Enterprise executives see big potential in artificial intelligence, and they're now devising ways to use AI to boost worker productivity, better engage customers and develop new business opportunities.
Nearly half of business leaders responding to McKinsey's 2020 global survey on AI said they have adopted the technology for at least one functional area, including product development, manufacturing, supply chains, marketing and sales, service operations, corporate finance and human resources.
Yet analysts, executive advisors and AI experts indicate that organizations are just beginning to tap into AI's potential. Most organizations, according to Gartner, are only at the first or second level in the research firm's hierarchical five-level AI maturity model, with companies expressing interest, starting to formulate ideas or experimenting with AI but not yet building strategies.
"Leaders are just now looking at the business challenges they have and asking, 'Is AI a solution for those problems?'" said Beena Ammanath, executive director of Global Deloitte AI Institute. Before moving forward with their AI initiatives, executives must first formulate a strategy, according to experts. They must create a roadmap that addresses the myriad issues involved in successful AI deployments as well as one that articulates how their AI vision supports the organization's overall goals.
"You need to open up your thinking to ask, 'What's possible?' but you also need to let your business strategy inform what your AI strategy should be," advised Seth Earley, author of The AI-Powered Enterprise and founder and CEO of Earley Information Science.
Experts offered 10 steps for building a successful AI strategy for a wide spectrum of applications in industry, government and education.
1. Establish a center of excellence
Implementing successful AI initiatives requires a multi-role team or a collection of these teams that can bring the necessary technical skills and business savvy to every project, said Gartner analyst Whit Andrews. Teams should include AI specialists, IT leaders and business-side workers.
The size and number of the teams required depends on the organization and the scope of its AI initiatives. Executives should adjust the number of teams and their composition as projects emerge.
A company that's developing several AI use cases, for example, may need separate teams for each use case with no overlapping personnel. Another enterprise may be able to assign the same team to more than one initiative without overloading team members. And yet another organization may find that it needs to configure multiple teams but can assign some workers to more than one team because their skills won't be needed full time on each team.
2. Set business priorities and identify opportunities
Enterprise executives should identify the business processes where AI can add value, ensure that those processes are already in good shape, determine the potential returns that AI could deliver in each area, and designate three to five areas with the highest potential returns as top priorities.
Executives could develop a use case ladder, Andrews suggested, to identify how each AI project sets the stage for a specific follow-on initiative. "So, whatever you do is a setup for the next shot," he explained. "It should set you up to advance your skills and capacity. Then you're going to be more successful on that next project because X percentage of the project is already done."
Successful organizations also find ways to support AI initiatives that don't just solve existing challenges but capitalize on the technology in its many forms -- machine learning, deep learning, natural language processing, robotics and automation software -- to enter new markets, create new products and services, and drive economic growth.
"What do you want your future state to be?" Early asked. "What do you want to do to leap ahead?"
3. Select and commit to a limited number of projects
Executives must select several promising AI projects and commit to delivering minimally viable projects -- not just pilots or proofs of concept.
Businesses often fail in their AI initiatives, Andrews said, because they start projects, move them into a pilot or proof of concept, and abandon them when they deliver limited results. Instead, they should refine the project as they move it into full production, where it's more likely to deliver valuable returns.
4. Assign executive-level project sponsors
An executive within a business function should have responsibility for each corresponding AI initiative, and the project's budget should be at the C-level. Executive-level accountability helps keep the initiative focused on delivering returns that the organization values.
The C-suite, Andrews said, "will want to see that the AI projects will do what you said it will do." The CIO should be the key stakeholder given the technology components involved with AI projects.
5. Determine and fill any skills gaps
Companies will need AI engineers and data scientists capable of filling the spots on multi-role teams. They'll also need business-side employees who understand workflow and business processes as well as the pain points and opportunities within the enterprise that could benefit from AI.
Organizations with successful AI programs use a combination of existing staff and new hires who often bring prior unique AI experience that adds to the company's culture. These new hires, Andrews noted, can work alongside and train existing employees who contribute institutional knowledge to the projects.
Successful organizations typically compile staffing roadmaps that identify skills gaps that need to be filled immediately or in the future and outline how to locate and acquire the necessary talent. These companies further address how they'll train the employees who will be using the AI-fueled systems.
6. Define how AI fits into your business strategy
Organizations have various reasons for adopting AI and different definitions of exactly what constitutes success. In all cases, however, that definition needs to relate to the company's overall strategic goals. Some executives, for example, may want to use AI for more accurate supply chain forecasting to reduce costs, while others see AI as a way to improve customer engagements and boost sales.
Executives should also know their baseline metrics to accurately measure results, while being realistic with their expectations.
"The chance that your project is only going to be partially successful is very high," Andrews explained, noting that an AI project can be considered a success for something as simple as helping a company do a task better moving forward.
7. Deal with your data
AI requires vast amounts of good data, so executives must have a plan that ensures the availability of a sufficient abundance of authoritative data.
"The more data you have, the more likely you can be to develop machine learning models and AI systems that can solve problems for you," said Luis Ceze, a professor at the University of Washington's Paul G. Allen School of Computer Science and Engineering and co-founder and CEO of machine learning startup OctoML. To ensure they have the needed data, Ceze advised, enterprises must not only have the right systems of record in place, but must also identify the internal and external data sets required.
In addition, businesses must build the technical infrastructure to gather, clean, move and store all that data and deliver it to the AI systems at the right time and speed. "You can't start with a bunch of Excel files and build out an AI solution," Deloitte's Ammanath said. "You have to enable a robust and reliable data infrastructure."
8. Plan for adequate resources
Along with building the required data pipeline, Ceze said CIOs should take stock of the compute resources that will be needed to advance AI initiatives. Some use cases can require significant computational resources and specialized infrastructure such as hardware accelerators. Latency concerns may also require some edge or on-premises processing as well as the cloud for additional processing and long-term storage.
These elements often translate into significant additional costs that need to be anticipated to ensure cash flow problems don't stymie AI projects. "It's not uncommon to see seven-digit bills a year or even each month just for compute costs," Ceze noted.
9. Address security, privacy, regulations, legalities, ethics
AI comes with significant security, privacy, regulatory and compliance concerns as well as legal issues and ethical implications. Executives should address all these areas from the start and throughout as they mature their AI programs. "You need to ask," Ammanath warned, "'What are the ways this can go wrong?'"
10. Establish parameters for acceptable AI performance
AI is not infallible, Ceze said, so executives must think about what failure modes are acceptable. A chatbot, for example, may direct a caller to the wrong customer service representative, and a predictive maintenance system may falsely alert that a machine is failing.
Executives need to plan for such possibilities and establish acceptable parameters for performance so AI teams can plan and design a project accordingly. If the probability of an AI component failing is only one in every 10,000 computations, that may be acceptable for a customer chatbot but not in a self-driving vehicle.
Before deploying an AI application, executives need to determine what is and is not an acceptable performance.