More enterprises are deploying machine learning models than ever before as AI tools and data become cheaper and easier to use. These models can help businesses forecast demand, augment employees' workflows and better attend to customers' needs.
Enterprises with a firm AI-first strategy -- collecting and using data to constantly improve predictive models to automate core functions -- can quickly adapt to changing market factors and customer demands. As a result, businesses can react more proactively, instead of reactively, to new situations and tackle existing business problems faster and more intelligently.
According to startup investor Ash Fontana, author of The AI-First Company, businesses that incorporate AI into their every move can quickly outpace companies that don't in the near future.
Experts agree. Deloitte's 2020 State of AI report found that about 80% of business and IT executives believe AI will be critical to their company's success over the next two years. In addition, about three-quarters of the respondents believe all businesses will use AI in the next three years.
Deloitte notes that soon, just adopting AI tools won't be enough for companies to gain a competitive advantage. Instead, companies will have to find ways to apply AI more creatively and responsibly and think about how AI can enhance their products, services, business models and overall strategy.
In this Q&A, Fontana discusses what makes an AI-first company and offers advice on enterprises looking to begin incorporating more data and AI into their business strategy.
What is an AI-first strategy? How does a company put AI first?
Ash Fontana: You put AI first by putting it first in every conversation, so having the vocabulary to say, when talking about building a product feature, for example, what feedback data does it get us? Does it improve our core data learning loop? That's what it means to put AI first.
First, it's in conversations about people to ask is this person on the research side, as a data scientist? Are they going to be on the experimentation side? Are they going to be in the production side?
It's when you're purchasing products to ask questions like 'How does this fit into our model management loop?' It's when you're purchasing data to understand, 'How are we valuing this data? How is it going to be more determinative of the outcome? What predictive values can it have? How can we verify the veracity of this data?'
Having that language allows you to have every conversation going in the direction of 'How do we fundamentally improve our intelligence systems and the competitive advantage we get from data?' Without that language, you can't have everyone in your company asking those questions and speaking that way about all of the key areas within your company.
AI is a broad term -- what exactly do you mean when you say AI?
Fontana: It is an extremely broad term. I'll say a few things.
Intelligence is the ability to learn fast, and so, therefore, artificial intelligence is something that is not running on our own wetware, it's running on a computer. It's something that's different from our own form of intelligence, probably, that allows us to learn fast. To me, artificial intelligence, in the context of the AI-first company, is something that helps your company learn faster -- learn faster about what your customers want, about how your processes work, about where your supplies are coming from. That's what I mean here, a form of intelligence that helps you learn faster.
A lot of companies use analytics and AI; are they all AI-first companies?
Fontana: An AI-first company is about building an advantage through AI, where your core advantage is built through it. A company may use AI to automate things or save costs, but that can just be a one-time improvement, it doesn't necessarily compound over time.
An AI-first company is one that has a unique advantage over what I call data learning, which is something that is compounding and automatically using information over time from which you learn, and therefore automatically increasing your competitive advantage over time.
That's what an AI-first company does. It involves a lot more than just buying or using an AI product. It involves accumulating unique data on a constant basis, constantly improving your models, etc.
What type of company can become an AI-first company?
Fontana: They can all be AI-first. The thing is, there is a transition point. A lot of people have been collecting data, accidentally or deliberately. A lot of people have the expertise in house, either accidentally, like they just got a lot of highly quantitative people that do other things, or deliberately, as in they've hired people to build models. So, a lot of companies have some of the core components and data capabilities to process that data into information and talent to help them build models.
A lot of people have this, and it's just a question of bringing it all together to build a data learning effect.
How does someone start to bring everything together?
Fontana: I think it's fundamentally a strategy problem -- you've got to decide that this is something you want to build, this is the competitive advantage that you will have. Once you do that, then you start making decisions to allocate capital to get more data, to experimenting with models and whatnot.
So, it starts with strategy, but I think then it very quickly moves into tactics, it moves into experimentation. It moves into building AI into your culture, and just providing very basic tools for people to very quickly start experimenting with models and see if they can very quickly build some sort of effective predictive model.
It can be expensive for a company to put together an AI-first strategy. What should a company do if they don't have the money to spend on that?
Fontana: They can start with what I call lean AI, which is about narrowing in the questions you want to answer with a predictive model to one thing that you can run, or that you can figure out from one data set with one type of model and generate one output, one report. Once you can narrow things down to that, it's actually very cheap and tractable to run experiments.
Just do something that may indicate what you want to do is possible. There are so many tools out there for running basic statistical models or playing around with pretrained machine learning models, so that can actually be a really cheap and easy process. Using this technique called lean AI to narrow things down really effectively can allow companies with any reasonable level of resources to get started.
Do you have any suggestions on AI tools for newcomers?
Fontana: It depends on what you're trying to do. What I would generally say is a lot of the big tech companies have a lot of tools that are not really obvious now. They just have so many available.
So, what I would generally encourage people to do is go through the documentation offered by companies like Amazon, Google and Microsoft. Go through their machine learning product pages and read through the documentation to really understand what they've got out there. Because, you may be surprised to discover some of the really powerful stuff they have.
What companies can you identify now that are AI-first companies?
Fontana: I really think the original AI-first company is Google. From day one, they were using methods from AI to build a search product, and then they were really strategic every step of the way about gathering data that improves that core product.
To this day, they're really strategic about that. They give away certain products for free so they can collect certain data, they collect data in lots of different dimensions on customers. They do a lot to improve their core learning loop. They are phenomenal at this, and they are really the original AI-first company.
Now, there are a lot of companies out there that enable others to be AI-first. So, companies like Palantir, UiPath, SAS, Microsoft and others, they allow other companies to be AI-first companies, but they're not AI-first companies themselves, they're not fundamentally building these learning loops themselves.
Then there are lots of earlier stage companies that are AI-first companies, ones that are focused on getting the right data early on and building models around that, and then deploying a product in such a way where it is able to improve over time by getting data from customers.
So, you know, there are lots of examples out there of smaller companies, but of bigger companies? No. There really aren't that many, because there are very few that put AI in the beginning of every strategy decision.
Editor's note: This interview has been edited for clarity and conciseness.