What a successful AI team really looks like

As more companies scale AI projects, turning proof-of-concepts into drivers of business transformation, a clearer picture of what it takes to succeed with real-world AI is taking shape.

When it comes to AI teams, a broader set of skills are required than previously known, with a particular need for people with experience in operations and in translating AI concepts into business terms and vice versa. In other words, AI success no longer hinges on just a group of data scientists anymore.

[ Cut through the hype with our practical guide to machine learning in business and find out the 10 signs you’re ready for AI — but might not succeed. | Get the latest insights with our CIO Daily newsletter. ]

In fact, enterprises need blended teams to succeed with AI, says Louise Herring, partner at McKinsey & Co. “If you look at the technical side, the emphasis is increasingly on how we can make sure we have production-ready code and we have elements available for reuse throughout the organization,” she says. “But the key area of emphasis that we see first of all is about translators: people who can make the connection between the business and the technical side.”

Here is a look at how several organizations are assembling AI teams to solve business issues — and how advances in AI technology are changing the baseline skills necessary for success. 

The fundamental roles of a successful AI team

Successful AI projects require team members with a broad range of skills, says Herring, including data scientists, data engineers, machine learning engineers, product owners, change management experts, and translators.

With proof of concepts and pilot projects, data challenges are different. Production-ready, scalable AI deployments need data in larger quantities, from more disparate sources, and at high speed.

“Data scientists and data engineers are the unsung heroes,” she says. “You have legacy landscapes and it’s not so easy to find the data and extract it — and ensure that the data pipeline is set up and ready to scale.”

According to a May report by IBM and Morning Consult, almost 90% of IT pros say being able to run AI projects wherever the data resides is key to the technology’s adoption. Data complexity and silos are the second biggest barriers to AI adoption, the report found, after lack of AI expertise.

Machine learning engineers take the code produced by data scientists and turn it into something ready for production.

Successful AI teams also need product owners and, depending on the degree of intervention required, change management experts, Herring says. But the key skill is that of the translator.

“It’s still something that I see missing in organizations,” she says. “But they need translators to maximize the value of their use case.”

Sometimes translators come from the AI side, for example, data scientists who have been embedded in business units.

“But they normally come from the business side,” she says. “They need to understand the business deeply, and it’s easier, in some ways, to learn the data science principles. Some organizations actually set up academies to teach data science skills, so they understand enough to engage confidently with data scientists and engineers.”

According to a survey released by Deloitte last year, half of the most in-demand AI skills are related to connecting AI projects with business needs. These “translators” include business leaders well-versed in AI, change management experts, user experience designers, and subject matter experts.

On the AI side, according to Deloitte, the most in-demand skills are AI researchers, software developers, data scientists, and project managers.

The power of blended teams

Online marketing company Urban Airship provides a textbook example of the shifts under way in how successful organizations approach AI. When the company first began thinking about using artificial intelligence ten years ago, it hired a PhD.

“The first machine learning model we introduced was around influence,” says Mike Herrick, the company’s senior vice president of product and engineering. It’s easy to track whether a person clicks on a link in their email. But tracking whether they visit the site later, and through some other channel, is a lot harder, he says, and that’s where machine learning came in.

The company then added predictive intelligence to figure out the most optimal time to contact a particular customer and how often customers should be contacted. The latest addition to Urban Airship’s platform is about using AI to kick off and manage a multistep customer journey.

Today, Herrick says, at least 40% of new business deals come from that tool. “It’s been huge. We’ve got unique capabilities that differentiate us against our competition. And it helps us retain our customers because we’re providing value to them.”

The skills required for the company’s AI projects include not just data science skills, but also product management, user interface design, software engineering, and product marketing, he says. “AI and ML really does take a cross-functional team to deliver on this type of technology. It’s been borne out by our experiences.”

An AI team composed of different kinds of experts is a scalable strategy, he adds. “Sometimes you can get someone who has all the skills, but they’re exceedingly rare. And if they have all the skills, they can’t do all the things.”

Having team members who understand the business cases well also helped the company adapt its products to be more useful in providing the analysis in a way that its customers needed.

The rise of the translator

For two decades, Company Nurse has been helping companies and educational institutions handle workplace injuries, with a staff of medical professionals just a phone call away.

The company now handles 100,000 health-related transactions a week. With the pandemic, Company Nurse began offering digital screening solutions.

“We’re actually screening for health symptoms on a daily basis,” says Henry Svendblad, CTO at Company Nurse. “And a lot of schools are using us not only for their teachers but also students, as part of their ‘return to school safely’ programs.”

In the past year, the company has begun applying AI to some of its business challenges. AI projects include a system that classifies sensitive healthcare documents and a speech-to-text conversion system for the call center.

Using AI has enabled the company to classify millions of documents, which will help Company Nurse to implement better cybersecurity measures. And in the call center, it has seen a decrease of more than 10% in average handle times. Future plans include using intelligence to compare health outcomes with the information initially provided on the calls, to identify potential fraud, and to help new agents become more productive.

Company Nurse uses outside vendors to help with this, including Concentric, Genesys and Salesforce’s Einstein platform. The company also has a data team to prep its data for ingestion into the AI systems, as well as subject matter experts.

But a critical role is that of bridging the gap between the AI technologies and the business case. “That’s my function as a CTO,” says Svendblad. “I try to marry, ‘How does this impact our business?’ with ‘Will this technology actually produce results?’”

And in pulling together internal and external resources, data sets, and third-party tools, Svendblad is also taking on a newly emerging AI role: solution architect.

“The solution architects are thinking about the ML and AI technologies that they need to solve problems,” says Dan Simion, vice president of AI and analytics at Capgemini. “What’s happening right now is there are a lot of solutions out there, a ton of technologies.”

Solution architects don’t just figure out which technology to use, he says, “but also how the technologies work with each other. They put the puzzle together.”

Change management and the agile business

The biggest mistake people make is thinking that AI is a technology project, says Tamim Saleh, senior partner at McKinsey & Co.

“They treat it as an IT project,” he says. “They think that they can get a small group of technologists and mathematicians together and magic will come out. You get a black box and give it to the business and great things will happen.”

It doesn’t work that way. “Around 50% of the effort in any AI project is people,” he says. “Change management. Training.”

For example, in one recent project, a client in the steel industry wanted to help improve demand forecasts. “Initially the forecasters were not interested,” he says, adding that it took getting them involved in creating the model to help them realize that it didn’t replace them. “It enhanced them,” he says. “But to get there was a combination of getting them involved in the design and the execution of the project.”

If the model is a black box, where the users don’t understand the logic behind the algorithm, they won’t adopt it, he says. “And you end up doing pilots and spending money and not scaling solutions.”

Or a company could have a pilot in one part of the business, but it doesn’t spread to other geographies or product groups.

One banking client trained 1,800 people to be translators, Saleh says. “The CEO was a visionary, and as a result, they dramatically accelerated the deployment of AI at the bank.”

Successful AI projects may also require businesses to restructure some of their operations. One utility client, for example, created a fantastic marketing model that can do 500 campaigns in an hour, but the business was structured around separate campaign management and marketing teams, so it wasn’t able to take advantage of the speed that AI now made possible, Saleh says.

“The business had to change,” he says. “They themselves had to become agile, so that they were able to make decisions really fast.”

When companies roll out AI applications, they often have to ask themselves whether they’re organized in a way that lets them make full use of the technology. “And almost always the company comes to the conclusion that they can organize very differently and make decisions much faster,” Saleh says.

Include outside expertise

Blended teams are critical for companies that want to operationalize AI, says Mark Beccue, principal analyst for AI and NLP at Omdia. But many skills are hard to find.

“Data scientists, especially experienced ones, are scarce and will remain scarce for a long time,” he says. “So even with building teams that combine other skill sets, there is a trend towards outsourced AI.”

As a result, organizations such as Company Nurse are looking for outside vendors to provide tools, platforms, and expertise. And the tools are getting better all the time. Data tagging and cleaning, for example, can be outsourced, Beccue says, and do-it-yourself cloud-based AI platforms and tools, as well as no-code options, are helping to democratize AI.

SaaS and end-to-end solutions providers are also bringing AI capabilities into the enterprise, he says. Salesforce, Adobe, and Oracle, for example, are building AI capabilities into their tools, including predictive analytics and virtual assistants. And companies such as Nuance, Interactions, and IPSoft are offering virtual assistants that require no in-house data science expertise.

“For many AI use cases, this approach will continue to make sense for the foreseeable future,” he says.

Some businesses will see big gains using products and components from outside vendors, says Scott Likens, emerging technology leader at PricewaterhouseCoopers. “It may help take costs out, make decisions faster, and help make decisions they couldn’t make before.”

Building an internal AI team is expensive, he adds, given the current marketplace for AI and ML talent. But making this investment can help companies differentiate themselves in the market, and optimize their use of AI to suit their particular business needs.

Organizations can accelerate building their AI team by involving less experienced people, says Ken Seier, chief architect for data and AI at consulting firm Insight.

That’s what his company is doing, he says. “We are getting supersmart internally about how to use more and more junior resources in our own organization.”

One strategy is to adopt a mentorship model, he says.

“I’ll run a project, and have another data scientist shadow it,” he says. “So your most talented are assisting the next tier down and you wind up with a great bootstrapping motion, where your teammates are growing as fast as they can and there’s a backup at every tier.”