Before you build another machine-learning startup, read this

The hype around artificial intelligence/machine learning has reached mythic proportions. Some commentators are calling AI the fourth industrial revolution. Others are calling it the new electricity. And I’m a believer. An incredible amount of money is pouring into companies focused on AI/ML, as it has the potential to revolutionize most, if not all industries. See graph below.

Avoid the trap — focus on real problems

Such a major technology revolution deserves broad and deep financing, potentially justifying the dollars being invested in the space. But I’m concerned that many entrepreneurs are falling into the trap of focusing on the infrastructure of AI – the algorithms and platforms – rather than the applications.

Successful companies usually start by solving a specific customer problem and evolving over time to deliver wider “platforms” based on the solution. What matters is the specific business problems being solved, not the technology itself. But, given how exotice the technology is, it’s easy to get caught up in the algorithms and models and ignore the applications.

So, what makes for a great AI/ML startup? If you try to generalize it, it boils down to a company that is focused on one or more of the following items:

• Eliminating/reducing human labor in areas previously believed to be hard to automate
• Exploiting a white space that has emerged due to a new capability (new products/services that weren’t previously cost-effective/possible)
• Making traditional applications significantly more valuable by embedding ML techniques into them

Why you should avoid a horizontal ML platform

Did you notice there isn’t a category for machine learning platforms? Here are a few reasons for it. Web-scale companies like Google and Facebook are not only investing heavily in AI/ML but have adopted the strategy of open sourcing their tools and platforms. Web-scale companies are hard to compete with given their access to immense resources and unique proprietary data sets. If you try to differentiate your startup based on these capabilities, you’ll be at a big disadvantage versus ‘the big guys’.

Additionally, the widely reported shortage of data science talent impacts a customer’s ability to take advantage of platforms and algorithms. This lack of AI expertise means customers do not have the capability to build their own AI/ML, so a startup with a horizontal platform ends up in a professional services role, helping each customer define and accomplish their specific goals.

One other important aspect that entrepreneurs building horizontal AI platforms should think about is the complexity of the go-to-market process. Different verticals may have different buying behaviors. And you may need to approach different verticals via different channels. Of course, before you choose a vertical, you should be sure it promises sufficient scale and growth to support a large business.

Choosing an industry to focus on

If your ML model can be applied to multiple industries, then here are some of the variables you should think through before deciding which one to focus on:

• Cost to deploy. How much will it cost your customer not just to purchase your technology but also to change from their current solution to the new one? For example, if a worker in manufacturing in China earns $6,000/year and a robot to replace this worker has capital costs of even $40k, there is a minimum payback period of 6+ years (not including operating costs). This is unlikely to be attractive to the average factory manager.
• Added value beyond cost. What value does your ML-based software offer beyond labor substitution? Better quality, enhanced customer satisfaction, fewer errors, higher performance or throughput, etc? For example, when it comes to hiring, people have biases and predilections. So startups like Gild, Entelo, and Textio have developed ML based software that automate hiring without these biases.
• Regulatory/compliance issues. Is there a lot of red tape that might complicate the adoption of your offering? An obvious example here is autonomous cars.
• Conflicting goals within potential customers. The scale at which AI/ML will eliminate/reduce human labor is likely to be significantly larger than any prior technology, resulting in a much higher resistance. Will the human teams you’re selling to lose their jobs as a result of your technology? For example, a major concern among IT outsourcing companies that bill by the person hour is the reduction of “routine” maintenance work due to automation.
• Industry readiness. Sometimes, an industry is just not ready to adopt a new solution due to extreme risk-aversion. We see this in industries that have incentives focused on uptime rather than efficiency and could also face penalties for downtime. A perfect example in this category would be operation of the electric grid. Obviously, if the market is huge and sufficient capital is available, an alternate strategy may make sense – such as Uber’s willingness to fight regulation and Taxi unions around the world.

In summary, to make the most of the huge opportunity in AI/ML, you should:

• Avoid areas where the large web-scale companies have structural advantages
• Pick a beachhead industry/set of use cases where your product solves a clear pain point and the buyer is not inherently conflicted
• Select a target industry based on the degree of readiness of the sector to adopting AI/ML technologies, without major regulatory hurdles to adoption.

Preeti Rathi is a Principal at Ignition Partners.