Artificial intelligence has hit the mainstream. Across industries, companies have rolled out successful proofs-of-concept and have even been successful in deploying AI in production. Some organizations have even operationalized their AI and machine learning strategies, with projects proliferating across the enterprise, complete with best practices and pipelines. Today, companies at the leading edge of the AI maturity curve are making use of AI at scale.
This overall maturation of how AI is deployed in enterprises is shifting how companies view the strategic value of AI — and where they hope to see its benefits realized. Here is a look at 10 AI enterprise strategy trends that industry experts are seeing unfolding today.
1. AI gets down to business
In the early days of artificial intelligence, projects were entirely driven by data scientists. They had the data and the algorithms, and they were given latitude to look for ways to apply their new tools to business problems. Sometimes, they succeeded. Today that dynamic has flipped around.
Business leaders have learned from the examples of successful projects and are more educated about what AI can do for them. As a result, companies are now less willing to invest in proofs-of-concept with unclear business value, a trend that sees business units increasingly in the driver’s seat for AI adoption.
“When I see companies doing AI well, it’s business-driven,” says Alex Singla, global leader for QuantumBlack at McKinsey & Co. “AI and IT are there to help them solve the problem, but it’s not technology pushing the solution out. It’s business taking a lead, saying, ‘I was part of the solution, I believe in this, this is the right answer.’”
Honeywell, for example, is using AI throughout its internal operations and is building it into customer-facing products and services, says Sheila Jordan, the company’s chief digital technology officer.
“We’re very connected to the business,” she says. “We’re driven by value. It’s going to be customer-facing value. Internal value.”
2. AI pervades the enterprise
When Jordan came to Honeywell two years ago, her first big project was to implement a data warehouse strategy to bring together all transaction data from all sources.
“Every function, every business unit, has a digital agenda,” she says. For example, Honeywell has digitized all its contracts. That’s more than 100,000 contracts total, she says, noting that this gives the company a wealth of data to use to help build AI solutions for almost any function area.
For example, with AI, all Honeywell contracts can now be reviewed automatically for areas where they are affected by inflation or pricing issues, Jordan says. “There’s no way any human being can go through 100,000 contracts.”
Similarly, with complete inventory data, Honeywell is now able to understand which inventory is scrap and which is reusable, and can thereby make smart decisions about managing raw materials more efficiently, Jordan says.
“We’re seeing AI pop up in every function,” she says. “In finance, in legal, in engineering, in supply chains, and of course in IT.”
3. Supercharging automation with AI
This is Honeywell’s third year into an aggressive automation program. If there’s a repetitive task, the company will try to automate it. “We probably have 100 projects this year,” Jordan says. “These are tasks that we’re automating across the entire global company.”
And Honeywell is working to make those automations more intelligent, she adds. “We’re going to be inserting more AI in more of these automated bots,” she says. “It’s about the automated bot getting smarter.”
Another company that started with basic, rules-based automations is Booz Allen Hamilton. Now the company is progressing to integrating AI and machine learning into those automations to make them applicable to a broader range of tasks, says Justin Neroda, vice president in Booz Allen’s AI practice.
People start with the simplest automations, he says. “Then they ask themselves, ‘What else can I automate?’ And they find that it needs AI and ML.”
AI-powered automations can help companies deal with staffing shortages or high volumes of work, he says. “Or half of the task can be automated and then people can do the hard part of it.”
4. Baking in AI for bigger benefits
There’s a major change-management component to doing AI at scale, says McKinsey’s Singla. It requires understanding how people are going to be using it, and that doesn’t come from the technology people working alone, but by a combination of technology people and subject matter and business experts, he says.
“If I have to get the adjuster and tell them to go to three different applications for AI, the odds of them applying it are zilch,” he says. “But the more it’s automatically baked into the workflow, the more we increase the probability of success. The less I have to change someone’s behavior, the more likely I am to grab adoption.”
5. AI strategies take federated turn
After companies are successful at initial proofs of concept, they often build AI centers of excellence to operationalize the technology and build talent, expertise, and best practices. But once a company reaches a level of critical mass, then it makes sense to break up some of these centers of excellence and federate AI, moving experts directly into the business units where they are needed most.
“For those companies that are less mature, there is value in having a center of excellence that is housing talent and learning across the institution,” says McKinsey’s Singla. “Without that, companies usually don’t have the ability to scale. Talented people want to be with other like-minded people. And less experienced people benefit from being in a center of excellence because they can grow or learn.”
Distributing them too early would dilute their impact and reduce a company’s ability to iterate and duplicate successful projects across multiple business lines.
“But as you get to a layer of maturity and scale, longer-term, the benefit of technologists having both a deep AI expertise and domain expertise is a real home run,” he says. “But only when you have scale.”
Business problems are distributed, says Amol Ajgaonkar, distinguished engineer at Insight.
“The business problems aren’t in one place, so you cannot expect to have centralized AI deployments,” he says. “They have to be distributed as well. But you do need to have a centralized AI strategy that is tied to a business impact.”
Or multiple business impacts, he adds, such as revenue, cost savings, or marketing positioning.
Like many other companies, Booz Allen Hamilton started with a core AI group. “But in the last year we’ve really been pushing it out,” says Justin Neroda, vice president in Booz Allen Hamilton’s AI practice. “We have sub-cells through that firm that have those experts in AI. But you have to build to a critical mass before you spread it out or it will all fall apart.”
“That is something that we’ve seen within our own organizations and the clients that we work with,” he adds.
6. AI triggers business process transformation
When companies first start using AI, they often look for individual steps in business processes where AI can make a difference. “You break down the process into pieces, digitize each piece, and put in the AI to make it efficient,” says Sanjay Srivastava, chief digital officer at Genpact. “But at the end of the day, the process itself is the same. Each part of it is better, faster, cheaper — but the process itself doesn’t change.”
But AI also has the potential to fundamentally change business processes, he says. For example, Genpact does a great deal of accounts processing work for clients.
“When we apply AI to invoices, we can tell which invoices are going to be disputed,” he says. “We can figure out which part of the portfolio has the highest risk.”
With the predictive powers available with AI, the entire process can be restructured, he says. “When you apply AI, you can think about the end-to-end value chain and completely re-engineer it.”
7. MLOps gets real
According to a McKinsey report released at the end of 2021, one of the factors that distinguishes companies that get the biggest earnings boost from AI is their use of MLOps.
This is the next big trend in AI, says Carmen Fontana, IEEE member, and cloud and emerging tech practice lead at Augment Therapy, a pediatric physical therapy technology company. Fontana was previously the practice lead for cloud and emerging technology at Centric Consulting.
The goal is to bring machine learning from theory into production, she says. “Two, three years ago, this was a burgeoning field and people were thinking that they had to do it,” she says. “But we didn’t see it a lot in practice.” Today, however, she’s seeing established tools and methodologies that enable organizations to become more rigorous in how they train, deploy, and monitor AI models.
“That goes a long way to making AI and machine learning institutionalized,” she says. “I observed all of that at our clients. The market has changed significantly.”
8. Enterprises lay down AI pipelines
Booz Allen Hamilton currently has about 150 different AI projects with its clients, says Booz Allen’s Neroda. But over the past year, the company has begun moving away from that one-off model.
“Over the past year and a half we’ve been investing in modular capabilities and end-to-end pipelines,” he says.
Successful AI requires more than just a working model. There’s a whole process that’s required to maintain the model over time as the data changes and as the models get continually refined, he says.
“The biggest challenge is how you tie all the tools together,” he says. “We’ve been doing work to standardize that and to build reusable pieces to use across projects.”
9. Organizations look to build AI trust
As employees and executives get more familiar with AI, they are increasingly putting their faith in it to make business-critical decisions — even when those decisions go against human gut instincts.
Michael Feindt, strategic advisor and founder at Blue Yonder, recently worked with a large British food retailer struggling with pandemic-related supply chain issues. When the company used manual processes to manage its supply chain, there were a lot of empty shelves, he says. Plus, there was a shortage of people with the knowledge and the ability and the willingness to do the work.
Automated, AI-powered systems could offer reduced costs and better performance. When the pandemic hit, however, people wanted to shut off the automatic systems. “But then they saw that the automatic systems could adapt much faster than humans could,” he says.
So instead of shutting down the systems, the company expanded to include not just the stores but also distribution centers. The result was both fewer empty shelves and less food waste to throw out. Plus, store managers could stop spending two hours a day fine-tuning their orders and instead spend more time improving customer satisfaction.
There are also other ways to build trust in AI, says Feind. “Some people are critical and don’t have the trust that the AI can make as good a decisions as they can, with their years of experience,” he says. Adding explainability can help alleviate some of these concerns. Explainable AI is when the system explains to human users what factors went into the decision it made.
10. New business model possibilities arise
In some areas, AI is starting to create opportunities that never existed before. Autonomous vehicles, for example, have the potential to transform societies and create entirely new kinds of businesses. But AI-powered business transformations can happen at a smaller scale, as well.
For example, a bank that requires human review cannot afford to offer small loans. The cost of researching and processing them would be higher than any interest revenues the bank could earn. But if AI was used to evaluate and process, the smaller loans would allow the bank to serve entirely new groups of customers without having to charge exorbitant rates.
“These use cases are still not as prevalent,” says Jai Das, president and partner at Sapphire Ventures. “They fundamentally change the way we do business, and enterprises don’t change that quickly.”
The tide will start to shift once AI and ML become tools used by every knowledge worker in the company, he says.
“We’re not there yet. It’s probably another five years until everyone will use AI and ML to do their job.”