The following interview was between Omar Tawakol of Voicera and Sean Jacobsohn and Fay Hazaveh of Norwest Venture Partners. They spoke about the development of the AI space and where they see great opportunities emerging. Read on to find out more…!
“Tell us about Norwest and yourselves.”
Norwest Venture Partners, the second oldest venture firm in the country at 56 years old, has 150 active portfolio companies. Over the past three years, we’ve had 35 companies go public, or get acquired.
I’ve been at Norwest the past three years, I’m a partner on the enterprise team. And, most of my career has been in go-to-market executive roles at enterprise cloud companies such as Cornerstone OnDemand, WageWorks, and Upwork.
Thanks, Sean. My name is Fay Hazaveh, I’m an associate on the enterprise team, I work with Sean pretty closely. I’ve been with Norwest for about a year. Before Norwest, I was at Hewlett Packard Enterprise, where I helped launch their strategic venture arm, Hewlett Packard Ventures. Prior to that, I was an investment banking analyst.
“There are lots of companies in the AI space, but can you help us understand the different layers, in terms of how you value the AI space? In particular, how do you separate platform companies from vertical application companies? And what are the most interesting AI startups you’re seeing at some of these layers? And feel free to define layers differently than I did.”
We think about the full AI stack from the technology building blocks up to the applications. To start off, we have a number of what we call Core AI Technologies, such as speech recognition, natural language processing, image recognition, and so on. There are also open-source libraries also offering from companies like Microsoft, Amazon, and Google. Plus there are a number of startups in the space, such as Clarify for image recognition, or Vicarious for Computer Vision.
The next layer that we see are what we call Core AI Platforms. Examples of this would be Amazon Alexa, Microsoft Cortana, IBM’s Watson, and Apple’s Siri.
The next category is what we’re calling Enabling Technologies and Services. This could include anything from conversational management platforms and chatbot frameworks to companies offering “training data as a service.” On the application side, we categorize into both horizontal and vertical AI applications, with the difference being that vertical applications are targeted toward a specific industry, while horizontal applications apply across a variety of industries.
We also see a lot of interesting examples of companies using AI technology to build great Horizontal Applications. Gong, which is one of our NVP portfolio companies, records and transcribes sales calls, and is also used as a sales coaching tool. Another area on the horizontal side, are AI assistants. Examples of this would be X.AI and Clara, that help people with scheduling calls, and meetings, and e-mails. And lastly, another horizontal application that is one of our portfolio companies called Cognitive Scale, which is building cognitive enterprise applications across a variety of industries.
Now, moving over to Vertical AI applications. We see many industries applying AI for interesting use-cases. For example, in the healthcare space, Qventus, a Norwest portfolio company, is leveraging AI to improve hospital operations, and Augmedix is helping automate transcriptions for doctors during their meetings with patients.
In the financial services space, Kasisto is a financial services chatbot. We’ve also seen an interesting range of vertical AI applications spanning everything from hotels and hospitality to industrial manufacturing. So, that’s how we see the AI landscape in a nutshell.
“So, when you bet on an AI company, what attributes are most important in driving the bets for you?”
We look at three different areas. The first would be the team. Is this a team that we can partner with for the life of the company? Initially, you need to have a very strong technical team, and until you get to product-market fit, you don’t need to invest much into sales and marketing. Another area is are you solving an important enough problem for the buyer? Buyers will pay to solve one of their top three problems. The first problem you’re solving doesn’t need to have a big market size, as long as solving that problem earns the right for you to solve other big problems over time. And then is this solution 10x better than the alternative? The most common decision is no decision, keeping the status quo. So, if it is 10x better than the alternatives, people will be more inclined to switch from whatever they’re currently doing.
“What is an example of a success, where a company solved its target problem very well for their buyers?”
Salesforce started in the early days just focusing on one problem —creating a CRM solution focused on SMBs. At the time there were either on-premise solutions, or manual ways of doing CRM through Excel, e-mail, and other systems. They laser-focused on solving this core problem for SMB customers, and from there, they have added other products for sales, marketing, and service. Over time, they also moved from SMB to mid-market, to the enterprise. If they would not have solved that first problem well for a core set of buyers, they would’ve never earned the right to expand.
“The thesis we have is that if we look at the 19th century it ended up creating and destroying jobs at the same time, but in the long-run, they had a net positive effect on the workforce. There are several reasons to think this disruption has some different attributes to it. Back in the 19th century, automation threatened the jobs of the most skilled artisans, because if you were making furniture, you would do it at a high cost, laborious process, but you had a very precious skillset. And then automation came and allowed you to scale a factory that built furniture. Which ended up creating more unskilled labor opportunities, which created a middle class, and brought down a little bit, the high perch of the artisan. What do you think about this disruption? Is it more similar to the past, is it different? What should we be thinking about?”
This is a great topic. As you said Sean, throughout history, periods of rapid technological advancements have stirred up concerns about effects in the workforce. So, the concerns that we’re hearing now, this isn’t the first time people have worried about technology taking over humans. The textile workers protested in the industrial revolution. In the 1930s John Maynard Keynes coined the term “technological unemployment.” In the 1960s, JFK said, “A major challenge is to maintain domestic productivity at a time when automation is replacing men.” This has been happening repeatedly over time, that’s not new. But, so far, to the point you made, and we agree, so far, all these ways of disruption have led to an increase in overall economic productivity. When we think about what may be different this time, there are two things that come to mind.
The first is, in the past, many people whose jobs were displaced by automation could easily switch from one form of unskilled labor to another. But, what we’ll now start to see is that these new jobs created from AI and automation will require new skills and re-training of a large part of the workforce.
The second part, is that this time around, almost every industry is being affected by automation all at once Whereas, in the past, disruptions were more limited to one industry at a time. So, change is happening much more rapidly and much more widespread than ever before. McKinsey actually estimates that up to 50% of people’s job activities can be automated with the technology available today, so, it’s a cost/ reward.
Ultimately, AI will bring about both job enhancement and job replacement. The vision is obviously to automate the routine and the repetitive tasks to elevate humans to a higher value of work. But, a critical factor in managing the effect of AI on the workforce will be re-training and education for the people whose jobs become fully automated.
“When you think about individual laborers who want to produce more can you think of examples where AI helps?”
Three areas come to mind. One is sales, one is healthcare, and one in customer service. With respect to sales, AI can help leverage best practices across different sales reps, for coaching purposes. It can instruct you if you should get off the phone if it’s unlikely the deal will close, and it can give you advice about what to say in real-time during the calls. Another area is healthcare, where you can analyze data in real-time, to help doctors make better decisions. And the other area is customer service, allowing reps to focus on higher value-added tasks and have AI handle the routine, one call close situations.
“And are there areas that it might hurt a labor force?”
A few areas come to mind, where there’s a high level of routine and predictable behavior, where most or all of the jobs could be replaced. One would be in manufacturing, where each robot has been replacing six people, on average. Trucking is an area where there’s a lot of debate, where people can be eliminated. But, that could be 10 to 20 years out. And then even journalism. There’s an opportunity for AI-written news articles.
“What will the future workforce look like, and what will be their opportunities and struggles?”
Right now, over 50% of the workforce are millennials or Gen Z. And, their average tenure is three years at any one job. So, what they’re looking for, in a lot of cases, are mobile-first tools, and they expect the same experience from enterprise apps as they’re experiencing with consumer apps. A lot of these workers have an entrepreneurial mindset and they’re building self-empowered teams. They are more focused on the work responsibility than their titles. Another area that I’m seeing is that there’s less reliance on office space. Collaboration tools are becoming more and more important, and this is really allowing people to work anywhere, and employers to hire someone anywhere.
“As a company at Voicera, we tend to focus more on areas we call augmented intelligence. Areas where every individual will start to be more like ‘Iron Man’. You find yourself having grammar improved, because of tools like Grammarly. You’re in sales, you get coaching … you’re writing an e-mail, you have an opportunity for coaching. You come into the meeting, you can focus on the meeting. All these are exoskeletons and not robots. We think that if you focus more on the creation of augmented intelligence, in the long-run, you help the whole labor displacement issue.”
We agree with the distinction. I think this aligns with the view that we had said earlier. We see this AI that enhances humans and AI that replaces humans. As we touched upon, also, earlier, both of these types of AI, or automation if you want to call it that, will affect our workforce in different ways. Some of these jobs that robots will be taking over, should actually be taken over by robots. Maybe they are hazardous for humans or require a high degree precision. So, I don’t think we make a blanket statement of “All AI should be augmenting humans.” The end vision for both of these types of AI is that the end vision is to elevate humans to higher-value tasks, in one form or another. But it does touch upon the re-training for the people who are getting fully replaced.
“We focused on augmented intelligence, and our initial product Eva, is an intelligent meeting AI who takes notes with action items and decisions. And she’ll help you interface with the enterprise systems, what would you like to see Eva do?”
So, I love the idea of Eva, first of all, and you touched on this earlier, just eliminating the need for having to take meeting notes, allows everyone to be 100% fully engaged in the discussion. Every time you pause to write something down, you’re missing what the person is saying in that moment. Things like highlighting action items and decisions will be really helpful to make sure nothing slips through the cracks.
My request would be for Eva to actively help us be as productive as possible, in the meeting, relative to what we have set out to accomplish. So an example could be, track how we’re doing on time, relative to the meeting agenda. Make sure we don’t spend 90% of the meeting on the first agenda item and then scramble to make follow-up meetings, things like that.
Omar: That’s a great feature request – thank you.