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
SigOpt has built technology solutions for some of the earliest adopters of modeling since 2014. These users treat modeling as a core business strategy for driving revenue or differentiating products. This gives us a unique – if biased – perspective on the evolution of enterprise AI adoption. Similar to the purpose of TWIML’s ML Platforms Guide that we sponsored, we hope that sharing these observations brings a bit more clarity to decision makers in a position to fundamentally transform their companies with investment in AI.
Enterprise AI has become a catch-all term for any technology that captures a portion of the billions of dollars that companies are now spending on AI. But this broad definition obscures the nuance that shapes this market. In short, we see three distinct markets that make up Enterprise AI:
Some processes can benefit from applications that have models embedded within them, rather than building new products and models from scratch for different use cases. This is most often the case for problems that are relatively uniform across companies and have a mature software market. Chatbots are an example of a competitive, saturated market of application vendors that obviate the need for you to build your own capability unless you expect this to be key to your product or revenue strategies.
Some modeling problems can be solved relatively easily with basic, off-the-shelf models and others are more like extensions of classic analytics problems. In these cases, you do not want your data scientists or ML engineers wasting time building a custom model to solve the problem where performance isn’t a priority. Instead, enterprises have started to arm citizen data scientists or business analysts with GUI-based AutoML solutions that make it relatively easy to quickly build these simple models. These models are more likely to be used in cases where a team is seeking either better (or more predictive) insights than traditional analytics methods or implementing models to cut costs for their business.
The most important modeling problems for any business are those that are relatively proprietary to them as a business. Solving these modeling problems tends to boost revenue or differentiate products. Data scientists and ML engineers spend outsized portions of their time building these models because they have significant upside and tend to be difficult to solve. This process requires an entirely different set of technology solutions that are designed to augment these experts. In some cases, this technology solves a DevOps problem that is non-core to data science. In others, it automates a non-domain specific part of their workflow like hyperparameter tuning and experiment management so experts can spend their time instead on areas that benefit from this expertise like feature engineering. In all cases, these technologies are critical to the Machine Learning Operations – MLOps – process any organization needs to implement to standardize and scale their modeling.
Each of these is a market rather than a segment because they tend to have mutually exclusive buyers, technology needs and business goals. (In some cases, the buyer may be the same for Basic and Differentiated Models, but this is often not the case.) But each enterprise typically invests in each of these types of AI at the same time. Take a banking example. Customer success purchases a chatbot to predict and respond to a portion of customer requests. Investment banking upgrades their Looker subscription to include an AutoML solution to accelerate a pricing project. Algorithmic trading combines best-in-class modeling solutions with their own homegrown software to generate additional trading revenue.
Summary of each of these markets
|Approach||Embedded Models||Basic Models||Differentiated Models|
|Goal||Apply AI to standard business process||Make simple modeling problems easy||Make hard modeling problems possible|
|Users||Functional non-technical staff||Citizen data scientists or business analysts||Data scientists and ML engineers|
|Uses||Upgrade apps or automate processes||Cut costs or enhance analysis||Grow revenue or differentiate products|
|Technology||Out-of-the-box applications||GUI-based AutoML or embedded AutoML||Modeling Platforms and Solutions|
Each of these markets offers a valuable service to their customers, and, as a result, will be a multi-billion dollar market (if they aren’t already). But we believe the market for differentiated models will be an order of magnitude larger than the others for three reasons. We will explore these reasons in our next blog post.