AI Startups: Don’t Just Focus on the AI
Nowadays it’s hard to come by a pitch deck without the phrases “AI” or “ML,” as if they are magic bullets. Unfortunately, the quality isn’t very high most of the time.
We have several portfolio companies that use AI/ML to power their products — for example, Qordoba uses ML to suggest translations for words, Imbellus uses AI to generate art for their games, and GumGum uses ML to identify images with specific features in order to place relevant ads — but they don’t bill themselves purely as AI companies. What’s going on here?
If you are working on your pitch right now, here’s a pro tip: don’t just focus on the AI.
Instead, focus on your business model, the opportunity you’re addressing, your go-to-market, and so on, like any other startup.
Good ML, poor business case
I’ve seen too many examples where the ML is fine and technically robust, but the business case hasn’t been thought through. For example, one startup picked where to focus based on where the data was easy to come by and more structured, without thinking about the competition (plenty) and differentiation (just ML is not enough).
Another startup applied ML in such an avant-garde fashion that only the most enlightened and forward-thinking customers would understand the product. In enterprise sales, it’s often better to cater to the common denominator than to hope that the industry will advance rapidly.
Yet another one focused their time on building a product to solve a problem, only to discover that their intended customers don’t have the ability to pay. They then realized that monetizable customers do exist but those are at least one-step removed, which complicates their go-to-market.
Finally, some startups are great extensions of university research, have great talent, and probably are on their way to making fundamental breakthroughs, but they are far from having a business yet. The VC model is not built for funding research.
Great business case, poor ML
Of course, companies also err the other way: great business case, poor ML. In fact, when the business case sounds too good to be true, it is often correlated with a limited underlying technology that is far from realizing the product promise. Building a regression is not ML. Setting up a rule-based Q&A engine does not make a conversational agent. In today’s competitive environment for AI talent, I’m hard pressed to imagine how companies without AI DNA on the founding team can attract the necessary talent to build the products they envision.
Some of the best and incredibly powerful applications for ML are where the product evolution is so natural that it’s obvious in hindsight: Google Maps has steadfastly built a moat such that the product is undisputedly best-in-class. Business press, to its detriment, is full of rhetoric making lofty promises about AI, but real, incredible products powered by ML are built from adding up incremental, everyday investments.
Solid business case, great ML
The magical combination is a solid business case supported by an astute application of AI. AI is an enabling technology. Here are some questions I ask when I evaluate AI startups:
- Does it leverage a large dataset? Ideally, the said dataset is not in places where Google, Amazon, or Facebook would have even larger datasets.
- Has the AI technology just reached or is approaching the tipping point between research and becoming useful in real-life applications? For example, image recognition and autonomous driving are relatively mature, while there has been rapid recent advances in NLP.
- Does it solve an immediate need? Is it a killer app? Can it solve the cold start problem so that there is a customer at the ready and they are willing to pay?
✌ I am an early stage investor at Upfront Ventures. If you’re working on AI/ML, drop me a line at chang@upfront.com.
