Every day, multiple companies reach out to me on LinkedIn, making claims about their groundbreaking A.I. In some sense, this is good — it shows the field is maturing into the mainstream, but it also creates a ton of noise, which results in executive confusion, lost time, and wasted budgets. All things I try to avoid as much as I can. Cutting to the point, the majority of these companies that claim to have an "A.I." that does a task such as anti-money laundering, churn optimization, or anomaly detection products, etc., are generally no more than technical consulting services that use your data to train models using open-source algorithms/packages e.g., H2o, TensorFlow, BERT, SK Learn, XGBoost et al. To sift through the hype,over the years I've developed mental hacks to quickly vet firms to increase my chances of hitting on my technology "draft picks," while avoiding A.I. start-ups with marginal upside. I should also note that this guide is for companies that claim their core attribute is their "A.I." Not firms who leverage A.I., but do not claim it as a fundamental competitive advantage. There is a big difference.
In short, here they are.
Funding — Building novel A.I. takes more money than most think (tens if not hundreds of millions).
Technical Talent/People involved — there are a few people with the knowledge and skill to create something unique. Most of the legit people are connected by <2 degrees to the top V.C.s, founders, or thinkers in the space. It's a small world. Red flags if there are zero or few connections.
Location — 95% of the time, if you aren't in one of a handful of cities, you won't have access to 1 and 2.
Documentation & Open Source Code — Firms that are genuinely building, especially small ones, will want to get their code in the hands of data scientists/developers. It shows confidence.
For more context, continue reading.
1.) Investors + Money: Look the company in question up using Crunch Base or a similar database to see who's investing in them. Are investors such as Lux Capital, A16z, or DCVC involved? If you have a budget and want to dig deep - these are multi-million dollar investments, after all, firms should invest in augmented intelligence and alternative data capacities. Quid (graphs below), which has been my long time "go-to," combines OSINT, Crunchbase data and natural language processing to generate a network that surfaces how firms, capital, product domains, and investors connect.
Network graph of companies focused on GraphQl, Graph Databases, and Edge Computing.
Does the data show V.C.s investing in the firm have technical depth and a history of hitting on game-changing technologies and products? Or are they a government or corporate venture arm run by people with a sparse track record of being associated with technologies before they were mainstream. Could they be there for for "innovation P.R." or "tech tourism" VS driving legitimately towards the bleeding edge? To my second point - how much funding is the firm in question getting? Many top A.I. start-ups are in "stealth mode" during seed funding and only work with selected partners. And legit A.I. firms will have raised $30–$60m at series B(!). In London, I often see firms that have raised between $2–$5-$10m at Series B trying to pitch their A.I. to me. While I may love the passion, and those amounts of funding would be significant for many business ventures, in machine learning, that amount doesn't provide enough runway to build anything novel. Key point? Machine intelligence is an arms race, and funding enables companies to hire the best and brightest that actually have the potential to build a product that improves the current in-state tech. The former is as limited as the number of people that have the raw athletic potential to run a sub-4.4-second 40-yard dash (not very many).
The graph shows how investment from 2016 in "Deep Learning" has cascaded to "Quantum Computing" and "Transfer Learning." Due to its lack of investment, firms that focus on Transfer learning have the most upside, while firms that claim Deep Learning expertise should already "levered up" and thus have a mature product stack.
2.) Talent & Network: Do they have the brain trust/talent/investors that could outthink current companies like Element A.I., Facebook, Google, Primer, H2o, Tencent, Baidu, or Microsoft? If they haven't raised requisite money or network, which is a result of being connected to the right funds, they won't be able to afford the opportunity. Does the founder have a past track record of innovation in start-ups or a corporate (potential clients & future funding to reach the critical mass needed)? Machine intelligence is a small world. Anyone who can legitimately build good products are connected to the top firms, people, corporates, and investors in some manner. If they lack these connections, which are easily found on LinkedIn, be skeptical.
Cities where machine learning companies are receiving the most investment which is dominated by the USA and China. And Europe is mostly absent.
3.) Location: If pure A.I. is what you are trying to build, there are few places outside of San Francisco, Toronto (Research), Shanghai, or Beijing with the concentration of elite technical talent needed (and the funding to pay for them). While great products are getting built in places such as NYC, or Israel, these firms are focused on specific tasks/products (security or finance) that use open-source frameworks rather than building their own. Unfortunately, Europe isn't relevant due to low funding levels, and it's unclear how they will compete in the future. The last meaningful A.I. company - DeepMind, was bought by Google more than five years ago. Years before the massive uptick in demand for talent that there is now. Quite simply, anyone who is elite at machine learning will start at $175-225k per year (+ LTIs and or Equity) in the U.S. and, in some cases, Asia. About two-four times more than most jobs on offer in E.U. or the U.K. If you have the skills and are one of the best, it's obvious where you go. The same reason why you typically don't see NBA level talent in the EuroLeague. Or top soccer/football talent (in their prime) playing in the MLS.
4.) Documentation + Open Source Code: Does their website have elegant documentation on their supposed A.I.? Are there technical white papers and or an open-sourced version of their algorithms or package that your team can test? If not, I can say with almost certainty they are using the same machine learning packages that everyone does. There isn't anything novel about what they are doing from a pure machine learning standpoint. While the former can be a legitimate approach, know that you're paying them to train machines. And your firm is missing out on the opportunity to build the ML muscles it likely needs to stay relevant in the future. My policy is such is that if I am not connected to the A.I. company in question through my network, and there is no documentation, 95% of the time, it's a non-starter. Alternatively, if a firm has excellent documentation, in addition to an API or an open-sourced version of their technologies, it gets my attention almost every time.