by Maria Korolov

8 key roles of successful AI projects

Mar 12, 2019
AnalyticsArtificial IntelligenceData Science

As enterprises further develop artificial intelligence projects, they are finding that some roles are essential to success. But the right talent can be hard to find.

virtual brain / digital mind / artificial intelligence / machine learning / neural network
Credit: MetamorWorks / Getty Images

Artificial intelligence offers ample opportunities to reap business value. When done right, AI can help improve sales, optimize operations, and free up staff for higher-value work. It can help reduce costs and empower organizations to create new products and pursue new markets.

And enterprises are diving in. According to a recent Deloitte survey, 55 percent of IT executives say their companies launched six or more AI-related pilot projects in 2018, up from 35 percent in 2017. More than a third have invested over $5 million in cognitive technologies, and 56 percent expect AI to transform their companies within the next three years.

But getting there isn’t easy, and certain key skills are required — but hard to find. Here we take a look at eight key roles for AI success, according to those making early forays into artificial intelligence for business.

AI researchers

It might seem counterproductive for the average enterprise to get involved in research. After all, AI researchers are often the PhDs who perform fundamental research that could, someday, lead to a breakthrough in machines’ abilities to think. Plus, going after AI researchers means competing against universities and tech giants like Google and Microsoft for near unicorns who might not immediately conjure business benefits.

But there’s always hope that a breakthrough will catapult them into the lead. This promise alone may speak to the high demand for AI researchers. According to Deloitte’s survey, 30 percent of IT execs view finding AI researchers among their top priorities, more than any other role.

“People want that shiny object,” says Vivek Katyal, global leader for analytics and data risk at Deloitte Risk and Financial Advisory. “But will that shiny object make a difference to what they’re truly after?” Unless a company wants to be the next Facebook, he says, maybe not.

But many business executives making funding decisions don’t understand the difference between AI research and AI applications, he says. “It’s not the data scientists funding these projects.”

For those companies where AI is critical to their core business, however, research is not a luxury, but a necessity. AppTek, for example, was founded around 30 years ago as a speech recognition company. The entire field of speech recognition has been transformed by AI, and AppTek has had to invest in research to keep up. For example, its latest published research focuses on identifying different speakers in a conversation.

“That’s a real commercial need,” says Mike Veronis, the company’s chief revenue officer. “We did that to solve the problem, and to push capabilities.”

AI software developers

AI software developers take fundamental research, such as the latest developments in deep learning or generative adversarial networks, and turn them into usable products. Some companies leave this work up to the big vendors, relying on commercial platforms rather than developing their own approaches to AI. But even if companies are using known AI techniques, they might still want to build their own platforms, says Deloitte’s Katyal. This in part may explain the high demand for AI software developers, a top priority for 28 percent of respondents to Deloitte’s survey.

One reason to build your own is the “black box” problem of current AI frameworks. Without the ability to see the source code of off-the-shelf products, some companies, especially in regulated areas like finance or healthcare, might rather pursue their own course.

“Maybe I should develop something on my own, where I know what I’ve built, I own the code, I control everything about it,” Katyal says. “That discussion is very prevalent.” When they build their own AI software, they can also get a better understanding of the built-in biases of the tools, he adds.

That is the case for AppTek as well. Instead of having a black box commercial system that can’t easily be tuned, it gets a product that can be customized as needed, in addition to having unique features based on the company’s own research. “We can adapt and train and continuously improve the speech recognition engine,” says AppTek’s Veronis.

Data scientists

When companies think about overcoming AI challenges, they typically think of creating new AI algorithms, Katyal says. But they would probably get more value from improving their data. “That is the usual barrier to functional AI,” he says.

This makes data scientists the most important AI role of all, according to Katyal. Sought by 24 percent of respondents, data scientists prep a company’s data for use in AI systems. They also identify the data a company needs to meet its goals — data that is either generated internally or gathered from third parties. Data scientists can also spot when data is missing, know when there isn’t enough data of a particular type, and recognize when a data set is biased or out of date.

They are also the ones who identify the right algorithms to use on their data sets, train and tune those algorithms, and work with subject matter experts to validate the results.

“In the old days, they would have been advanced statisticians,” says Katyal. “They are the users of the AI research and the AI software.”

Data scientists are at the heart of Sumitomo Mitsui Banking Corp.’s recent AI projects. SMBC, a global financial company and the second largest bank in Japan by assets, is using AI to improve customer service in its data centers, to make it easier for employees to find information, and to better identify potential corporate customers.

The bank already had a data management department and data scientists on staff, says Akinobu Funayama, the bank’s executive director. At first, the data scientists would manually set up use cases, identify the data points most relevant to those use cases, and create the algorithms to analyze the data. For example, when scoring potential new customers for profitability, data scientists would look at thousands of factors to see if any turn out useful.

The entire process would take two to three months per use case, translating into 10 to 15 use cases per year. Using technology from dotData to help identify data points most useful for creating new algorithms, SMBC has reduced the time it takes to create a new model to just a few hours. This has increased the number of use cases the bank can tackle to about a 100 per year, enabling it to apply AI to more areas in the bank, including finance, treasury, and compliance.

“We are working on improving the performance of the whole group,” says Funayama.

The data scientists are still critical to the process, he says, but instead of doing repetitive feature engineering work, they are now tackling a much wider array of business use cases of AI technology.

User experience designers

As AI is incorporated into more products and services, user experience design is becoming increasingly important. Instead of opening menus or clicking buttons, people now expect to be able to ask plain-English questions, or have applications deduce what they need from context.

“We’ve always thought of user experience as being web-driven or mobile-driven,” says Brandon Ebken, CTO at Insight, Tempe, Ariz.-based technology consulting firm. “In the AI world, we’re interfacing with chatbots or Siri or Cortana, with voice.” It’s created a whole new type of user experience design, he says, and is a critical piece when creating new AI-powered tools.

“The connection between AI-powered things and human experience is evolving,” Deloitte’s Katyal agrees. “I think that’s the next revolution, one that we’re already starting to see.”

As new tools are created, people have to be able to use them, and that can require new kinds of interfaces, as well as accompanying changes in the way that an application or business processes are structured.

To find people with these skills, companies should look for experts in customer service, he says.

Change management experts

Change management is the single most overlooked aspect of AI deployments, says Deloitte’s Katyal. And it’s not just enterprise employees that benefit from change management, but also users and customers. “It’s the hardest thing,” Katyal adds. “This is an area most ignored and undervalued in the enterprise.”

Still, change management experts remain in high demand, a top needed skill for 22 percent of respondents to Deloitte’s survey. AI projects can have a large impact on knowledge workers, who may refuse to accept AI recommendations if they have not been involved in the development of the solution, according to Deloitte.

“The fundamentals of fostering organizational change can get lost amid excitement around pilots, grassroots experiments, and vendor-driven hype,” says the Deloitte report.

Moreover, 63 percent of IT managers surveyed said that, to cut costs, their company wants to use AI to automate as many jobs as possible — further underscoring the need for change management expertise.

Project managers

Many AI projects are plagued with issues because they are often not managed with the same rigor that companies use with more mature technologies. Project managers capable of leading AI implementations can help integrate AI into a company’s roles and processes and help measure and prove business value, a top-three challenge for 39 percent and 30 percent of Deloitte survey respondents, respectively. They can also deal with skill shortages in other areas related to AI.

It’s difficult enough to find data scientists, much less data scientists who are also software engineers, user interface designers, security professionals, and subject matter experts. Because of this, AI projects include complex teams of people, says Marty Young, managing director at Slalom Build, the technology consulting division of Seattle-based Slalom.

Project managers are needed to wrangle all these roles. Moreover, project managers will help multi-disciplinary teams move AI from experimental pilot projects to becoming just another aspect of software engineering and the software lifecycle, says Steve Herrod, managing director at General Catalyst Partners, a venture capital firm focusing on high-tech startups. Herrod was previously CTO at VMware.

“We shouldn’t lose sight of the project and program managers that need to understand the unique aspects of the models and fit them into the broader software releases that they must be part of,” he adds.

As the field advances, there will be an even broader range of roles that will be relevant, such as people to handle audit and certification-related questions, Herrod says.

That’s going to create more work, and more need, for project managers.

Business leaders to interpret AI results

Even when a company uses outside vendors for much of their AI functionality, having in-house business expertise is critical.

That was the case for Spoton Logistics, an India-based shipping company looking to use AI to help with customer service, sentiment analysis, and automation in the finance department. For example, one specific use case is to solve the company’s “first mile” and “last mile” address problem.

“India addresses are not standard,” says Satya Pal, the company’s head of business engineering. It only gets worse when the company is working with addresses that have not been fully filled in. “This takes away the possibility of central planning and vehicle utilization.”

The company decided to use outside vendors for much of the work, instead of building the technology in-house. However, the business leaders who were needed to interpret the AI results were on the company’s internal team. They had business knowledge of the specific problem the company was trying to solve, and an understanding of various AI models and frameworks, he says. For example, they were able to understand the applications of classification models versus reinforcement learning, and supervised versus unsupervised learning.

“Generally, they were from a computer science background with Python knowledge,” he says. Some additional training was required, but this was usually independent research as well as AI-related online courses.

This allowed them to determine which AI approach was best suited to solve particular products and validate progress.

Subject matter experts

Because off-the-shelf AI tools don’t always work for all use cases, subject matter experts are key. Take, for example, product recommendation engines, which are typically designed around the needs of online retailers, says Michael Rigney, SVP of client solutions at EnergySavvy, a software company focusing on the utility industry.

Online retailers collect data about the shopping habits of their customers and can compare that to the shopping habits of other customers. But past purchases aren’t useful metrics for those who get electricity from the local utility company. Here, expertise from companies such as EnergySavvy can help.

“We know how to identify which customers are benefitting from energy-efficiency projects, how much they’re benefitting, and who else would be similar to those customers and would benefit as well,” says Rigney. That has helped EnergySavvy to serve clients such as NationalGrid in Massachusetts, he says.

The new AI capabilities are responsible for the vast majority of the company’s recent revenue growth, says EnergySavvy’s VP of marketing Ryan Warren. “Most of our new customer growth, the lines of business that are the future of our company, are fundamentally tied to technologies that are underpinned by AI.”