How to Build an AI-Ready Organization

October 23, 2019

In our “How to Build an AI-ready Organization” Master Class, Joe Brown and Mike Stringer discussed how companies can create the right strategy to deploy artificial intelligence in their organization. During the call, they focused on three key questions:

  • Where should teams begin when they want to work with AI?
  • How can teams grow their AI capabilities? 
  • How can organizations scale their AI solutions?

Brown is a Senior Portfolio Director at IDEO San Francisco. In his role, he helps companies reignite their cultures in order to sustainably launch new products and businesses. Stringer is the executive design director at IDEO Chicago and has a data science background. Download the slides from the presentation.

Brown explained that IDEO works to design not just products, but also services and organizations. As an example product, he cited the Willow Breast Pump; as a service, he mentioned the PillPack Internet-based prescription delivery service; and as an organization, he touched on IDEO’s work with Ford to get employees more comfortable with experimenting around new kinds of mobility, not just automobiles.

Survey Data Says…

In a pair of surveys conducted during the webcast, participants were asked about their organization’s current status when it comes to AI. More than half (53 percent) say their organization is “figuring it out.” Just 4 percent said their organization was “living and breathing AI.” Forty-three percent described their organization as either “completely lost” or “a little confused.”

Participants were also asked about where they are applying AI in their organizations. The largest segment, 48 percent, said they are building AI into products and services; 30 percent said they are applying it to “everything we do”; 22 percent said they are mostly using AI to improve strategy and operations.

Why AI Initiatives Fail

When organizations feel they are behind the times when it comes to AI, Stringer says, they often react with an “infrastructure-first” approach. But think of data infrastructure as plumbing. It’s the equivalent of running pipes to any room, so anyone can install a sink or toilet wherever they want. But do you need a toilet in every room? Or dangling from the ceiling? That approach leads to a high failure rate of AI-related initiatives. 

To improve the odds, Stringer suggests, ask the three questions above. But question #0 is “What exactly are we talking about here?” AI is being used in 2019 to refer to “almost anything digital,” he says — from business intelligence to data visualization to robotic process automation.

“If you’re feeling confused about what AI means, you’re not alone,” he says. It’s essential to have a conversation with senior leaders about what outcomes you’re aiming for, so that AI isn’t seen as a magical force that solves every problem. 

Stringer says that organizations can rely on a framework he calls “Sense, Act, and Learn.” And he defines AI, in this context, as “designing the ability to adapt to change.” That ability is “only going to get more important over time,” he says.

Start Small

You can start small, Stringer says. Think of the first phase being chalking up a few successes — how can you use AI, data, and algorithms to help the organization in some way. 

The next phase is resisting the urge to prematurely scale as you build on the early successes, and create more fluency within the organization.

Then, when appropriate, build the infrastructure to scale the AI capability more broadly throughout the organization.

As an example of starting big, rather than small, Stringer cited a challenge that Netflix tossed out in 2006, to help improve its movie recommendation algorithm. A 10 percent improvement would pay out a $1 million prize. While the final result took three years for a team to develop, it was too complicated to implement. Netflix wound up implementing a solution that was developed in the first few weeks — even though it didn’t hit the 10 percent threshold. 

As an example of starting small, Stringer mentioned work that IDEO did with Procter & Gamble, to help identify internal experts with knowledge about particular topic areas. Rather than approaching it as a data problem, IDEO talked with people about how teams come together and how they seek expertise. After many pencil-and-paper iterations, they hacked together an approach to finding people with specific knowledge — but required almost no infrastructure. The end result took about 24 weeks to create, rather than several years. 

How to Find Early Use Cases

On one slide, Stringer and Brown collected their advice related to finding early use cases:

  • Start small.
  • Build one generalist team, led by a translator.
  • Find one specific business or customer challenge, not a “data challenge.”
  • Consider data as a partial resource for a solution. (Asking “what can we do with this data?” is a recipe for useless outputs.)
  • Prototype to learn. Build as little as possible to test how you created value.
  • Rinse and repeat.

In hiring data scientists, Stringer said it is important to “resist the urge to specialize too soon,” and instead hire generalists for the early stages of projects. “It’s only when you’re scaling that you [should be] starting to specialize.”

When It’s Time to Scale

When it’s time to scale early successes, Stringer and Brown advise companies to:

  • Stay small.
  • Share the results of your pioneer projects to generate demand.
  • Bring specialists in to expand upon your pioneer projects.
  • Launch new pioneer projects into different corners of the business.
  • Gather results to build a playbook for AI in your business.

They cited IBM’s Watson Health initiative as an example of not scaling well. Early success with Watson in general AI realms, like winning at the game show “Jeopardy,” may have created hubris. The company poured money into health care applications for Watson, helping doctors diagnose diseases, for example. They brought up the agricultural giant Cargill as an example of chalking up “small wins” before scaling — spending several years looking for early applications for AI within their business, like gathering data about how much feed shrimp are getting.

What to Do Next

As a final set of recommendations about how organizations can expand AI capabilities further, Stringer and Brown say that once progress is starting to become evident, companies should:

  • Develop and deliver formal trainings to collaborating teams.
  • Formalize a set of data science processes, metrics, and tools.
  • Pool data science teams to determine what infrastructure changes are likely to deliver the greatest value for the greatest number of teams. 
  • Work with each business unit to develop AI projects and plans. 

In the webcast’s final poll, participants said that “increased efficiency” is the primary yardstick by which they are measuring the results of their AI investment.