by Jason Costa AI and deep learning will be a fundamental game changer, akin to the wave of innovation that was ushered in as “mobile” and “cloud” became major category enablers over the past decade. When we think back to the advent of the internet, companies needed a different kind of DNA to build a massive organization in the age of a new platform. In other words, it wasn’t enough to take an offline shopping mall, with brick & mortar stores like Sears or Macy’s, and give them a website — that just didn’t equate to an ecommerce internet company. It took a brand new organization with a different way of thinking to build Amazon and completely overhaul the ecommerce landscape in the two decades that followed. Brick & Mortar + Website != Internet Company A big part of that was because there were too many new roles, methodologies, and processes that required developing new skill sets that the old guard was either unable or unwilling to do. This included processes that previously didn’t really exist, such as A/B testing and rapid experimentation. Companies just getting onto the internet hadn’t seen A/B testing and experimentation at this scale. This gave way to an ever exploding need for data engineers to outfit existing systems with a scalable framework to routinely deploy experiments, and a slew of data scientists to interpret the results of said experiments that were coming in on a daily if not hourly basis. Data Science in the age of the web is a fairly new and unique role that hadn’t really existed at this scale previously. Furthermore, there was suddenly a change in development cycle time for product. The web enabled companies to deploy a new product literally whenever they wanted. Sears put out a catalogue once a quarter; Amazon could put out a new “catalog” every day (or hour). This again was something totally new to the nature of software and the internet as a platform, requiring a different type of preparation and planning to fully exploit. Once teams could get coordinated and on a release train cadence, they were able to upgrade or fix systems with users on an extremely quick basis. This also allowed for a highly compressed product feedback loop, allowing teams to improve their product ever faster than before. The new organizational DNA included the creation of a “Product Manager” role as well as elevated power for Engineering. It was no longer the case that engineering was relegated to a role in a dark corner, taking their marching orders from sales & marketing. Suddenly, engineers were celebrated as heroes who built and had a major voice in the product development cycle (rightfully so). The arrival of Product Managers gave rise to individuals who were technical yet had a keen business sense as well — they were equally adept at holding a conversation with an engineering counterpart or discussing user interface requirements with design, to chatting with sales & marketing about the expressed needs of customers. Most important — they were capable of cutting through all of the external and internal noise to ship products into the hands of users. Internet Company + Deep Learning != AI Company In the same way that a brick & mortar company can’t just deploy a website and suddenly become an ecommerce business, it won’t be enough for traditional internet companies to build an array of neural networks and then morph into an “AI company.” The coming platform shift to AI will require new organizations, roles, and methodologies. This will be particularly pronounced in areas where robotics are involved: self driving cars, industrial automation (robots in factories, warehouses, etc.), automated last mile delivery, and so on. Data Network Effects Suddenly, data will be even more important than it was in the age of the internet. Strategic data acquisition is going to be absolutely paramount in the age of AI. Extremely large, differentiated data sets that feed into training frameworks to generate learned models will be a major competitive advantage for new AI companies. Many companies will likely use hardware to collect this data. They can sell hardware as a loss leader, to penetrate certain markets (home, car, office, etc.). That hardware presence will give these outfits , which means that over time these companies will procure a compounding competitive advantage amongst other players in their respective category. The more data a given company acquires in a vertical, the better their models and automated decision making will become. Lastly, it will need to be “good”, “clean” data; not garbage in, garbage out. Better products lead to better usage, which in turn provides more data — hence the advantage of the data lifecycle. data network effects Decision Automation Decision automation will also become critical. By this, I mean the ability for a machine to make a decision without requiring a human in the loop. The faster that a company’s product can reach autonomous decision making, the more competitive advantage they will have. For instance, in the case of last mile delivery robots, how often will a human need to perform teleoperations on the robot? If a human doesn’t have to remotely interject when the machine crosses the street or has to avoid on obstacle on the sidewalk, the more economies of scale a company will see. Then the question becomes how quickly the producer of such a robot gets to 75% autonomy, or even 90% autonomy — where there’s one teleoperator managing fleets of these robots, when they hit an edge case requiring some kind of human intervention. This will be true in factories and warehouses as well, with these robots needing to reach the highest possible levels of automated decision making. New Product Teams There’s likely to be a plethora of expanded roles within these next generation orgs as well. Product management will change, as will engineering. Over the course of the internet and mobile eras, most companies in these respective sectors have opted to focus their attention away from hardware. Instead, there was an emergence of the modern product organization triumvirate: product, design, and engineering. With autonomous vehicles, industrial automation and robots, smart homes, and so on — there will be a need for mechanical & electrical engineers, industrial engineers, potentially even a need for physicists and mathematicians as well. Certainly there will be a need for more data scientists, and data science itself may become a tradesman craft, where schools like Hack Reactor train a tier of data scientists to apply industry standard tools such as TensorFlow to specific datasets. PhDs who develop new modeling strategies may become more like building architects, contracted out to deliver blueprints but possibly never even setting foot on a construction site. The Role of PM The Product Manager role is also going to change dramatically, and I suspect will be cut into two roles. On one side, there may become a much deeper emphasis on program management and product marketing for a set of customer facing product managers. These PMs will be constantly interfacing with customers, more aggressively monitoring timelines, and procuring feedback from the market. They’ll also work closely with marketing on narratives for go to market, focusing on making robots and AI services friendly to users and businesses. On the other side, there will be highly technical PMs who work closely with engineering (and likely even report into engineering). In the context of AI & deep learning, these PMs will need to understand what is possible right now, and then plan for what data they need to collect to make more things possible in 2 years, 5 years, and so on. That means it will be much more than just building a feature roadmap; it will also require a data roadmap. Furthermore, it won’t be just a matter of getting any data and massaging that set — products will need to be built with data collection in mind from the start. PMs will need to figure out what signals are important, and then build the product to collect the best data sets providing those signals. Conclusion While it’s going to take time for AI to evolve and more deeply penetrate the market, I do believe that deep learning will become a pervasive part of our everyday lives — from how we get to work and our productivity once we’re in the office, to how we get our food and interface with our home living space. Organizations will have boundless opportunities to capitalize on this, but also many pitfalls to avoid (see ). If the internet wave was any indication, there will be many winners and many losers in this next game. For existing players, their success will heavily depend on how well a given company can evolve their DNA and ride a new platform wave, taking cues from lessons of our internet past. The next decade is going to be amazing. Blockbuster Jason Costa is currently an EIR investing at GGV Capital. This post is part of an ongoing series aimed at exploring topics such as consumer product development, platform analysis, and strategy.
