This post is part of a five-part series. Follow these links to read any post in the series:
- Post 1 – Why Enterprise AI is Actually Three Markets in One
- Post 2 – Differentiated Models are Eating the World
- Post 3 – Technology Considerations for Machine Learning Operations
- Post 4 – Modeling with the Modern Machine Learning Stack
- Post 5 – How Teams Use SigOpt to Build Differentiated Models
Software is still eating the world and modeling is eating software development. But the models that will eat the world aren’t embedded AI or basic models. They are the differentiated models we discussed in the first post of this series. In this first post, we divided the Enterprise AI market into three – embedded AI, basic models and differentiated models. Revisit that post to learn more about these definitions. In this post, we will explore why differentiated models – and the technology supporting their development – represent the biggest opportunity in Enterprise AI.
Creation of Business Value
First, differentiated models will represent the lion’s share of value created by AI. Two Sigma, a SigOpt customer, has grown its assets under management from $8B to more than $50B due to the performance of their differentiated models. More famously, Google’s pagerank algorithm is a large part of why it is now valued over a trillion. These are the models that change companies and owning them can be a decisive advantage in a growing number of markets. As an example of this, a company that works with us publicly values their modeling ensemble at over $1 billion. A 5% improvement to the performance of this ensemble represents $50 million to them.
Democratization of Resource Availability
Second, a growing number of companies have the resources to build world-class models, and you should expect this trend to accelerate. Building these models is largely contingent on availability of compute and talent. The combination of hardware innovation, elastic cloud services and distributed computing is facilitating unprecedented access to compute resources for training and tuning models. OpenAI estimates a 300,000x growth in compute required to train the most cutting edge model from 2012 to 2018, which suggests exponential growth in availability as well. At the same time, our own research shows that more than 200 companies have hired more than 10 data scientists, 150 companies have hired deep learning engineers, and over 100 have invested in building their own modeling platforms to boost team productivity. AI specialists are the fast growing job listing according to LinkedIn, and all of these are not just listings from Google, Facebook, Amazon and Microsoft.
All Modeling is Differentiated Modeling
Finally, companies that create embedded AI applications or AutoML supporting basic models must build their own differentiated models. As they invest in their own modeling teams and these teams grow larger, they will need the same modeling stack that any enterprise needs for their own modeling teams. Rather than dogfood their own product (in the case of AutoML) or pull a pre-trained algorithm from DeepMind off the shelf (in the case of embedded AI), my bet is that the successful companies in these categories will build their own differentiated models. Even if the adoption rate assumed in my second point is lower than expected because of embedded AI and basic modeling technology providers, this should still create more opportunity for differentiated modeling – just a different customer than anticipated.
So if building differentiated models is going to be a big trend in Enterprise AI, then how do you go about doing it well? In the next post of this series, we explore the technology requirements for these teams and why they may look a bit different than traditional software requirements.
We’d love to hear how it stacks up to your experience. Email us at [email protected], try our product or sign up for blog updates.
Use SigOpt free. Sign up today.
