “Okay, based on who you are, where you are, what your goals are, here’s where we think you should head.” So that’s basically our strategy.

Jordi Torras:
Got it, got it. So it’s like understanding the problem and what is the use case and the objectives and say, “Okay, if that is what you want to accomplish, this is the technologies out there that help with that“m right? So I’m assuming that in order to do that, and that’s how you and me, we know each other, working together on some specifics of this project. So I’m assuming that you have to, on one hand, be talking to a lot of customers, potential customers, potential companies that are dealing with AI, and then AI companies, right? And how do you do that? How do you manage all these relationships?

John Stroud:
Sure. Yeah. And so, it’s about keeping track of what some of the developments in the industry are, plus we’re networked into a couple of different communities that have an interest in exponential technology. And so it’s trying to stay up on the latest literature and being connected with the community. The hardest part is understanding and being abreast of all of the new offerings that are coming out, because there are so many different ones, and there’s just a lot of good choices out there for people. So that’s the good part. The hard part is trying to decide from the dozens or hundreds of choices that might be relevant for them, what’s the best fit.

Jordi Torras:
Totally. I was talking to Gartner the other day and, of course, they talk to many companies, and the guys told me that just in the space of chatbot, which is one of our areas, but just in that space, they have counted 2,000 competitors. That’s crazy. And just like that, AI Guides makes a lot of sense, because how am I supposed to choose? I cannot have an RFP and have 2,000 companies running in there. That’s nuts.

John Stroud:
Exactly. And so, we wouldn’t profess to know all of those chatbot companies, but we reach out to people and then engage with the vendors. So just like we sat down with you to get some detailed specs on how it works and so forth so we can do a vetting process. That way, we’re not saying that we know every good vendor out there, but what we’re saying is that we know good vendors and it’s possible to tell a good one from a bad one, and that all of the names that we recommend to a client has gone through that vetting process.

Jordi Torras:
That makes a lot of sense and diminishes the risk of going south with one of these projects, particularly in a space, and I don’t know if you agree, but in a space that is, on one hand, very exciting. Of course, Jennifer as the data scientist, you know the amazing outcome and results that are possible, but as well, there is a big hype around the entire AI and some of the potential outcomes are sometimes exaggerated, right? So I guess that’s one of the things that you have to make your customers realize, what is realistic on the approach.

Jen Schellinck:
Absolutely. You do sometimes have to get under the cover, open the hood to get a sense of that. So I think that that’s a really important piece of this process. There’s that piece, seeing what is the actual functionality of this piece of technology, and then there’s also that matching piece, because, as you say, if we just look at the chatbot landscape, I’m willing to bet that all of these ChatBots have a slightly different technology under the hood, they’re all going to perform in slightly different ways. So some of them are going to match well for a particular use case, some of them are going to match well for another use case. So there’s the hype piece you have to get past, and then there’s the actual, what is a really specific functionality when you start to look under the hood, as well.

Jordi Torras:
Yeah. Just for the chatbots, some of them, they’re what I call ‘glorified menus’. So it’s essentially, okay, you can click A, B or C, I cannot type anything. So, I mean, a bot? Okay, but where is the chat part of the chatbot in there? It’s more like a menu. Guys, could you tell us some experience, some example with a specific use case of a company that has benefited from your activity?

John Stroud:
For sure. I mean, one example was… So it was a Canadian company and they have X-ray machines all across the country. If these machines break down, then there’s a really big operational impact on them. So the machines were all networked together and they were each generating a lot of data. So what the company wanted to be able to do was to say, “Okay, we’ve got all of this data. Are we able to predict when they’re most likely to break down so that we can engage in preventative maintenance?” And they thought, “This seems like the type of problem that AI should be able to assist with.” And they’re right. Their first call was to a very large, well-known vendor who said, “We would be happy to help, and we can provide you with an enterprise solution. And the cost will be about $400,000, and it could take a couple of years.

John Stroud:
And so their second call was to us. And they said, “That’s just beyond our scope and we don’t want to spend that long on an enterprise solution.” So what we were able to do with them was define precisely what their requirements are, do a deep dive on what data they had, what their infrastructure was, and then we ended up matching them just with a boutique shop that was able to come up with a custom solution for them at a fraction of the cost. And so what the client liked was, well, A) there were savings. It was a lot cheaper to go with a $20,000 option instead of a $400,000 option. But also, by making that recommendation to them and connecting them with this vendor, we were able to give them peace of mind. It was a topic where they felt nervous, anxious, uncertain. It’s a complicated field and they were really worried about making a wrong decision. And so by giving them that confidence, we helped move them along their journey from lagging behind to being more of an industry leader, and they really appreciated that.

Jordi Torras:
Wow, that’s an amazing use case. X-ray machines and predicting, right? Use the past to predict the future based on all that data, which is a classic use case for machine learning.

Jen Schellinck:
Absolutely. It’s really a net ballpark. Yeah, exactly.

Jordi Torras:
Yeah. And 20 grand is very different from 400 grand, right? So, that’s a big impact and an overall solution. And I’m assuming they were happy with the final result, right?

John Stroud:
Exactly. And it was faster, it was easier, and it was just more of what they needed. Yeah.

Jordi Torras:
That’s fascinating.

Jordi Torras:
Uh-huh (affirmative). Yeah. That’s amazing. So speaking of that, which kind of verticals or use cases do you think can benefit more from AI? This X-ray machine predicting engine is surprising to me. So what else do you think is out there?

John Stroud:
Sure. I mean, this is a question that it’s easy to answer and it’s hard to answer. It’s easy in the sense that there are so many use cases, it makes me think, if we were to travel back in time to the last industrial revolution and you were to ask, “Well, what are some good use cases for electricity? And which industries would benefit from electricity?” The answer is, “All of them“m right? And it’s just trying to figure out, well, where do you start? And so that’s the harder part, is picking the good categories. And so for us, when we try to simplify it in meetings with clients, because sometimes they’ll come in with a, I guess, specific problem, but that might only be a symptom of a larger problem in the organization.

John Stroud:
And so we want to make sure that we’re broad enough in considering what are the use cases. And so we try to bucket them into three main categories. One is around customer experience, and for that, in a number of cases, an Inbenta offering makes sense. So, customer service, then operations and that’s where the predictive maintenance, that would be an example of an operational problem.

Jen Schellinck:
Very specific to that organization. Yeah.

John Stroud:
Yes. Thank you. And then there’s back-office support. So it’s just trying to make the business run a little more efficiently on the administrative side. And in terms of verticals, where we’ve been able to apply it, it’s certainly in government, but financial services like insurance and even in professional services like law firms.

Jordi Torras:
Wow. That’s interesting. And, yes, as you said, Inbenta works… I mean, artificial intelligence and machine learning are big spaces. I guess there’s always the controversy, if machine learning is a kind of AI or if AI is a kind of machine learning? There is always, I guess, that we can have all kinds of opinions, but when it comes to AI and especially for user experience, what is the application of electricity there? I mean, everywhere, right? But what do you see that AI can be changing the way companies and organizations are dealing with customers?

Jen Schellinck:
I’m going to field this one because I think it’s really very exciting, the potential impact that this technology can have in that space. So the short answer to that is, basically, I think it’s going to have a massive impact. And I think the reason for that is that, if you think about what customers want in terms of their customer service, customer experience, they’re always going to, in terms of the high level, they’re always going to want more and they’re always going to want better. That is never going to change.

Jordi Torras:
That’s not going to change.

Jen Schellinck:
Right? That’s the way, right?

Jordi Torras:
Like that seems the same industrial revolution. Yes.

Jen Schellinck:
Yeah, exactly. And so when you think about it in those terms, you can see how AI technologies, like the Inbenta technologies, are going to really enable that. I’m very much an optimist on that front because, within the last five to 10 years, we’ve really seen this technology catapult ahead in terms of its ability to do natural language processing, text processing, and start to interact with people in a way that’s just intuitive and familiar. And so I really see that there’s going to be these great… What John and I sometimes call these super workforce teams that are going to be made up of the AI customer experience, customer service part, plus the people.

Jen Schellinck:
And they’re going to come together and the AI will be on the front lines helping people whenever they need it. No more waiting in a queue for an hour to get some help, so that’s going to be great. Then on the human side, they’re going to come in when more nuance is required, more decision-making, more responsibility for customers in that sense. And so, I hope, anyway, it’s going to be a really good partnership on that front.

Jordi Torras:
I’m pretty sure. I’m pretty sure. I guess, I also, right, back to you, the initial point, we have seen and we will see as well, some fiascos along the way, but as well as the great successes in many, many use cases, right?

Jen Schellinck:
Well, I think one of the shifts that has to happen there is moving away from just automation to intelligent technology. Because you were talking about the ChatBots, they’re really just a menu, basically. They don’t have the intelligent part. To me, that’s where the optimism comes from, is that the customer service, if you have these intelligent technologies, then they’re going to be able to provide that level of service that will be genuinely valuable.

Jordi Torras:
Absolutely. That’s what I imagine, right? That you would call using Zoom call or similar to customer service, you would talk to a digital person with… If you could say, “Well, that looks like an actual person and is effective. Always there solving my problems.” You might prefer a digital assistant to a real one. Inbenta tries, of course, to work on that, but we are still far from having customer service that is undistinguishable from humans. But we’ll get there. Pretty sure that everybody is after that because everyone wants it.

Jen Schellinck:
Absolutely. And to some extent… Sorry, John, I was going to say, it’s okay if they’re not completely indistinguishable from humans as well. Sorry, John, go for it. Yeah.

John Stroud:
And I think that can also be good for the customer service reps, too, right? If they’re able to offload some of the more routine calls to their digital coworker, then it can be a better experience for the customer and for the customer service rep.

Jordi Torras:
Absolutely. Right. So you guys have been working with several companies. I’m assuming talking to hundreds of vendors and you know a little bit of Inbenta, right? We’ve working to enter with some examples and use cases. What do you think about Inbenta?

Jen Schellinck:
It is my pleasure to field that question. Yes, Inbenta is great. I mean, we first came across you when we were looking for a good AI-as-a-service for a client, and I’m going to speak to the tech side of things, as I sometimes do. I was looking around, we were talking with you and I was excited to learn about your semantic engine technology and approach because I think it really fits in very nicely with some of the stuff I was saying before, where you’ve done a lot of that legwork to get the technology piece meshing well.

Jen Schellinck:
This is really important from that democratization of this technology point-of-view because a lot of organizations want to get in on this technology, but they have this problem where they don’t have enough data to really fully power the technology. So I see you as having found a way to get around that problem with your semantic engine, where you can say, “No, we’ve done a lot of the initial work for you and we have supplied the initial power to the engine, and now you can benefit from that by bringing your specific use cases to us.” So, yeah, I’m very excited about your offering.

Jordi Torras:
All right. Thank you. Thank you so much. And I can tell you that sometimes we talk to data scientists, and then when they see the semantic model, they say, “Well, that’s not data.” Well, it is data, it’s just a lot of data, pre… I mean, if you speak English and you live in America, you know a trillion things already, right?

Jen Schellinck:
Yes.

Jordi Torras:
… if you just use the raw bits and bytes of language without the nuances of what this language means, then the data, well, you have the opportunity to not really use it the way it should be used. So that’s what we try to do, build this common sense behind the natural language information, which is, to me, and that’s my personal opinion, among all the AI problems that the technology is facing, driving sales, driving cars, predicting the future, I believe that human natural language is the most complicated. And that’s the analogy that I use. Is like, “Hey, by now we should have a lot of self-driving cars driving around our cities and they’re not there.” So it was not as easy, but you know who is doing that, no problem? Cats, dogs, pigeons. They do that. You will not see a dog crashing against something by accident.

Jen Schellinck:
No, they’ve got it figured out.

Jordi Torras:
There’s no accident. They’re smart. However, you will not see a dog speaking English. That’s a whole another level. So that’s the way I try to think in terms of AI. Listen, I will be talking to you guys for hours. That’s amazing and I really appreciate your time. What would be the best way for our audience to find you? To locate you, or to get in touch with you, guys?

John Stroud:
The easiest way is to go to our website. So we’re aiguides.co. C-O. So if you go there, then there’s an opportunity just to book a meeting with me and Jen. And if you’ve got a problem, we’d be really happy to talk about it. See if AI can help.

Jordi Torras:
Amazing. Amazing. All right. So, John, Jennifer, thank you so much. For the audience, remember, aiguides.co. That’s what is going to prevent your next fiasco. So don’t forget that website. Thank you so much. Have a wonderful rest of your day.

Thanks so much for tuning in. This podcast was brought to you by Inbenta. Inbenta symbolic AI implements natural language processing that requires no training data with Inbenta’s extensive lexicon and patented algorithms. Check out this robust customer interaction platform for your AI needs, from chatbots to search to knowledge centers and messenger platforms. Just go to our website to request a demo at inbenta.com. That’s I-N-B-E-N-T-A.com and if you liked what you heard today, please be sure to subscribe to this podcast and leave us a review. Thank you.

Read more
Ebook

How to build a successful chatbot project?

Trying to design your next chatbot project? Is it covering the basic guidelines to be set for success?

Download our ebook and find out the best tips and practices to build a chatbot that gives real value, provides natural conversations, and guides users to meet their needs.