Success Factors for AI Business Models – A Venture Capitalist’s Perspective

Daniel Faggella

Daniel Faggella is Head of Research at Emerj. Called upon by the United Nations, World Bank, INTERPOL, and leading enterprises, Daniel is a globally sought-after expert on the competitive strategy implications of AI for business and government leaders.

Success Factors for AI Business Models - A Venture Capitalist's Perspective

Episode Summary: Saying that your company does artificial intelligence might still have a slightly cool ring to it if you’re talking to one of your peers at a conference, but it doesn’t mean very much to venture capitalists today, who’ve been battered with machine learning and artificial intelligence in every pitch deck they’ve seen for the last three or four years.

I wondered, from a venture capitalist perspective, what makes an AI company’s value proposition actually strong? What is it that makes an AI startup actually seem like a company that maybe could use AI to really win in the market? Not just to be another company that says they’re going to do it or says they are doing it, but where can it actually provide enough of that competitive edge to make a VC want to pull the trigger?

Getting a grasp of the answer to that question seems pretty critical.

This week, we speak with Tim Chang, partner at Mayfield Fund in Menlo Park, California. Chang and I both spoke at the Trans Tech Conference, held every year in Silicon Valley, focused on wellness and health-related technologies.

Chang talks about what it is about an AI company’s pitch, product, and market that actually makes AI an enhancement to the business in a way that’s compelling to someone who wants to invest potentially millions and millions of dollars.

Subscribe to our AI in Industry Podcast with your favorite podcast service:

Guest: Tim Chang, Partner – Mayfield Fund

Expertise: investing in consumer internet, digital media, and health and wellness companies.

Brief Recognition: Chang has been named two times to the Midas Touch List of top 100 venture capitalists. He was the lead investor in MOAT and Basis, which were acquired by Oracle and Intel respectively.

Interview Highlights

(02:30) What factors allow an AI company to break into and dominate a space?

Tim Chang: Honestly, I think we’ve shifted to an era where firms that aren’t as [involved with] AI and [don’t] have the talent and proprietary data sets to arm their AI are just not going to exist. I think AI machine learning is going to have…sweeping effect in terms of better understanding what your clients want 24/7, serving them on a one to one personal level…at scale to every single person while reducing the cost basis relative to humans and everything else out there on the cost side. This will have such a sweeping dramatic effect that I dare say the majority of companies won’t exist anymore in…10 years if they don’t harness this. Because it will be table stakes at this poker game and won’t even be a competitive advantage…If you don’t have this, the other large platforms will make your business go away.

(4:30) So companies that don’t use AI are going to go away. The firms that do really win and dominate, what are they going to do well?

TC: Well, let’s take two cuts at this. We’ll look at AI applied to horizontal markets, where you will take existing applications and turbocharge the capabilities of workers with those tools and applications with AI. Then we’ll take the second cut, which is by vertical, and this will be specialists in verticals like aerospace or healthcare, pharma, whatever it is that leverage AI and machine learning to better innovate on products.

So on the first cut, horizontal, every business plan we see today…take a successful existing application by line of business, whether it’s human resource management, customer relationship management, customer support software…add AI machine learning, get 10x productivity out of one-tenth of the staff.

Quick example: one of our star companies in our portfolio,, up in Seattle, basically has machine learning applied to helping salespeople write better sales emails. So imagine a plugin in your Gmail or your Outlook and a salesperson is about to dash off. The sales emailing system says, “Whoa, whoa, wait a minute before you send that. Let me rewrite that for you because based on the billions of emails I see and the open rates and engagement rates, I know that these specific phrases will get a better open rate.” And so that in a case is turbocharging a salesperson to be able to better write emails, but not hard to imagine after awhile that the machine would just write better emails than any human period. Right?

Repeat that playbook for all the existing apps out there, the CRM, marketing automation, on down the line. Frankly, I think big parts of my own job as a venture capitalist would probably be done better by an AI eventually too.

The second cut: verticals. If you have AI as a better way to map out and do product R&D, something I like to think about AI is it can map out and simulate all of the permutations, all the edge cases, all possible scenarios faster, quicker, and less biased than humans can sometimes, depending on if you set it up correctly.

So let’s take in the case of say, small molecule discovery in the pharma industry, rather than trial and error by humans, you know, you can kind of turbocharge that and say, here we have the machine learning look at all this and start creating all the possible samples and permutations. At first, the human’s there to judge which cases are novel and worth further exploration, but eventually even that might be able to be something that the machine can spot out the notable examples as well.

The playbook for enterprise software I’ve seen with this would piggyback on an existing behavior, existing application. Behavior change is really hard in trying to get new markets to go. So instead of doing away with that salesperson, ride on top of what the salesperson already does.

For example, hence the plugin into email or Salesforce, learn from that, provided that you have the instrumentation setup to see what are win cases that you want to reinforce. Nowadays all business is done via digital and email, where you end up going to tracking pixels, engagement, checks, all that kind of stuff so that you have a feedback loop.

And so you can be able to instrument and be able to analyze and have analytics on what the positive conditions are that you want to train for, and then if you have enough of a corpus of data, in this case, millions of emails go out a day, right? And then as you adopt this plugin onto those enterprise applications your system learns on top of it, and it provides benefits by seeing not just the emails and the workflow of the employees within the same corporation, but of employees across all corporations.

This is a good use case because it’s a well-defined area. It’s an existing application, tends to follow a specific workflow, and therefore the problem is usually constrained pretty well.

Part of it is if you see enough top docile salespeople in that vertical and see their lingo or whatnot, you could probably train off of that, right? And know what language and what works for that particular domain. I think …the big productivity gain here, is eventually [the AI writes the emails] for me to everybody in the whole world. Period. Done.

I’ll give you kind of a crude example, but I’m a musician and if I had enough pieces of data I could feed into a machine, there’s a Ted Talk about this, then the algorithm would kind of know stylistically what my fingerprint music is, and I could say, “Okay, go do analysis on The Beatles and then go mash that up and go write every possible piece of music I would do if I were a member of The Beatles.”

And then I’ll just sift through and see the most interesting examples or something, and then claim authorship and I’m done. So we’re moving from systems of record, which are just tools which put all the friction on the worker to do, then you shift to systems of engagement, then we’ve got more systems of insights and scoring and analytics. But the next is systems of just the automation, the workflow automation there, where it’s sort of, you know, “Help me do it. Tell me what to do and then do it for me.”

(15:00) Is the point to get to a place where one company becomes the layer that meshes with the behavior in the world and get so many instances of that behavior that their solution now is better trained, more capable than any other out-of-the-box solution would be? Then people have to basically go to the company because their snowball of machine learning, improving outcomes, ease of use, is rolling so much faster than everybody else’s. There’s really nothing left to do but kind of at least monopolize your slice, right?

TC: I believe so. First step is, “can I make my workers smarter or just more productivity by automating and take more things off their plate or doing things for them in a better way in that workflow?” What I do wonder though, is will that lead to an escalating arms race of these, let’s call them assistant bots if you will. So if I have sales bots that know exactly the language to get me to open [an email], but then I now have a sales defense bot which is screening all the inbound ones and what you get is this ongoing cat and mouse game of who can hack whose attention.

In the startup world, you have limited time and money and your biggest bottleneck is go-to-market strategy, and so if you have to evangelize or paint pictures of what could be done with AI machine learning as a platform or general tool, you will run out of money. And so, again, back to behavior change, instead of a whole new workflow, it’s easier to say, “Look, you already use CRM or marketing automation, here are your ROI and metrics of success. What if I can double, triple, 10x that now? What if I can get you more productivity or cost savings on something you already do?”

It’s an easier sales pitch, and that’s been the flaw of the first generation of AI machine learning startups where they’d come with this general toolkit and say, “I can do stuff for you. What can I do for you?” And then the customer’s response is a blank stare, and they say, “I don’t know. What can you do for me?”

(18:30) How do we bring AI internal?

TC: On the big business side, the advantage they have is they probably already have giant pools of proprietary data, it’s just that nobody has taken the time to mind through or sift or know how to. And that’s where you can engage a startup or specialist to come in and do that. It’s sort of like, I have this backyard full of stuff, I know there’s jewels in it, I don’t know where to look, can you go prospecting for me?

Subscribe to our AI in Industry Podcast with your favorite podcast service:


Header Image Credit: EECU

Stay Ahead of the AI Curve

Discover the critical AI trends and applications that separate winners from losers in the future of business.

Sign up for the 'AI Advantage' newsletter: