In a world where self-driving cars can move across Europe by themselves and claims staff are being laid off at an insurance company because of an effective algorithm, how can SaaS companies adapt to this new reality? In this talk, Ludo Ulrich, Head of Startup Relations at Salesforce for Startups, and Tomasz Tunguz, Partner at Redpoint Ventures, take the stage to discuss how machine learning will transform SaaS.

Tom also asks a poignant question: how do you get people to interact with AI without being afraid of it? While Hollywood has done a good job of instilling fear into our minds about robots taking over, he believes that shouldn’t be the case. In a year from now, we’ll definitely be talking about machine learning and the companies that have started figuring out how to use it to scale.

And yes, it is possible to pitch your startup without saying “machine learning”. In fact, doing so shows that you’re interested in the value proposition for the buyer rather than just the technology itself.

If you want to know where machine learning and AI can take us, this session is a must-see.

You can see the slide deck here.

And if you haven’t heard: SaaStr Annual will be back in 2018, bigger and better than ever! Join 20,000 fellow founders, investors and execs for 3 days of unparalleled networking and epic learnings from SaaS legends like Jyoti Bansal, Aaron Levie, Josh James, and Dustin Moskovitz. If you don’t have tickets, lock in Early Bird pricing today and bring your team from just $499! Get tickets here.

 

Transcript

David Appel:  Good morning. Hope everybody enjoyed that first strategy stage. My name is David Appel. I am the Head of the Software and SaaS Vertical at Intacct, the world’s largest, independent, pure cloud ERP provider. We’re thrilled to be sponsors for the tactical stage today. We got a great first session.

It’s going to be on “AI, The Next Platform for SaaS”, it’ll be moderated by Ludo Ulrich and Tomasz Tunguz. Ludo is the Head of Startup Relations at Salesforce.com. Tomasz is a very prominent venture capitalist at Redpoint Capital.

There’s three things we hope you guys learn today. Machine learning is real. It’s important and also practical ideas that you can take on in order to apply it to your business. As the three of us were prepping, it dawned on me this old adage, “Tomorrow’s crazy idea becomes today’s dogma, which turns into tomorrow’s kitsch.” We’re right on the cusp of this, all of us on transitioning from crazy idea to dogma.

To put in the plug for both guys, Tomasz is looking for fantastic firms to invest in. Ludo and team are looking for great firms that want to build and accelerate their development with Salesforce.com. Without further ado, Tomasz and Ludo.

[applause]

Tomasz Tunguz:  Hi. Thank you so much, Dave.

Audience Member:  Woo!

[laughter]

Tomasz:  That takes a nerve. They adjust the nerves, thanks. Hi, everybody, welcome to AIs, New Platform for SaaS. It’s a pleasure to be here. My name is Tom Tunguz. I’m a venture capitalist at Redpoint.

Redpoint is a venture capital firm based in Menlo Park. We invest in seed, early, and grow stage companies. We funded about 434 companies. We’re about $4 billion in assets under management. We’ve been lucky to have 136 IPO and M&A, the most recent of which you heard from earlier this morning is Jeff Lawson.

Let’s see. Here we go. These are some of the companies we’ve been lucky enough to partner with in the, as a service space.

Today’s talk is on a topic that I’m really passionate about. When I was in college, in grad school, I studied machine learning. When I was at Google, we also did a lot of machine learning. The world of machine learning has changed tremendously in 2016. Let me tell you about why I think it’s going to be so impactful for SaaS.

In 2016, it was hard to avoid machine learning. We saw Lee Sedol, the best Go player in the world, be defeated by AlphaGo, a Google algorithm. We saw the advent of the first self driving car, and then we also saw more than 10 million Amazon Alexas put into consumers’ homes. That’s just in the consumer world.

The challenge with machine learning in 2016 is that every startup that comes in to pitch at Redpoint now says, “Hey, we’re an AI company,” “Hey, we’re an ML company.” What does it really mean to be an AI or an ML company? I’m only going to use ML because I don’t really believe in the term AI.

Machine learning is really simple. All it does is it teaches a machine to find patterns in data. There are four things that you can teach a machine to do. If you used Waze today, you ran through an optimization algorithm. “What’s the fastest way to get to the Pearl?” which is a nightclub. [laughs]

The second thing you can do is you can do object identification. If you take a photograph of a cat with your telephone, how quickly can the computer tell you whether or not it’s a cat?

The third way you can use machine learning is anomaly detection. This is most prevalent in security, and also in antifraud cases. I have a $10,000 charge on my credit card for buying a piano yesterday. I don’t play the piano. It’s clearly a fraudulent charge.

The fourth thing that you can do with machine learning is segmentation. If you’re a very avid user on FarmVille, you might segment the user basin, decide to treat the whales differently. You could do the same thing for a SaaS company.

Given that all these technologies are more than 30 or 40 years old, why is there so much recent buzz about machine learning? What’s happening is the convergence of three different major trends that have been in the works for those three decades: really cheap compute, more data storage than we’ve ever seen before, and advances in algorithms.

The key trend is something that everybody is calling deep learning. Neural nets, I studied them in college. They’ve been around since the late 1970s. Neural nets do something that we’ve never been able to do before, and they’re a consequence of that convergence that I just talked about.

To break down machine learning really simply, there are two different steps inside of it. The first is feature selection, and then the second is model tuning.

When I was at Google, the data engineers working on machine learning algorithms, they were doing the first part by hand, the feature selection. That’s picking the columns in the Excel spreadsheet that you are going to use in order to create an equation to predict something, optimize something, segment something, whatever it is.

The second part is the model tuning. That’s taking the huge amounts of data, putting it through the algorithm, and figuring out which of the algorithms that you’re trying are the most effective at getting to the result that you want.

Feature selection for this cat, that might be the angles of the ears, or the shapes of the eyes, that kind of thing.

The big advance in deep learning, basically, is that deep learning automates both processes. Now, you can do everything in a computer. The results that have come out as a consequence of these advances are fundamental. Let me walk you through three.

Google has a product called WaveNet. It’s a computer that speaks so well that no human can tell it’s a computer. Microsoft released some research in 2016 that a computer can understand human speech as well as another human. Google released a machine translation algorithm that will translate from English to any other language, even if it’s never seen that language before.

Those are fundamental machine learning advances. This is not the kind of simple linear regression stuff that we’ve been doing for decades. This kind and this scale of innovation is really going to impact the startup world. Let me walk you through a few ways that we’re thinking about it.

The first, we were lucky enough to partner and announce an investment in a company called Chorus yesterday, which is leveraging a lot of these speech recognition and natural language processing advantages to analyze conversations for salespeople and help them sell more effectively.

In the insurance world, we saw a Japanese company lay off 30 percent of its claims staff because it’s being replaced by an algorithm. There’s a British company that allows you to take a photograph of the accident that you just had in your car. The machine learning algorithm will tell you exactly how much damage you’ve done to the car and the insurance company will pay you that instant.

In the world of logistics, everybody knows this example. There’s a truck that drove across Europe entirely by itself. We’re seeing self driving Ubers. In the world of construction, there’s a robot that lays bricks three times faster than a human.

In the world of medicine, algorithms are looking at radiological scans and detecting Parkinson’s and cancer at far greater accuracies than a human ever could. In the world of agriculture, a young farmer’s son is using tensor flow in order to mechanize his dad’s cucumber farm and categorize them for less than $1,000.

The big thing about machine learning is it’s going to change the world of SaaS because all of a sudden, the machine learning innovations will happen in the core categories that we’ve seen in the past, those horizontal companies selling to customer support, customer service, sales, product.

It’s also going to broaden out SaaS. We’re going to see an expansion in the world of vertical SaaS. These are just some of the examples.

All of a sudden, all these offline industries, their key processes are going to be revolutionized. That’s a huge potential for innovation, the scale of which is similar to the mobile phone and cloud. That’s how big of a wave we at Redpoint think it’s going to be.

In order to prosecute the strategy, we have five different precepts about the kinds of companies we want to invest in. The first is we’re looking for companies with proprietary access to data.

The algorithms themselves are not that innovative, they’re not that closed source. The company that’s going to really innovate is going to need its own data set to be able to train its own models.

The second is we are not interested in platforms. Google, Microsoft, Amazon, they will be releasing the APIs that lots of companies will be building on. You may want to compete with them, but we will likely not invest in that because it’s a very challenging go to market strategy.

What we’re more interested in are machine learning applications that solve an end to end problem where a customer can say, “This is generating me more revenue. This is reducing my cost.”

The third is a machine learning innovation in and of itself is not enough to generate substantial customer demand. It has to be able to change the go to market. The trope that we use inside of Redpoint is we’re looking for a fundamental technology innovation that enables a go to market advantage.

The fourth is we’re looking for experts in the field. You can take a generic NLP or speech learning model and you can get 80 percent of the way there. In order to really have a fundamentally new experience, you need to get it 95 percent of the way there.

You can’t do that with just off the shelf algorithms. You’re going to need special talent in order to deliver that magical experience. The fifth is we will occasionally invest in potential algorithmic advances that are closed source.

For us, machine learning isn’t new. We’ve been investing in lots of machine learning startups, startups that are not necessarily exclusively focused on machine learning, but use machine learning in order to create a competitive advantage in the market. Because these advances in machine learning are so…they’re of such great magnitude, we are going to continue to invest in more.

That’s an overview of the way that we think about machine learning. Now, we’ll have a conversation, Ludo and I, about how we’re seeing machine learning.

Ludo, you work with a startup incubator at Salesforce. What kinds of trends are you seeing in the world of machine learning?

Ludo Ulrich:  Thanks, Tomasz. By the way, amazing presentation. Congratulations, of course. I don’t know about you, but I think there’s nothing like human being intelligence, versus artificial, in this particular case.

To take a step back, indeed, in my team, we’re engaged with a lot of startups. It’s separate from the M&A efforts. I’ve seen, like you, it started there. We invested as a company, as Salesforce, in many, many companies and acquired a lot of companies.

Companies, like RelateIQ, with a product, as well as technologies, who have more, almost, lab technology, like MetaMind, like PredictionIO, like BeyondCore, Implicit. We’ve been really active. My colleagues have been very active on this front.

On the ecosystem development side, with activities like the Incubator, Salesforce startup, all those activities, we really agree with what you put on the slide. It’s all about the go to markets, and basically revolutionizing the scenarios and the use case for customers.

We’re on the tactical stage. It’s basically trying to make something improve the productivity. A lot of people talk about AI as the new UI, so how we make that experience better.

We just reviewed the next batch. We’ll announce very quickly the next batch of the incubator focused on AI.

I’ve seen a lot of vertical. I’ve seen a lot of new, new businesses, like identifying unconscious bias within a company, optimizing the composition of a team based on AI, all those kinds of things that we haven’t seen before. I think you have clearly those trends.

We’re trying to help them go to market, access to the Salesforce customers. We are eager to get more AI, basically.

Tomasz:  That makes sense.

Ludo:  Tomasz, I want to ask you, because one of the things we see as well beyond the trends, as we engage with startups, specifically in my team, more generally Salesforce see a lot of demand from our customers. The whole industry is taking about AI.

You probably came to that session saying, “Wow, OK, let’s try to go beyond the buzz words and beyond all that stuff,” but it’s real. Beyond the end users, the consumers who are using Facebook, Uber, and all those AI powered applications every day.

Beyond the VC community wants AI. We heard that loud and clear. I said your investment is clear. The C suite, the CXOs are there. Accenture is a evangelizing all their customers about the era of a new enterprise, intelligent enterprise.

Salesforce have been going big in the last year with the brand Einstein, which is basically the umbrella brand under everything that we do AI wise, to be able to tell our customers that they should be more demanding with us, and get all those features.

How do you see some of your portfolio companies play with that phenomenon?

Tomasz:  The most important thing…maybe I’ll tell it in the form of a story. I’ve loved speech recognition for a long time. About three years ago, I bought Dragon’s Nuance and I started dictating all my emails. I started noticing that I was two to three times faster dictating than I was typing.

Then, I saw this movie “Her.” That’s a brilliant movie, not for the romance. It’s for the human-computer interaction.

You watch the way that the main character dictates a letter, says print, the thing prints, or the way he interacts with his email, where he says, “Next email. Next email,” he replies. He’s only really looking at the telephone when he wants to see an image.

That’s a great analogy for how machine learning succeeds. What I mean by that is it’s not in your face. It’s hidden in the background. It’s doing the work behind the scenes.

That passion for voice recognition technology was what led us to spend a year looking at different companies taking advantage of voice recognition and natural language processing. That’s ultimately why we invested in Chorus, because they can analyze conversations in real time and help salespeople sell better.

That’s why we’re so excited. For me, that’s a great example of how, because it’s a seamless experience, you just dial into the call, and then all of a sudden you get value, you don’t see the machine learning. You don’t feel the machine learning.

It doesn’t feel like a different product, but what’s happening is the product is giving you that magical moment. It’s telling you, “Here’s the battle card,” “Here’s the case study,” “Here’s how I handled that objection.” “You should be talking slower,” “You should be talking less.”

That’s what we’re looking for when we’re looking for companies. We’re not looking for companies with people who are beating you over the head.

In fact, one of the things that we were talking about in getting ready for this was a really great point that you made, Ludo, which is, can you pitch your startup without saying machine learning?

What that means is you’re not focused on the technology. You’re focused on the value proposition. You’re focused on why the customer’s going to care about it, why the buyer’s going to be promoted. That’s the key. Yes, the technology’s there, and yes, it’s going to be a fundamental transformation, but you can’t lose sight of the fact that you have to deliver ultimate value to the buyer.

Ludo:  Indeed, that’s actually a good tip. I really recommend. Again, it’s a tactical stage. Ask yourself if you can do that without saying AI and surfing that wave.

One of the company, for instance, Salesforce Incubators called Quarrio. They have a massive team around the world. They’re highly distributed, and they invest in a lot of the technologies, ML, and everything that you discussed. They also have this very approachable voice activated interface.

Again, coming back to AI being the new UI, how do you help coach the companies in your portfolio, the ones you talk to, to basically keep it super simple, approachable, etc. in the way they build their pitch. Whether they’re in a portfolio or they’re pitching to you, what could be a tip that you give them, outside of the ones that…?

Tomasz:  We have this fear that’s created by Hollywood about AI. We have an annual investment meeting every year, and one of our partners is Andy Rubin. Andy Rubin created Android. He brought one of the Boston Dynamic dog robots. Have you guys seen this thing? It runs 40 miles an hour. It’s exactly what you’d see in a post apocalyptic film.

[laughter]

Tomasz:  There’s this video where they’re testing it. There’s a human that’s hitting it with a hockey stick as hard as he can. He starts kicking it and hip checking it.

Andy gets up, and he says, “This is why they will rise up. They never forget.” It’s the whole “Westworld” thing.

Ludo:  Good storytelling.

Tomasz:  [laughs] That’s a completely irrational fear. Whenever anybody interacts with AI, they have that fear somewhere inside of them.

When I was interviewing PMs at Google, we would constantly see people who had studied human computer interaction. Now, there’s a new field. It’s no longer human computer interaction. It’s human robot interaction.

How do you get a machine learning algorithm to interact with somebody so that they’re not afraid of it?

Ludo:  Not being afraid of it. It’s funny you say that, because a lot of the companies I interact with, another one is Nova.ai. It’s as simple as they help you draft a sales email. The clear impact is that you have five X the open rate. It’s pretty down to earth. It’s more like productivity, doing simple things.

Do you have example of that? Again, to try to dumb it down, and try to say, “OK, it makes it approachable. It’s not a robot. It’s actually just a core component of what you’re doing.”

Tomasz:  The key thing in human robotic interaction is you have to set the expectation. You can’t say, “Hey, this is a chatbot that is going to be able to answer any question that you could possibly imagine in an intelligent way.”

If you do that, what’s going to end up happening is the user is going to use it once, figure out the limitations and then say, “Hey, this doesn’t work with my mental model of the world or how a robot should work.” They’ll stop using it.

This is particularly relevant in the chatbot world. The reality is that there’s a limited syntax that you can use to interact with the chatbot for it to be effective and you have to train people. People are becoming more sophisticated. We’re all getting to the point where we use Google more effectively. We’re more comfortable with command line interfaces. We’re using text message on a daily basis.

When it really comes down in order to get people comfortable with UI, particularly interactive UI, is being able to set the expectation of the user appropriately and that means much lower than you think it typically ought to be.

The second thing that you really don’t want to have happen is if you’re wrong, you lose the trust of the user whether it’s a recommendation about, “Hey, you should bring up UCLA in the sales email, or how we both went to that college, or whether it’s a slightly incorrect answer.” The second somebody if they think it’s a human and then it acts like a robot, it breaks down people…

I was talking to one entrepreneur this morning, he said, “We’re very explicit that this is a robot. These are the limitations and we’re going to escalate to a human when and if necessary.”

We’re starting from a place where whenever you’re using machine learning with humans, you need to build trust. You need to build that relationship just like your or I would. We’re starting to build a rapport and we can’t forget that when we’re building robots.

Ludo:  Yeah, understood. Another thing is we keep hearing about AI from all the big players in the industry. If you go to Facebook, Microsoft, Salesforce, we do a lot of moves. What’s your suggestions to the founders you work with in terms of buying, building, etc., and building teams, because talent is scarce as well? Maybe some tactical, practical advice to the founders here.

Tomasz:  It’s awesome that this mono clouds are building all these AI algorithms. They’re publishing all this research. They are offering them startups for pennies on the dollar. You can start with an NLP algorithm for free. What’s going to end up happening is that layer’s increasingly become commoditized.

For a start up, the way to think about it is what value can I add on top. That’s in the four and five investment parts of our strategy. One is experts in the field. You need to continue to be able to add value on top of the mono cloud platforms in order to have a sustainable competitive advantage.

You have to assume like generic NLP, and generic automated speech recognition is a commodity and everybody has it. The question is, given that, what is it that I’m going to do next?

Ludo:  Specifically around data, AI without data doesn’t exist, so that’s why a lot of companies want to deal with our customer’s data. You mentioned in your Investment Strategy that you care about proprietary data. Can you elaborate a little bit? Again, where do you get that? As a founder where do you get, unless you have that unfair advantage?

Tomasz:  We’ve probably seen two key strategies in creating proprietary data. The first is creating it yourself. You build a workflow application and by virtue of people doing work within this application, they generate a data set that you can then mine that becomes proprietary.

That’s a really good strategy. The challenge in the short term is that data set needs to be really big in order for many of these algorithms to work, so you can’t really offer a whole lot of ML features right out of the gate, you have to be patient.

It does make sure that the core value proposition of the workflow product is on the money. One great example of that’s LinkedIn. That’s an older example, but they had a huge data network effects as a result of all the traffic and all the information that they have on their system, they can continue to build different products on top like Sales Navigator or whatever it is.

The second example that we’re starting to see more and more of is it’s particularly in enterprise. An early stage SaaS company will go to a large like Fortune 100, Fortune 1,000 and say, “Hey, there’s this is really critical problem that you have that you don’t have the expertise to solve but we can solve it. Why don’t you put in a little bit of money, why don’t you give us proprietary access to your internal data and then we will build a system that isn’t necessarily built to suit for you, but will definitely benefit you, but that we can scale across.”

This is happening a lot in the Internet of Things world and the construction world. Caterpillar has made this substantial investment in a Chicago-based company where they’re doing analytics on top of sensors for all these equipment for predictive maintenance. How do you repair a Caterpillar Excavator, that kind of thing.

Or, we’ve seen a recruiting company that partnered with lots of Fortune 500s to get access to their hiring and recruiting and promotion information so that they change the way that the applications or job requisitions are written. If you can start with that kind of corpus out of the gate, then what you can do is you can offer those ML features right out of the start.

Those are the two strategies. I’m sure we will see other strategies. One that we haven’t seen yet is a cabal of startups getting together and sharing information across, but who knows.

Ludo:  Not to get too much into a political thing, but I think AI could also help a lot, like I said, create a composition of a team or identifying unconscious bias, maybe help us do a better immigration, identifying all these things.

Hopefully, AI will help us in many ways. Can you just predict…We’re running out of time. We have this panel in AI today. No startup here is probably defining themselves as a mobile cloud startup, so probably next year, they won’t say mobile cloud AI. What do you think we’re going to talk about to tomorrow or next year?

Tomasz:  A year from now, we’ll still probably talking about machine learning. We’re in a world of…there’re very few SaaS companies that use machine learning at scale.

Maybe next year, we’ll be talking about the first handful of them that are 10 or 15 million in ARR and we’re going to be trying to dissect how it is that they got access to their data set. What was the team composition? What was the kind of expertise they needed?

How did they translate the technology advantage into a go to market advantage? That’s the part…technology advantage into a go to market advantage – it’s still not clear.

Let’s say, you create the next version of Zendesk that’s got a whole bunch of ML. It’s not necessarily that Intel Inside technology is going to, all of a sudden, reduce your cost of customer acquisition relative to an incumbent. Next year, we’ll be talking about those strategies, I think.

Ludo:  Great. Let’s finish with a tip. I guess I stopped to let you think about it. Again, we see all of customers trying to reinvent themselves. As I said, the C suite is being educated by all the key players about the fact that AI is going to transform their business. That’s the new way they need to engage with their own customers with personalization at scale.

This is a great opportunity to come in. It’s always hard for start ups sometimes to scale to large accounts. You can probably build on ecosystem like Salesforce and try to leverage access like marketplaces like the AppExchange, but this is the time trying to go beyond the buzzword. Help those customers reinvent their own experiences.

Again, I’m going to say one more time, AI being the new UI, how can all those big accounts basically have using that technology a very disruptive approach? Really, take your chances, it’s now or never basically with large companies. There is demand and we’re evangelizing that on your behalf.

Tomasz:  In closing, thank you very much to Jason for inviting us. Thank you for coming to the session.

View the slides here.

Related Posts

Pin It on Pinterest

Share This