Drivers of change: the role of Canadian startups in the AI landscape
What do Canadian AI startups need in order to compete on a global scale? Are we doing enough to support their lead in AI development? Amanda Lang and Jodie Wallis speak with Mohamed Musbah, head of product at Maluuba, a startup he founded that was acquired by Microsoft in 2017. They also speak with Peter Misek, partner at BDC’s IT Venture Fund, and John Ruffolo, CEO of OMERS Ventures.
Appearing in this episode
Amanda Lang: [00:00:30] I’m Amanda Lang, this is the AI Effect, a new podcast series that takes you to the frontiers of the artificial intelligence economy in Canada. This country happens to be a world leader in research for AI. We’re asking the question: Does that translate into new businesses and jobs for the entire economy? Joining me in Jodie Wallis. She is the managing director for AI at Accenture. So, a big part of her job these days is helping Canadian companies understand this space and what they need to do with it. Today we look at the pioneering startups and the opportunities and challenges that they encounter.
Jodie, it’s such a new space, so let’s start with a look at what Canadian startups, uh, need to do to have a shot at big US and Canadian companies.
[00:01:30] I think the initial concern, when you think about this, is will we ever get enough capital to compete with the US and China, and- and Germany and- and some of the other powerhouses. But, I think the answer lies in something other than just capital. We can compete on the talent, we can compete on the research expertise, and we need to continue to compete on creating an environment where startups and entrepreneurs want to be in proximity to our research centers.
[00:02:00] So, what’s interesting, that a lot of Canadians might not realize, is that Canada actually produces more businesses than any country in the world. As much as Silicon Valley, as much as Israel, and those startups are as efficient as any business anywhere you can find. Where we fall down is when they start to try to scale. They start to actually grow to some kind of meaningful size, more than 50 employees is one measure you might use. Then, we fall off. We either sell ourselves to some big US company, or we just kind of bump along at that level.
Only a handful of Canadian companies get to world-class size. Is- is it going to be different this time?
[00:02:30] I- I think the jury’s out on that. And, the startups that we’ve talked to will tell us that one of the big stumbling blocks is, of course, revenue. And, the reticence of large Canadian companies to buy from Canadian startups. Large Canadian companies tend to prefer proven solutions. Solutions that have already made a mark in the US, or already made a mark in Europe. And, so, some of the startups CEO’s we’ve spoken to have said they developed their product here, they would never want to live anywhere else. But, when it comes to selling their product, they need to look across borders.
[00:03:00] And, then, in this very Canadian fashion, it’s suddenly validated because Americans buy it so we’ll buy it, too. The thing that makes that even a bigger problem is that we don’t have a giant market here. So, if our companies aren’t buying, it really does force Canadian startups to look elsewhere. To take their products- and often that does mean their talent- somewhere else.
Jodie Wallis: And, I’m not saying that Canadian startups shouldn’t be selling their products elsewhere. I’m just saying that, perhaps, there is a role for corporate Canada to play in making it a little bit easier- and then, the government as well, by the way, who’s a huge procurer of software- make it a little bit easier for Canadian startups to generate revenues here, and give them another reason to stay.
Amanda Lang: One thing we do know, though, is when there is as hot a new technology as this, and we have the lead on research, you’re going to get a bunch of startups that are Canadian.
Jodie Wallis: Meet Mohamed Musbah, co-founder and vice-president at Maluuba. A Canadian startup focusing on one of the core applications of AI, language. Basically, how we can talk to a machine and have it understand us. “Kudos to Canada,” he says, “for our academic work in the field. But, the real challenge is building a business.”
Musbah: So, it’s great that we’re doing awesome work for research, and we’re on the cutting edge and the state of the art research is being- is happening here in Montreal and Canada, which is- which has been incredible. But, the next step to this is, “How do we leverage that and take that and apply that into the industry?” I think that’s an area where we could be doing better, and I think a lot of people recognize that, and we’re trying to focus on that, and trying to solve problems in that area.
[00:05:00] We started in Waterloo in 2011. The core founding team and myself are from the University of Waterloo. That’s where we all met. We were, essentially, studying, um, you know, computer science and machine learning at the University of Waterloo. Um, and, really the essence of what we wanted to do at Maluuba was- was simple and it was ambitious, but really simple. We wanted to teach machines, um, to give machines the ability to speak and understand humans from a language perspective. And, the reason why that was really exciting for us is because, if you teach a machine to understand language, um, you’ve, basically, taught it to think. You’ve given it the ability to solve intelligence, which is absolutely incredible.
Like, we believe that language understanding, for example, is at the- is at the heart of solving intelligence. Um, so, it’s been a really, really exciting endeavor for us. And, the first way that we started focusing on this was by building voice assistance. A- at that point in time, back in 2010 and 2011, um, they most, um, I’d say, the advent of the space has been in the context of personal assistance. And, that was when Siri was first purchased by Apple, for example.
So, we set out to build out these voice assistants. And, in the process of doing so, um, we launched our own technology and we had hundreds of thousands of users using it at that point in time, which was really exciting. But, we also started understanding the limitations of language understanding in this space. And, that’s when we decided to g- delve deeper into solving more fundamental problems that relate to language understanding. An example of that is comprehension. Can we teach a machine to understand unstructured texts or, like, long passages of text?
[00:06:00] Can we teach a machine to read a book, for example?
If we’re able to do something like that, then that’s actually an incredible breakthrough. So, we got really excited about going into further depths of teaching machines to understand language. And that’s really where our transition from Waterloo to Montreal came to be. Um, we had- we started conversations with professor Yoshua Bengio, who’s known as one of the pioneers of deep learning in artificial intelligence. Um, you know, after multiple conversations with him, he agreed to become an advisor to our team.
[00:06:30] Um, we decided, at that point, that it made sense for us to invest heavily into developing research. And that’s where we decided to setup our first research lab, which was based in Montreal. Uh, this was in 2015.
Amanda Lang: And, from that purch, you made this agreement with Microsoft. Why was that the right thing for Maluuba?
[00:07:00] Well, if you think about Microsoft and the different products that Microsoft has, I mean, text is really a cornerstone of a lot of Microsoft’s products. Across the board. Whether you look at, you know, the Office suite or Skype, or Bing, for example. Um, it’s- it’s incredible how- how much, you know, work that Microsoft does in this space. And, for us, the ability to work with, you know, a large organization that has deployed products, you know, to, uh, hundreds of millions of users, if not more, across these different realms was a really, really exciting avenue for us to take our research and apply it into these different realms.
[00:07:30] That was one aspect of why we decided to go forward with the Microsoft. The other aspect was the fact that they really believed in the vision of what we’re trying to do. When we first spoke to them and we had conversations with them, at our early stages it wasn’t really about, you know, being acquired by Microsoft. It was about them working hand-in-hand to support our endeavor. The more we had these conversations, the more we understood that Microsoft has an incredible vision for artificial intelligence. A lot of what Microsoft tries to do is build AI systems that can work hand-in-hand, uh, with human beings so that we can be more productive as a society.
We have become Microsoft’s first, uh, Microsoft research lab based in Canada with a focus on artificial intelligence.
Amanda Lang: So, it’s obvious there are a lot of pieces you need to get right to get commercial success. One thing that feels different about Maluuba’s story is that big US tech company chose to move to them, and not force them to the big US tech company. And that’s new.
And then committed to growing the Montreal lab over the next two years, which is also new. When we think about commercialization, or becoming commercially successful, it requires a lot more than a great idea and a great piece of technology.
[00:08:30] It requires commercial talent. And, what do I mean by commercial talent? It requires the sales acumen, and the executive connections, in addition to the technology prowess. It also requires, of course, entrepreneurship, it requires investment, and, ultimately, global scale.
[00:09:00] One of the big questions for companies that are looking to attract capital is, what kind of investors you’re attracting. And, for investors, are you looking for businesses that are solving short-term problems, kind of the more immediate AI applications? Or, are you investing in the long game, that general intelligence, the more human kind of AI that could be a long way out?
Jodie Wallis: And, I think the other question, the other dimension to it, is are you investing in AI technologies that solve a specific business problem for a specific business industry? Or, are you looking for AI technologies that can be applicable across industries and across functions?
[00:09:30] Here’s Peter Misek. He’s a partner at the Business Development Bank of Canada’s IT Venture Fund. He say investors, as always, are looking for home runs. But, the holy grail of AI may just be too risky for some.
[00:10:00] So, one of the things that we struggle with, and, uh, you know, I believe there’s a good chance we’ll see in my lifetime, and, um, certainly in my kid’s lifetime is artificial general intelligence. It is literally a self-thinking, um, conscious machine. Uh, that in an investment world is epically challenging. It will require more capital than I have, and I’ll probably ever have access to. And more time than I will ever have and my investors will tolerate.
[00:10:30] So, we, generally, stay away from that. What we mean by verticalized AI’s, where you have really deep vertical data that’s very specific and large and, uh, it doesn’t necessarily have to be structured, but it has to be able to be crawl-able. And, by that, I mean an algorithm has to be able to go through it and understand the uniqueness of each piece of data. After that we think the algorithms in the world have now been very much open-sourced, so the innovation on the algorithmic side is, probably, going to slow. Uh, where we see the innovation is, uh, on the last part. The reward function. And what that is, is imagine you wanna move a cup from one spot to another. Well, the reward is you’ve moved it.
[00:11:00] Well, in the real world, when you design reward functions, like a cars thing on the road, uh, it’s not hitting anything. It’s staying within the lines, it’s obeying speed limits, et cetera. And, environmental factors start to impact it. Human behavior, other cars, and it becomes exponentially difficult, and what you want to see is, you want to see hundreds, if not thousands, of companies created. Because the survival rate is so low.
[00:11:30] So, once we have company creation, and let’s say we get to the point where we’re really blessed and we have 1,000 companies in AI that we make every year; roughly 100 will survive over time. And, by time, I mean three to five years. After that they start entering into my world, and in our world what we’re looking for is companies
that are generating revenue, and we want to help them scale. So, we’re looking at the late series, A series B. So to demystify that a little bit, companies that are doing roughly three to five million of revenue, growing at more than 100% year over here, have product market fit, a technology that works, and, uh, we help them scale.
[00:12:00] We help them get talent, we help them with their product roadmaps, uh, we help them with revenues, business development, you name it, across a suite of products that we offer. Uh, and then, we work with them to get them to, say, $100 million of revenue, and at that point we give them a bunch of options.
Amanda Lang: So, one of the challenges that businesses have in this country is the scaling part. What do we need to be doing differently to help improve that? ‘Cause that’s one place we really do need to move the dial.
[00:12:30] I would tell you that scaling is significant challenge for the ecosystem. Uh, and, it’s primarily because, um, very few people have been there, done that. So, imagine you’re an entrepreneur, you’ve built your company $10 million revenue. Everyone’s patting you on the back. Everyone’s saying what a great champion you are, how awesome you are. And you’re feeling pretty good about yourself. But, the challenge is going from $10 to $100 million, is actually harder.
[00:13:00] It actually is harder, and, knowing what to do, knowing even when to start is so daunting. And, you’ll also find that the number of people on the team that ever done anything like this is right around zero. And, so, what we try and do is, um, we bring in management that we go out and we recruit. Uh, and, you know, frankly, I’ve had the privilege of working with companies that have scaled, uh, beyond that. And walk them through primarily what not to do. And, primarily where to start.
Jodie Wallis: So, what advice would you give to young AI startups as they’re looking to develop new solutions?
[00:13:30] Uh, some similar advice and some different advice that I’d give to a regular startup. So, the start- the advice I give to every startup is the best capital is revenue. Uh, it is no other panacea works as wonders as revenue. If you focus on revenue, all other problems tend to go away. Um, that’s the first of advice. The second piece of advice is don’t try and boil the ocean. Solve a very particular problem, and solve it in a way that has exponential impact. Um, if you’re going to make a 10% improvement, or a 15% improvement, and you go for the game, “Oh, it’s a trillion dollar market, I’m going to get 1% sure, look how big I’m going to be,” that, really, ultimately, kills ya.
[00:14:00] What you really should be looking at … Find a market that’s, I don’t know, a few billion dollars. But, you’re going to make a 99.999% improvement. Or, you’re going to increase revenues or growth by thousands of percent. Don’t think 100%. Don’t think 5%, 10%. Think in exponential terms, non-linear terms.
Jodie Wallis: So, we’ve heard from some startups who are looking to scale that it’s very difficult to scale revenue in Canada, and that they’re finding much more receptivity to their
innovative products outside of Canada. How do you see that issue?
Peter Misek: I totally agree with it. And, I would argue that that makes perfect sense to me. Canada is- as a wonderful place as it is, is a tiny market on the global scale. Uh, companies scaling to $100 million revenue in Canada would be virtually impossible. Uh, almost all of our companies generate the majority of their revenue outside of Canada, and we don’t think that’s a bad thing. We think that’s a great thing.
[00:15:00] So, there’s some great advice there for startups. I love the idea that you shouldn’t boil the ocean. Uh, and that you should try to dominate a space that you can dominate. N- don’t try to do everything. Jodie, you talk to big companies all the time, and big companies acquire startups. What do you tell them they should be looking for?
Jodie Wallis: Fundamentally, two things. So, the first is that the startup has developed an applied solution. A solution that actually addresses a specific business problem in the real world. And, then, second, that that solution can be extended to tackle other problems.
Amanda Lang: One of the things businesses want to do is get their product to as broad an audience as possible. And, somebody who thinks about that a lot is John Ruffolo, who’s CEO of the big pension fund, Owner’s Venture Arm. But, he says, when we go to the global market, we need our eyes open about some of our weaknesses. I talk to him about the fact that we have this lead in intellectual property in ideas in AI and that that gives us an advantage. He was not so sure.
[00:16:00] So one of the things that we are, I think, justifiably proud of in Canada is that we have the early lead on IP when it comes to parts of the artificial intelligence landscape. And, we have a supportive environment. We’ve got government funding, we’ve got hubs that have sprung up that you’ve been part of, uh, a couple of them. Our we- do we have an advantage here, or is this a kind of a Canadian patting themselves on the back story?
[00:16:30] Well, you just said two different things. We have no advantage of the IP, we gave it all away for nothing. Right? So, we have the greatest grandfather in Canada, Jeff Hinton, and all of it is not owned in Canada. But, thank you very much for the Canadian taxpayer who fou- who funded a lot of this stuff.
Amanda Lang: Who owns it?
John Ruffolo: Uh-
Amanda Lang: Who did we give it to?
John Ruffolo: It- it- it- it- it rhymes with “loogle.” (laughs)
Amanda Lang: (laughs) Fair enough.
[00:17:00] But, no private enterprise, or public enterprise, in Canada understood how to monetize it. What I’m particularly thrilled about is that is increasing the general level of education in that space, which, again, kind of falls more from a gut- into a government policy perspective. And this is where the government’s really did a phenomenal job coming together. So, I give an A-plus on that.
[00:17:30] The second part is, okay, now that we’re using taxpayer dollars to fund all this thing here, let’s not give it away again. So, everything here, anything that’s developed in here, needs to remain, uh, owned by Canadian entities so that we can exploit the IP. We could certainly license it, and share it with the world, but in exchange for a fee. As opposed to continuing to give it away, again, and that has been the Canadian history of giving it away.
[00:18:00] We should be welcoming any country, or any company in any country, that wants to come here and utilize our great resource, will have to pay for it like every other Canadian company does. But, knock it off with giving them incentives on top of that. Uh, the value that they’re getting is, they’re getting the best people in the world, and that’s good enough. And, we need to be proud about that and stop being insecure Canadians that think that we actually have to give, you know, the whole- the house and the kitchen sink, uh, on top of that. So, I’d say that’s fine.
[00:18:30] Uh, where we develop it and use taxpayer’s money, and then just, in essence, give it away, uh, to a- and not for the highest bidder, for no bid. That is- th- th- that is, uh, unacceptable and needs to, absolutely, stop.
Jodie Wallis: So, which of the Canadian organizations that are stepping up to be the ones to take advantage of the research?
[00:19:30] That is a great question. So, you’re getting a lot of the startups who, at the end of the day, when these individuals are starting to get trained through the AI folks, it is a lot more manpower for our startups. And, the access to talent is a massive issue. The second part is the corporates. The corporates, who right now probably have the greatest need for the AI talent, and I’ll use the banks as an example. Um, if they’re helping and they’re employees of the banks, just by v- by that nature, the IP is remaining, uh, in this country. It’s being monetized by a Canadian institution. And, that’s absolutely fine.
[00:20:00] What I see right now, though … So, it’s still early days, because they still have to, you know, run through the programs, the Canadian corporate companies, where they really are, is looking at AI on the cost side of the equation. How do they reduce the cost? They still are not really at, “How do I gain new insights for new revenue models, and new types of goods or services that no- no- no one else has?” We really quite haven’t hit that level, yet.
Jodie Wallis: Where do you think the greatest value is going to come from, in terms of startups? Um, are there companies out there that you think are really leading the charge?
[00:20:30] Well, the one thing that eh- it’s getting very interesting, and if you kind of look at our portfolio, and certainly, it’s been more pronounced over the last 18 months … Um, I have zero interest if a pizza could get delivered to your house 30 seconds faster. And, th- it was starting to get ridiculous that we were getting involved in incrementalism on things that just didn’t matter.
[00:21:00] When you look at the data, Canada does not have a shortage of startup. Uh, and- and let’s make it easy. Let’s define a startup as zero to $10 million dollars in revenue as a proxy. Where the problem hits is the $10 to $100 million dollars. And why do I say $100 million? Because that’s usually the point where you actually, seriously, think about going public. And now you start to have the separation of management and shareholders. So, that’s kind of a different massive phase of- of formation.
[00:21:30] That 10 to 100- when you really look to see what are the actual problems in there, it’s not the access to capital. They’ve already been funded. And, it’s not perfect by any stretch. But, just as you’ve said, over the last seven years we’ve had more capital deployed ever, and that includes the dot- whole, dot com, uh, uh, time period.
[00:22:00] So, the two areas are the access to talent and access to customers. The talent we’ve talked about, but the access to customers … It is frustrating where a number of these companies, when they’re really trying to scale, and a lot of them are selling into the public sector. I’m using this as one example. Uh, when they sell to US public sector first before Canada, you’re really are shaking your head and saying, “What’s going on?” And, you start to look at our procurement policies.
[00:22:30] Uh, health tech … Great example, uh, Province of Ontario, $54 billion dollars of procurement. How much goes to startups? As far as I know, if it’s not zero, it’s pretty close to zero. That’s just- that’s just ridiculous. The money is being spent.
Amanda Lang: This next step you’re talking about. It is a role for government when we talk about education, when we talk about creating, sort of, the pipeline for expertise. When you’re putting your money to work, how do you keep wanting to keep it here if that doesn’t happen?
[00:23:00] It’s a great question. So, let me give you an analogy, and I’ve used this before. Uh- uh, I use the analogy of a farmer’s field. The role of government is to clear out the weeds, lay the fertilizer, make sure the ground, uh, is ready so it could support the growth. The entrepreneur is the farmer. The farmer chooses the seeds, and is the one that manages the growth of that plant. And, the VC’s and lawyers and accountants, et cetera, that’s the sun and the rain, just to make it go faster.
And, our role is simply to help it grow faster. And the role of the government is not to select the seeds or make it go faster. E- it’s too hard, but if they’re not there plowing the field, then, there is no farmers planting the seeds. And, so, um, you know, human capital, immigration, standards, the area of the role of the
[00:24:00] government, in my view, in the- to me, the way I think about it is, if a private company can’t really do it, then it’s the hint that it’s government.
[00:24:30] So, when it’s- when it’s- when it comes to the availability of capital, private companies should do it. Maybe the government could play a small role in giving it a- a kickstart. But they should get out of the way on that. But, when it comes to standards, or comes to immigration, no individual company can actually do that. Hence, that’s where the roles of where government should play. Every single country in the world that I visit, all say the same thing. Innovation is the future of that country. So, we are in the dogfight of our lives right now.
[00:25:00] So, Minister Bains says we are working on an IP strategy for Canada, which is good news to a lot of people who’ve been pushing for that for years. Uh, intellectual property protection, it can be all the difference between success and failure for a business. My concern, Jodie, is what does it look like? How far away is it? Uh, the best intentioned of programs can take a long time to come to market, and we don’t really have a long time.
[00:25:30] Yeah, I’m not sure where IP strategy really belongs in the list of priorities. And, I’ll tell you why I say that. I think there’s really two ways to differentiate in the AI space. The first is in the way that AI is embedded into core business. It’s easy to take a piece of AI technology, or some AI software, deploy it on the fringes of the business. Show what interesting insights it can produce. It’s another thing, entirely, to embed it in the core operations of a business, and have an impact in the market.
[00:26:00] The second thing that we need to differentiate on is the data. So, it may be true that we all have access to the same algorithms because our academic institutions are publishing them, but if we can A- make sure we’re truly integrating into the core business, and B- we have enough data and great quality data to fuel our AI. That’s where we’re going to differentiate.
[00:26:30] Some of the greatest entrepreneurs in this country, including Jim Balsillie who helped build Research in Motion, Blackberry, into a global giant would say IP is critical. Getting that piece is critical. To your point, though, there are other components here that, if we do not get right, we fail.
One of the questions, I think, Canadians want to know is, where is the next Blackberry? Where is the next Nortel? We don’t- we don’t build many giants like that, but we should be able to build, you know, say once every 10 years. Are they coming?
[00:27:00] I think we should anticipate that they’re coming. I think we should anticipate that the ideas that we continue to generate in this country are going to lead to great startups, which are going to scale, and are going to lead to great, large companies. I think the question on the minds of many Canadians is, ” Then what? Where does it go from there?”
Amanda Lang: So, have you ever had a computer ad kind of strangely anticipate your needs? What
[00:27:30] you were looking for before you even thought you wanted it? That’s predictive AI and it’s based on the data that Jodie’s talking about. It’s a hot field for marketers, as you can imagine, but it’s also full of ethical minefields. Have no fear, we’re here to guide through. Join us for the next episode. We’re going to ask the question, “How much do you really want to share with the all-seeing mind of AI?”
I’m Amanda Lang.
Jodie Wallis: And, I’m Jodie Wallis.
Amanda Lang: The AI Effect is produced by Antica Productions, and hosted by Amanda Lang and Jodie Wallis.
Jodie Wallis: This podcast is sponsored by Accenture.
Amanda Lang: Our producers are Paula [phonetic 00:27:47] Fla-lo, Dela Valasquez, and Annalisa Nielson.
Jodie Wallis: Our executive producer is Stewart Cox.
Amanda Lang: Music for this podcast, by Poddington Bear.
Jodie Wallis: Subscribe to the AI Effect on Apple Podcasts, Google Play, Stitcher, or wherever you get your podcasts.
Amanda Lang: Visit our website, the effect dot a-i.
Jodie Wallis: And follow us on Twitter, @ A-I Effect.
Amanda Lang: Thanks for listening.