How To Build AI Capabilities Across An Enterprise

How can businesses ensure that their embrace of AI isn’t just hype, packed with chatbots but few real transformations? How can they make the most of the AI opportunity and pivot quickly enough to respond to new AI-enabled threats? They can learn from the masters of making the most of disruption – venture investors.

Venture investors, who see AI impacts likely across much of their portfolio, are reacting quickly to build AI capabilities. As one of the leading investors in this space, Bessemer Venture Partners, says of start-ups, “far more will be AI-enabled than AI-first.” What are these investors doing?

Prosus provides an example. One of the world’s largest tech investors, with more than 90 companies that it operates either as total owners or with an equity stake, Prosus operates across diverse verticals such as fintech, edtech, classified ads, and food delivery. It’s seeking to build AI capabilities across this broad portfolio.

Euro Beinat is Prosus’ Global Head of AI and Data Science. In an interview, he shared how the group is embedding a common philosophy and approach across its many companies, and how it is organizing to build capabilities without dictating too much to each firm.

Building a Common AI Philosophy

At Prosus, there are three elements of an AI philosophy that Beinat spreads. First, AI should be implemented by design instead of retrofitting. The group seeks to be customer-centered and first understand the full use case before jumping to technologies. He says, “We look at the entire workflow and understand what are the technologies used now, what are the problems that you have to solve, and in what ways are machine learning, AI, and data science solutions better than the status quo. We have this approach and have repeated that in essentially every company in the group.”

Second, Prosus sets common guidelines for responsible and ethical use of AI. This helps to de-risk projects and avoids the need to debate these issues over-and-over again.

Third, it wants to adopt AI as fast as possible where it makes sense. And that leads to its organizational approaches.

Organizing to Drive AI Adoption

Prosus has a small team at its corporate center in Amsterdam, and then separate data science teams at its many portfolio companies. The teams can range from a small group of five to organizations numbering a few hundred. No matter their size, these teams can take advantage of several core Prosus capabilities.

Hiring is one major, common need, and the central group helps to attract and vet appropriate candidates. It also assists with researching and vetting emerging technologies, which arise at a rapid pace in the AI world. Additionally, it provides educational resources that the portfolio companies can tap.

One of the most significant contributions by the central group is to provide a common set of tools to leverage. Beinat explains, “It’s important that everyone can figure out by themselves where these new tools work and where they don’t. We organized a process of collective awareness and discovery so that everybody in the group can experiment by themselves. We have wrapped 20+ models into our AI Assistant, so that it can be used for instance for image generation or code generation. We created an environment which is safe and private, so you can experiment without the risk of data leaking. There are now six or seven thousand users of this across the group who use it every day.”

Learning from Analogies

Companies building AI capabilities can also learn from how firms in analogous situations operate. As Henning Trill, VP for Innovation Strategy at the life sciences firm Bayer explained in an interview for my book The Innovative Leader, “This is what I focused on to start:

1) inspire people that this is an opportunity

2) learn common methodologies

3) connect the people who are already driving these innovations across the company

4) collaborate and bring people together from around the organization to jumpstart new ideas and projects.”

Pacing the Change

Beinat’s concluding words of advice can temper expectations. He notes that, “There was this expectation that there is this miraculous set of tools that are ready to click and you just apply them and everything will be great. In fact, our experience is that while all of the models that are in production create at least incremental benefits, it’s a process of continuous improvement and systematic adoption that leads to big results at the end.”

He also points out that, as ever, the hardest part lies in the execution. He says, “It’s now really, really easy to create a service that looks good and is promising. But that is misleading, and it still takes the same amount of effort and time to build something that makes sense and scales.”