When it comes to AI, companies typically test the waters proof of concepts or small-scale use cases, taking advantage of vendor offerings, such as new features in their existing SaaS platforms.
If things go well, they pursue another project, then another — and soon they’re relying on a sprawl of incompatible systems, competing data lakes, problems with cost overruns, duplication of efforts, and an inability to scale, not to mention privacy, compliance or ethics problems.
At some point, the benefits of AI become obvious enough, and the pain of continuing on their present path so acute, that companies step back to develop a cohesive strategy for an enterprise-wide AI-powered transformation.
“The tendency to get overwhelmed in individual technologies is not only drowning organizations in technical debt but discouraging them because they don’t see a path forward to sustainable and scalable AI,” said Traci Gusher, partner in data, analytics and artificial intelligence practice at KPMG.
Here’s a look at how organizations can ensure the shift from pilot projects to full-scale AI fluency goes right.
Start with core business priorities
To create a comprehensive AI-powered strategy for enterprise transformation, you have to start with the basics, Gusher says. “What are your top opportunities as an organization? What are your top challenges? What are your biggest risks?”
Then, look at how AI can impact those opportunities, challenges, and risks. “If you’re really trying to hit competitive advantage, then you have to look at it at a strategic level, across these big areas,” she says.
Companies should also avoid projects where the ROI isn’t going to be immediate enough.
“If I have an imminent risk to my organization that needs to be solved, AI probably isn’t the answer,” she says. “It takes time to get the data, time to do the learning, time to get the benefit from it. Long-term type of programs are what organizations should be looking at.”
Making these kinds of decisions requires a solid understanding of AI’s capabilities by senior executives and board members.
At UL, a 125-year-old company that tests electrical equipment, the AI transformation strategy starts at the top.
“The role of leadership can’t be underestimated,” says Scot Webster, SVP and chief technical officer at UL. “Leadership is important to help create a wider purpose across the enterprise.”
An enterprise-wide AI strategy can’t be run as a skunkworks project, he adds.
“We have 14,000 employees at UL, and I bet you that all 14,000 know that we’re focusing on digitization, on AI, on strategy 5.0,” he says. “The board is updated at board meetings, all the employees are updated. We have a center of excellence, specifically for the smaller RPA and bot applications.”
Shift your data strategy from silos to platforms
According to Anand Rao, partner and global AI leader at PricewaterhouseCoopers, once all the right people and leadership are in place, it’s time to focus on AI and data science platforms. Typically, platforms used for pilot projects don’t scale well. Instead, enterprise-wide AI requires a move to the cloud, which can provide additional benefits, including easier handling of large data lakes, better integration with external data sources and tools, and easy access to the latest AI technologies.
But to make the shift to the cloud, you have to get a handle on your data. And that means breaking down silos.
“The No. 1 problem is not focusing enough on your data strategy up front,” says Jamie Thomas, general manager of systems strategy and development at IBM. “That’s the biggest pitfall that we see.”
A comprehensive data management strategy not only covers collecting, organizing and analyzing data, but also infusing it with meaning and context that will make the data effective for AI. It also means having a plan for dealing with bias, which can arise if models are fed the wrong data, Thomas says.
In the past, teams at Fannie Mae were implementing AI and machine learning projects in silos, says Jay Rudrachar, director of enterprise monitoring, analytics and reporting at the mortgage financing giant. “But we had to take a step back. We said, ‘This isn’t working.'”
So, 18 months ago, the company began moving to a centralized, enterprise-wide AI strategy, one that includes multiple data science platforms and data lakes based on business needs, Rudrachar says. The consolidation of its AI strategy coincided with the organization’s shift to the cloud. Previously, the company ran its own data center in Urbana, Ill., with a second resiliency center in another location.
Now, as part of its AI strategy, Fannie Mae uses Tableau’s MicroStrategy AI platform on its customer-facing data to improve the speed and accuracy of its underwriting, servicing, and securitization business. For internal operations, Fannie Mae relies on Moogsoft and Splunk to analyze system logs and KPIs to improve performance and help resolve IT and cybersecurity problems. It also employs Blue Prism for business process automation.
“We’re already seeing the benefit of this investment,” Rudrachar says. “One benefit is breaking down the silos. For example, in IT, there are so many teams doing IT operations, at the platform level, operations level, business level — and they each maintained their own data sets that aren’t visible to other people. With machine learning and AI, we’re giving one pane of glass to everybody to see what’s going on in other areas.”
So if there’s an incident, people can find the cause of it, even if it’s in a different system or layer.
“That’s a game changer,” he says. “Every firm has to start doing this, otherwise efficiency in the IT operations space is a challenge.”
Spread your search for stakeholders
AI projects can easily create privacy, compliance, bias, or ethics problems for companies, and so it’s essential to enlist stakeholders from a wide range of lines of business.
“Often times, the business goals are what the company is focused on, and for good reason, since the business should be the driver,” says Michael Shortnacy, partner at King & Spalding, a global law firm. “But there can be a world of unintended consequences with AI.”
He suggests that companies also have legal, compliance, risk and other experts be part of the strategizing process. If AI is being used in the customer relationship process, there are a variety of legal and privacy pitfalls that companies need to be wary of. Overlooking privacy or security are common problems for proof-of-concept projects that can create problems when a company decides to scale up.
“It can induce massive compliance and regulatory risks, and create financial risks to enterprises,” says Jennifer Fernick, head of research at NCC Group. “By having security architects and other representatives from security teams at the table from the beginning, you can introduce security by design into the system.”
Getting all these stakeholders organized around the same table, instead of working in isolated groups scattered across various business units, can help ensure all viewpoints are covered.
“When you have piecemeal teams, you may have missing stakeholders,” Fernick says. “You may not have someone representing privacy; you may not have a competent security architect.”
Pick the right platforms and partners
The scalability of your AI strategy isn’t all about picking the right cloud platform for cost-effective growth. It can also mean taking a different approach to staffing and picking the right business partners.
Five years ago, Washington, D.C.-based Hello Tractor launched with plans of being the “Uber of tractors,” helping connect tractor owners with farmers in Nigeria and other emerging markets. “When we started out, we were very simple fleet management, with a crude booking component,” says Jehiel Oliver, the company’s CEO.
Then, a year ago, as a result of feedback from customers, Hello Tractor began looking at using AI to help tractor owners better deploy their tractors and drivers and reduce unplanned downtime and maintenance costs. “That would be a huge value-add for the tractor owners,” he says.
Like many companies in similar situations, Hello Tractor went out looking for AI experts.
“We hired some data scientists, and it was difficult,” Oliver says. “It’s a highly competitive market for talent. Not only did we have to find folks who were comfortable working with these advanced technologies, but also have to be comfortable with the business models and the locations we’re operating in. We’re not promising cushy offices in Silicon Valley. We work in agriculture in emerging markets.”
And when the right talent was available, it was expensive, he says. “We can’t pass on exorbitant costs to our customers. We have very price-sensitive customers.”
As a result, the company churned through several people who didn’t work out. “One person was actually very good but we couldn’t afford to keep him on the payroll,” Oliver says. “And he was just a fraction of what we needed.”
So Hello Tractor got creative about engaging with partners — and what it had to offer to those relationships.
“Sometimes, the traditional channels won’t work,” he says. “Sometimes, you have to be thoughtful about how you get a bigger company with deep capabilities interested in a startup working on Africa food security.”
By shifting its sourcing focus on its core mission, Hello Tractor could provide partners something they can’t get in working with just any organization: meaningful work in a new market.
“We think that there’s some intrinsic value that we can offer to our partners,” Oliver says. “Good companies realize that their workforce also wants to do meaningful work.”
Once Hello Tractor gave up on following the traditional route to finding employees, or finding business partners, and refocused on that mission, it was easier to move ahead, he says.
“We’re not going to help IBM reach their quarterly goals,” he says. “But across all our partnerships, from IBM to John Deere, they all see this as an opportunity to dip their toes into frontier markets that are high growth but also high risk. We can help them manage that, by being the first mover. If it works, it could be huge.”