If you’ve already imagined how profoundly self-driving vehicles will change our lives, think of what self-driving companies could do. I’m talking about self-executing enterprise software that, when mature, could enable businesses to virtually run themselves. Autonomous companies may be less sexy than autonomous cars, but their impact on society will be just as significant.
The road to autonomy
Autonomous vehicle systems are ranked from zero (no automation) to five (fully self-driving). For example, Tesla’s Autopilot, which requires drivers to keep at least one hand on the steering wheel, is Level 2. To borrow this scale, today’s business software would rank somewhere between zero and one — the power-steering stage, let’s call it. Most current B2B software is workflow-based; that is, software that helps organize and facilitate routinized tasks. Salesforce, the cloud computing company, for example, is largely a workflow-driven software solution. To get paid, sales reps of enterprises using Salesforce have to input their activities, which allows supervisors to monitor their work and manage the sales pipeline more efficiently.
This type of business software has unlocked enormous productivity, and most multi-billion-dollar B2B-software companies today are some form of workflow solution. Over the next decade, I believe that these impressive results will be dwarfed by the value created when AI-driven business applications attain Level 4/5 autonomy. And as technology supplants human performance, the very way we think about work will shift from machines assisting humans to humans assisting machines.
Let’s imagine an AI-enabled, self-driving version of Salesforce. Sales activity would be input automatically. Even more powerfully, the system would source and prioritize leads that have a high likelihood of closing; it would draft correspondence for these leads; it would even reach out to them via the most appropriate channels (chat, email, etc.). Salesforce.ai would then go back and forth with these potential customers to drive them through the sales funnel, engaging a human agent only when the machine is uncertain or it’s time to take a prospect out to dinner.
It’s hard to overstate how transformative this would be for the company. If Salesforce’s software could find, prioritize, and reach out to leads without human effort and predict which leads are most likely to close, its utility to customers would increase by orders of magnitude. So much so that it might even be more profitable for Salesforce to shift business models from its current subscription-based fee to charging a percentage of new revenues generated for its customers. It would be such a game-changer that it’s difficult to see how non-autonomous companies could compete with a self-driving Salesforce or NetSuite or SAP.
We’re all drowning in data these days. For me, it’s everything from thousands of selfies with my two cats to hundreds of hours of uneventful video footage captured by my Nest cam. But what matters to businesses is meaningful, exclusive data. Whether a company will be able to make the leap to becoming an autonomous enterprise is contingent on one non-negotiable factor: access to high-quality, proprietary data.
A proprietary data set is one that meets at least one of the three following criteria:
- Uniqueness. An example of unique data is distinctive population data, such as the genomic data set of an unusually homogeneous country. But truly unique data sets are increasingly rare.
- Scale. LinkedIn has one of the largest résumé books in the world. Is each individual profile so unique? Not necessarily, but the scale of all them taken together is proprietary. More importantly, with new users joining every day and current ones updating their profiles, LinkedIn has an organic way to update and grow this asset.
- Weight. Facebook has profiles, and each profile is interesting, but what’s more interesting is the weight of the relationships — how significant the connections are between persons in that network. A strong relationship is heavily weighted, weaker relationships less so. The weighting of data-network relationships is important because it helps to train AI algorithms more accurately, resulting in better predictions.
Getting up to speed
So far so good … for Facebook, LinkedIn, or Salesforce. But will our future robot overlords simply be software upgrades of our current corporate overlords? How do companies that aren’t tech giants compete? As someone who, in my day job as an investor, has met hundreds of enterprise-software companies and worked closely with several, I have three pieces of advice for AI-aspirational startups:
1. The Day 1 Imperative. B2B founders understand that one of the biggest hurdles to developing AI-enabled business applications is acquiring their own proprietary data set. But some over-realize this challenge. Some AI startups expect to just conduct data collection for their initial lifecycle. Or they plan to pilot with a customer that will share its data but won’t receive anything in return until the AI is trained in six months to a year.
In other words, they’re so focused on amassing their data asset or differentiating themselves as an “AI company” that they lose sight of the fact that in order to build a data-first business, you have to build a business, first! An AI-enterprise startup — like any startup — must deliver a product with a compelling business use case and provide significant incremental value for its first customer on Day One.
2. The Golden Horde. Each enterprise customer that you acquire will contribute its data. Your proprietary data set should become more robust with each of these additional data contributions, because they further train your AI model. In other words, build a product and a company to harness the network effects of growth.
Take Mya Systems, an AI-powered recruiter in which I’m an investor. Mya’s initial customer was an English-speaking business in the industrial manufacturing space. For this first customer, Mya’s AI had to be taught basic manufacturing jargon. A subsequent customer in industrial manufacturing was Francophone, so Mya’s model had to learn French. But it didn’t need to be retaught manufacturing jargon. And now, Mya’s bilingual conversational AI can communicate with all existing and future customers in English and French — at least on topics in industrial manufacturing.
Network effects allow for the information and experience of one customer to result in a better solution for all customers.
3. Virtuous cycles. Ideally, a B2B AI startup would also build a solution that gets its customers to do work for them. That is, design your product to incorporate constant feedback from customers in order to further sculpt your AI algorithm.
For example, Teamable, another AI-recruiting startup, uses machine learning and social networks to drive job referrals. It works with companies like Lyft and Spotify. Teamable shows Employee X of one such company a job and asks her if she thinks the role would be a good fit for a particular friend of hers. When Employee X indicates yes or no, she essentially becomes a mechanical turk for Teamable by helping to build a data set with proprietary weights, which trains the algorithm. The idea is that, as the AI learns, Teamable’s model will become a better prediction engine for which candidates will match which jobs. Over time, as its software grows increasingly autonomous, the business will become increasingly self-regulating and self-perpetuating. And, eventually, the company will drive itself.
But we’re not there yet; we’re only at the power-steering stage. We still need genius gearheads and Fast & Furious drivers — i.e., entrepreneurs — to get us over the finish line. But they have the map, fuel source, and keys to the enterprise of the future — the autonomous enterprise. All that’s left to say is, ladies and gentlemen, start your engines.
To Read Our Daily News Updates, Please Visit Inventiva Or Subscribe Our Newsletter & Push.