A lot of the folks I’ve been meeting recently have been so eager to join an AI startup that they forget that not all startups are created equal. In this post, I want to share with you what to look out for when joining an AI startup.
To start off, there are a bunch of questions relating to the founders, team, funding, vision, business model, learning opportunities etc. that you should ask, these questions will probably cover 80-90% of what you need to know about any startup, not just AI startups. Paul Graham has written a ton about this subject and his advice is probably some of the best around.
Assuming you’ve done your regular startup due diligence, here are four AI specific questions you should be asking:
- Where does the data come from? You can expect that most of the competitive advantage you’re going to have as a startup will come from some really specialized data source that is hard for competitors to access. All the big guys have more money, better AI talent, and greater compute than any startup, the only way to have better AI than them is through the data. For example, this can be data generated internally from the users of their app (e.g. Facebook) or it can be access to specialized private datasets such as retail transactional and ERP data in the case of Rubikloud.
- How do they treat data? As a corollary to the previous point, data is the lifeblood of an AI company. Data needs to be a first class citizen, full stop. Good data practices are necessary for good AI. Companies that effectively use AI almost always take things like collection, cleaning, storage, processing, etc. of the data very seriously. Companies that don’t take it seriously usually end up hitting some wall on how good their models can perform. The good old saying still rings true: “Garbage in, Garbage out”. The decidedly unsexy truth about building a successful AI startup is that you need to have good data and data practices.
- Practitioners vs. Academics? It’s important to understand the type of AI experts that an AI startup employs. By and large, you want people who can find practical ML solutions for a given business problem — not academics (in the theoretical sense of the word). Very rarely will you need to invent some new algorithm or architecture to solve your business problem, you’ll just use whatever is on hand, usually an off-the-shelf library or something in a robust framework like Tensorflow. Importantly, it’s very rare that the model or algorithm is the competitive advantage (see Question 1). Instead, you want your AI experts focused on shipping value-producing models, not humming and hawing about the theoretical trade-offs of different neural network architectures.
- How does AI contribute to business value? AI doesn’t produce any business value in and of itself, it’s only valuable when it’s applied to some business context. Almost all value-producing use-cases for AI will be related to automation, optimization or prediction, each of which can be tied directly back to some sort of business value. If AI isn’t helping along one of these dimensions then it’s probably not contributing much value.
These questions cover a lot of ground and focus on the key points relating to building an AI startup. At Rubikloud, we’ve been focused on building AI systems to solve core business problems since the beginning. We’re serious about data, pragmatic AI practitioners, and are always focused on delivering business value.
If you’re wondering about how our AI systems directly contribute to the bottom line of retailers around the world, head over to https://rubikloud.com/careers/ and ask us these questions in person!