AI is a Game Changer, But Not When It Comes to Business Fundamentals
Back in the ‘90s when digital media was taking off, no one had really figured out how to make any money from it. In lieu of offering metrics showing how ads had influenced consumers, the industry settled on another measurement: eyeballs.
In 2018, no one cares about eyeballs because marketers now have more concrete metrics connecting digital ad exposure to goals like purchases and brand affinity.
We’re seeing a rerun of such questionable metrics with the rise of AI. Because it’s difficult to judge how an AI system might progress, some have suggested using different measurements for AI startups. The logic is that AI has created new dynamics that mean the standard rules of business don’t apply.
This is absurd. Business fundamentals don’t change because a new technology has been added to the mix. Focusing on AI as if it exists in a vacuum also ignores the fact that AI is only useful to the extent in which it solves a business problem.
AI does throw a spanner in the works because companies can possibly scale up more quickly than in the past. They might also become more profitable. The problem is that most AI in 2018 is a work in progress.
For example, if you need to transcribe recordings, then an automated solution is attractive because it’s faster and a lot cheaper. The problem is that what’s available now is still a bit rough. Even the best automated transcriptions are in the 90% range, which sounds good until you read them. Then the question is how long until the system is good enough to be truly useful.
That’s the calculation that investors make with AI. In addition to judging the business fundamentals and the human talent, investors have to make an educated guess about how well the technology will improve over the next few years. If you’re investing in an automated transcription firm, then the question you need to ask is whether the system’s accuracy will be close to 100% in two years, four years or never (as well as how easy it will be for competitors to do the same thing!).
Since all AI startups aren’t created equal, grading their technology is important. For instance, an intervention ratio tells you how much of the startup’s functions require human intervention over time. Ideally, in a year, an AI startup will rely less on human workers as more functions can be automated.
Other considerations include the amount of data available for the system and the way it’s organized. Obviously, the more centralized and accessible the data is, the better. Finally, as with the transcription example, you need to factor in future performance. If the system is able to get a critical mass of data and its algorithm is well constructed, then performance should improve over time, often dramatically.
Focus on business fundamentals
While these are all important factors, AI companies are really software companies. A biotech startup may have no underlying business, so its valuation is based on dates of testing on non-humans and then humans and the possibility of getting FDA approval. In that case, the company is built around an experiment and investors bet on how successful the experiment will be.
That’s a different model though. A successful drug will sell itself because it will meet an existing market need. But in order for AI to meet a market need it has to be focused and trained in a very specific way, by the right people. For instance, a company might build itself around the mission of using data about soil, fertilizer, water and weather conditions to help farmers get greater crop yields. In order to do that, the company has to recognize that market need and train a system to use that data effectively. That’s a lot different from making a drug in a lab somewhere. By the time you identify a market and a need, you already have the makings of a business.
Where AI startups differ from standard tech startups, though, is the extent in which their AI will evolve and mature. That is the experimental part of evaluating a startup, which is tricky since not every system is going to work well. You can throw a lot of smart people and resources at a problem and still fail. Look at the war on cancer, which has been going on for nearly 50 years and although there have been great strides in some areas, overall we haven’t made much progress.
Not as much of a differentiator
The final reason why AI startups shouldn’t be treated differently than standard software startups is that everyone is using AI to an extent. Over the next few years, everyone is going to be integrating machine learning to a degree.
The good news is that AI works best in very specific use cases. That means a startup can still trump Google or Amazon if they have a great idea. It also means that, despite the possibilities of AI, business fundamentals still trump everything else.
Achieving true product-led growth takes a winning combination of free parts of your product, virality, paying users, and more. Startups spend years (and thousands of dollars) trying to figure out the right model for viral growth – and many never do. So how do you succeed at PLG. Find out here.
Eraser founder, Shin Kim, shares why his company, Eraser, a whiteboard for engineering teams, built an AI sidecar that ultimately drove 30% of all product sign ups. Learn more here.
Miro’s Kate Syuma shares how the company’s growth team iterated smart to improve the user onboarding journey for their popular collaborative platform.