Scaling AI – how to make it successful with Luca Boschin

AI offers plenty of benefits, but before bringing those end-goals to life, AI has to be scaled and here the statistics are truly bothering. In fact, even though more than 90% of organisations are investing in AI today, only 17% have been able to scale it. It seems that finding a balance between the strategy and AI is not easy.

Today we are going to discuss if there is anything we can change in order to increase our success rate in that matter.

Michał Grela (MG): Hello and welcome to yet another episode of IT Leadership Insights by Future Processing. Today, my guest is the Luca Boskin, and we’re going to talk about scaling AI and how to make it successful because, well, it’s no secret that AI is at the top of every organization’s strategic and tactical agendas, and if not yet, then I suspect it will be really soon. And on the one hand its goal is to, of course, manage the costs more efficiently, et cetera, but on the other, there are more factors to it. And the factors that should influence in a company improving customer engagement, providing more insights from data and boosting growth, et cetera. There’s plenty of benefits of AI, obviously.

However, before bringing these end goals to life, actually, harnessing these benefits, AI has to be scaled and here the statistics are in fact truly bothering. As we’ve learned from recent research, for example, from KPMG, even though more than 90% of organizations are investing in AI today already, only around 17% of them have been able to scale it, which is devastating. It seems that finding a balance between strategy and AI, and scaling AI is not easy. So that’s the topic of discussion today. We’re going to discuss if there’s anything we can do or change in order to increase the success rate in that manner. And the subject matter expert today, Luca Boskin. Super happy to have you here with me today. Thanks for joining the podcast.

Luca Boschin (LB): Hi, Michael. It’s nice to be here.

MG: If you could please introduce yourself and share a few words about yourself, please.

LB: Yeah, sure. Absolutely. So my name is Luca and in 2014, with my co-founder Alessandra, we started an image recognition technology business. So what we have focused in this company was developing and licensing image recognition technology with a specific variance, which is that we have historically focused on the detecting things inside the images and videos which are related to branding, to brands, to marks everything that has to do with a company branding. And then over time, we have expanded those services with additional collateral image recognition service that allow to boost that core central service. So things like object and scene detection, visual search, custom object detection.

And what we do, we license this technology to organizations, enterprise, larger organization, who are looking to either build new products with our technology or power internal services with our technology. So maybe I’ll give you quickly few examples of the type of applications that are built using our image recognition technologies. It goes from things like social media monitoring where most of the big players in this space, so companies like Brandwatch or Sprinkler are using our technology to extract brand insights from images that consumer’s share on social channels. Or another example of sector where our image recognition technology is used, would be counterfeit detection where clients like eBay, the Marketplace or Red Points, which is a very well known counterfeit detection platform, use our technology as the visual signal that allows them to automatically remove counterfeits from those platforms.

And maybe just a quick touch again, to give an idea of the broad spectrum of this technology when properly applied would be things like cyber security. So an example of a client in this space is companies like Mimecast again, who use our visual signal in that case to understand if a phishing attempt has been trying to be made by looking at a phishing email in the same way you would do it with your eyes so that it can automatically detect that’s a phishing attempt or not. So that’s a few examples of where the tech is used and what we deliver to our clients.

MG: Wonderful. Thank you. I guess, it’s a very good example of how can innovative tech be used on daily basis. So I think it’s going to be a relevant conversation.

LB: Yeah, absolutely.

MG: So as a starter, my first question would be touching on the balance between the strategy and AI, what factors in your opinion should be taken in consideration when it comes to deciding on the balance that previously mentioned is often not in place?

LB: Yeah. Yeah. Well, the first point as you mentioned earlier, is that there is absolutely literally no doubt that AI will change the way we do business. And the main reason for that is that it allows us to take out the mundane and boring work that we usually have in our daily lives and businesses, but so that we can reinvest those resources in what we do best as humans and what we all love most as well as humans, which is all of those creative and corroborative tasks that generate real value in inside the business and are those tasks that create actual and true innovation for businesses going forward.

So when we’re looking at AI and working it in our strategy, what we need to make sure the first point is that we need to find tasks that AI can realistically take care of at very large scale. And only then we can start thinking of where we can reallocate resources accordingly, because first we need to see where AI can remove those bottlenecks. And then we can decide how to reinvest the current resources that take away in a manual way of the bottlenecks, so that we can reinvest those again, towards more creative tasks, like innovating the product or figuring out new markets or places where our product can deliver value.

MG: So it’s not about using AI just for the sake of using it. It’s rather finding an actual serious, business wise serious use case that AI will be good to help with.

LB: Absolutely. And as always making sure that it will work at scale because that’s one of the major fail points of AI is that it won’t work at scale, but just in nice small demos.

MG: That’s a smooth transition into the other question I wanted to ask you. As stated before, only around 15, I think 17% of companies have been able to scale AI successfully. So it tells me that there’s a huge gap to bridge still ahead of us. How would you approach scaling and executing AI successfully?

LB: Yeah, sure. So for us, the first question is always what goals are we looking to achieve? Especially keeping a mindset where we try not to think of all the legacy technology limitations we had in the past. And only then we can have a clear mind in order to deliver true innovation and be able to rethink completely the way we do things. And that’s when we can start discussing about AI and understand what it can solve at scale today, but also with a clear path to reaching those long-term goals. So, first thing let’s focus on the goals, remove all the legacy limitations of technology within in our minds, and then figure out what we want to deliver and see which elements that technology can take care of autonomously.

MG: So there’s definitely a clear roadmap for implementing AI and there’s a process an organization can and should follow. But when should companies think about scaling AI in this process of implementation?

LB: Sure. Well, rather than talking about scaling AI, I would rather make sure that the AI we are thinking of implementing will be scalable from day one when it enters production, or at least it has a clear path towards reaching very large scale of use. And the reason for that, is that you want to focus on an AI that generates longterm value rather than, as we were saying before, delivering some short lived great looking, but still short lived demo type of projects. And, the reason is that, those projects, those small demos are the ones that will get you fired because if you have something that works right at a demo stage, and then you want to scale it, but it doesn’t work that’s when are going to get in trouble.

So in that you need to test of core things in a smaller case, scenarios so that you can make sure that it does deliver value and only then you should scale it up, of course, but from day one, you need to make sure that the technology you’re implementing with deliver large scale implementation. Otherwise you’re going to hit a wall and all your work is just going to crumble into pieces.

MG: It’s about scalability by design.

LB: Yes, absolutely.

MG: Very interesting point. But regardless of the size of the project, my experience is that, almost every IT product eventually comes down, it depends on the human factor, people involved in the discussion, the decision-making process, execution delivery, et cetera. And I believe it’s the same with scaling AI. And here from your expertise, what’s your opinion on how important is the factor of these expertise and the human touch behind scaling AI?

LB: Yeah. Well, for sure, to me, you do want to bring an expert in the room when you’re thinking about implementing AI. And the reason for that is that you need to have someone that knows for a fact that what they’re implementing will be scalable. Or even better, they have already proven that the AI they’re looking to implement does work at scale because falling into non-scalable AI is way easier than you can think. I think it reflects those numbers KPMG, you were mentioned before. Despite 90% of organizations investing into AI, only 17% succeed. And I’m pretty sure that most of those cases are due to the fact that there have been some nice demos but then they’re not scalable. So once again, often you’re presented with some very nice demos, but the problem with them, they most often fail spectacularly as soon as you start implementing them at a production level.

So even worse case, if this works at a small scale, then you start making it part of your product. And as soon as you start scaling all the investment, the product and the company future will start crumbling in front of your eyes because you have built something small that looks nice and works great, you implemented in the product, in what you’re setting out to the market, you expose yourself, you start scaling it up, it doesn’t work at scale. And that’s when you get the bad news and the bad times.

MG: So from your perspective, the recipe for success, and in order to avoid disaster, you have to bring an expert into the room, but doesn’t it equals paying the price for it. It always comes with extra cost, et cetera,

LB: Like everything else, if you eat cheap junk food, you will have long-term repercussions on your body, in the same way as if you implement a non-expensive technology in your company, you will feel in the long-term. But in my opinion, even short-term repercussions, when you start scaling up that product on the market. So yeah, it does come with a price.

MG: I really liked this, this comparison. I also believe that it’s sometimes good to pay premium, but it’s always good to look for the value for money. It’s important to understand that, you just sometimes can’t move forward without, or can’t make money without spending money, actually.

LB: That’s exactly the point. So, you may find cheaper alternatives out there, but most of the cases, it’s not an alternative, and it’s going to be way more expensive than just spending a bit more to get up a proper team in your organization, or find a proper supplier, testing them thoroughly, making sure it’s the right partner to work with and then really deliver the success that will allow your business to move forward.

MG: Nicely, nicely said. So last but not least, are there any factors that tend to be missed when implementing AI solutions in a company? Because of course it’s an extremely, extremely important process. And we already said that when it comes to scaling, there are a few things to take a look at, but what’s usually the most overlooked bit?

LB: Yeah, so to make things a lot to work we were just discussing about. So making sure to implement the right teams or the right partners that are going to help you with AI, and it goes back to the people behind them, and it really boils down to values in my opinion. So in other words, making sure that those who would implement the AI also have very strong core values that will be inevitably reflected in the AI, et cetera. And, the values that I’m-

MG: That’s an interesting point. So your point is that the technology will, to some extent reflect the values of the people who build it.

LB: Yeah, absolutely. Not just the technology, but ultimately even the product that you’re looking to deliver. I mean, we all know about the usual well known by now AI by Yesware, it is humans that train the AI. And if the human puts in a bias element in the training, it will reflect in the AI, but to meet something even a bit deeper where I’m thinking, especially at three values that I think are very important for those who are building AI and they would be transparency, collaboration, and passion. So the reason for transparency is that you need to make sure that those who will implement the AI will focus on your success. And will have candid conversations, when something doesn’t work and you need to find a solution and make it work in another way, or figure out what the different way to gather data or implement the AI so that you can find and make the process actually work. So transparency is crucial and working together rather than apart, and making sure that we have all the conversation so that we can solve any challenges together so that we can be successful.

The second one links very closely. So collaboration and the reason is that why AI and technology is great. You need to make sure that it serves a purpose that creates value for the business and aligns with your goals and strategy. So if you don’t have a clear communication and collaboration channel between those who are building the AI and those who want to extract value from the AI. First of all, those who build the AI, they’re not going to be sure what the goals are, and they’re not going to be able to build an AI that the delivers value towards those goals.

But it’s also the opposite where those who want to deliver value with AI, they need to know what the AI can deliver in that collaboration so that they can decide again what their strategy, as we were talking before, end goals are going to be thanks to AI. And my last point was passion because you need to love AI to make it work, and you need to work that extra mile to make sure that it is scalable. And it is something that works because AI is definitely the democratizing more and more, but it’s still a tricky topic that we’re still learning about a lot. So having that extra passion will allow you to know that 10% more that makes a big difference when implementing a project at scale.

MG: Thank you. Thank you. I really liked a bit of both collaboration and passion especially. I also believe that the more you put in the more you can take out, and it’s as simple as that. Thanks Luca for sharing you thoughts. Were I to sum up this conversation, I would say that when it comes to scaling AI, in order to make it successful, you have to follow the scalability by design scheme from day one, you have to find the right task for AI to perform because it’s not a magic wand. It’s not going to do anything, and it’s not going to be scalable with anything.

LB: For sure.

MG: And last but not least find people that will know what they’ll do. And will be experts in the field because otherwise you’ll hit the walls sooner or later.

LB: And don’t be cheap. Not in the beginning at least.

MG: Never. Never be cheap.

LB: At least in the beginning, then you can take it easy. But in the beginning make sure you do things better.

MG: Thanks. That was IT Leadership Insights by Future Processing. Thank you, Luca.

LB: Thank you, Michael. It was great.