by John Edwards

7 reasons your AI strategy isn’t working

Dec 17, 20208 mins
AnalyticsArtificial Intelligence

Where did it all go wrong? Instead of achieving success, many new AI adopters are tripping into common pitfalls that derail the technology's benefits. Here's how to get back on track.

frazzled confused fragmented broken ai
Credit: Thinkstock

Artificial intelligence (AI) promises to help enterprises boost productivity, business agility, and customer satisfaction while shortening the time required to bring new products and services to market. Yet as more IT leaders plunge their organizations deep into AI science, many are finding disappointment rather than success. A 2020 IDC study, for instance, found that 28% of AI and machine language (ML) initiatives have failed.

Creating an effective AI strategy requires careful planning, setting well-defined goals, and building a strong management commitment, plus the ability to deftly avoid common mistakes. If your organization’s current AI strategy is failing to deliver its expected results, here are seven likely reasons why.

[ 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. ]

1. Insufficient staff training

Failing to adequately address user needs is one of the biggest obstacles to a successful AI deployment.

“Unless businesses prepare people to use an AI solution, it will not scale,” warns Charla Griffy-Brown, professor of information systems and technology management at Pepperdine University’s Graziadio Business School. This isn’t just about training, she adds. “It requires updating policies and putting business support in place, not just technical support.”

IT leaders must ensure their workforces are adequately trained to work with the new technology, says Ravi Kumar, president of global IT consulting firm Infosys. “They should have a plan in place to educate and empower their teams to work with AI and not just consume it.”

The idea that AI requires human effort should be burned into the initiative from the very start. “This can be more difficult to deploy than the AI itself,” Grifffy-Brown says.

2. Missing or inadequate governance

AI strategies can’t work or scale effectively without a fully deployed, enterprise-wide model governance standard. There are many facets to model governance, notes Scott Zoldi, chief analytics officer at credit scoring service provider FICO.

“It must incorporate the concepts of responsible AI, which is robust, explainable, ethical, and efficient,” he explains. The model should also focus on standard technology deployment practices and specify which AI methods may, and may not, be used.

“Finally, AI projects need a corporate, governed-model development process [so] that models are created to the corporate standard and not subject to the artistry of individual data scientists,” Zoldi adds.

3. Failing to understand AI’s true value

As AI becomes accessible to a growing number of enterprises, many new adopters are failing to fully recognize the technology’s real-world ROI benefits. “It’s essential to integrate AI into the core value chain of industry applications rather than treating it as an add-on,” says Lan Guan, a senior managing director for the applied intelligence unit at professional services firm Accenture. “When AI is embedded seamlessly, value tracking becomes effortless, habitual, and addictive.”

The AI value discovery roadmap differs from those of most other enterprise technologies. Software, for example, comes with its own value guardrails.

“It’s very clear exactly what value an enterprise will gain,” Kumar says. Since AI lacks value guardrails, its value could be very well be exponential. “Organizations often don’t understand how to discover the full breadth of use cases for AI,” he notes. “Further, enterprise embrace of AI has typically been focused on pointed problems or on addressing a specific challenge, not necessarily thinking ‘big picture’ about how the technology can be used across their value chain.”

4. Neglecting to fully embed AI into existing business processes

For AI to create value, it must be embedded directly into the target business process. This not only means that the business process will need to change, but the human role within the process will have to adapt, too.

“For most mundane tasks, AI can automate the entire process and take human [interaction] out of the loop,” says Shervin Khodabandeh, a senior partner and AI co-lead at Boston Consulting Group, a management consulting firm.

Khodabandeh notes that full, human-less automation is an important AI benefit, yet represents only a small fraction of value the technology is capable of providing. “In our research and work with leading organizations, we see that they often leverage AI beyond automation — they use it to drive growth, improve customer experience, and manage risk better.” The most effective organizations achieve this goal by implementing new models of human-AI interaction.

In customer service, for example, it’s not just what AI can do, it’s how the human customer service agents work with AI to serve customers better. “In order to truly adopt organizational learning and see human-AI systems flourish, companies need to start the AI initiative with a deep understanding of the underlying business processes that have to change and the potential multiple ways that humans and AI can interact in the new process,” Khodabandeh explains.

Selectivity is also important. AI has become an IT buzzword, and few CIOs want to find themselves trailing behind the AI bandwagon. Yet in their rush to catch a ride, many IT leaders feel they must throw AI at every possible business challenge, observes Manjeet Rege, director of the Center for Applied Artificial Intelligence at the University of St. Thomas in St. Paul, Minn. “Often, we see an AI department developed that doesn’t integrate well with the business units,” he states.

Rege proposes launching an AI initiative that will be funded for the first two to three years by the affected business units. “That way, the AI team is given enough time to showcase the possibilities of AI to the business units,” he explains. “At the same time, the business units develop confidence in AI and are willing to fund AI projects in subsequent years.”

5. Insufficient management and monitoring

CIOs are experts in delivering “five nines” uptime. Instilling AI rigor is no less important, because decisions made using the technology often directly affect human lives. “The same level of rigor that goes into making sure systems are up [and] running needs to be applied to make sure AI models are performant and being continuously monitored,” Zoldi says.

Zoldi points to a recent Corinium Global report on building AI in a disrupted environment, which found that 67 percent of chief data and analytics officers don’t monitor their models to ensure their continued accuracy, as well as to prevent model drift and bias. “Despite being typically overlooked, AI model deployment and monitoring are as, or more, important than the core model development,” he says.

6. Lack of upper management support

As many CIOs are well aware, there’s often a lack of data literacy among senior business representatives. Therefore, it’s up to IT leaders to showcase and visualize the impact and benefits of developing a vigorous AI strategy.

Enterprises will struggle with scaling their AI strategy when they don’t have complete buy-in from executive sponsors and aren’t correctly prioritizing and innovating their use cases, says Jerry Kurtz, executive vice president of insights and data for business and technology consulting firm Capgemini North America. “If organizations can’t see the long-term benefits and payback of their short-term investments, then it will be hard to get the buy-in to scale these AI strategies for long-term commitments,” he explains.

Convincing senior management that AI is a proven value-building technology can be challenging, Kurtz admits. “[Resistance] can be successfully overcome but requires a very carefully designed AI strategy and roadmap that addresses the data track in parallel with the business use case identification/prioritization process and effectively addresses the non-technical barriers to scaling,” he notes.

7. Neglecting adoption management

Resist the urge to blow the entire AI budget on technology purchases. “Instead, spend almost an equivalent amount on adoption management,” recommends Krishna Kutty, managing partner and co-founder of management consulting firm Kuroshio Consulting. “Setting aside funds for communication, training, workflow redesigns, and organizational structure changes from implementing AI is a necessity for success,” she says.

Kutty notes that many enterprises assume that investing in AI technologies and related data management tasks is enough to get the job done. That’s a big mistake. “The majority of problems happen outside the narrow IT-centric teams,” she warns. The entire organization, from operations to finance to HR to marketing, must be included in both the operating and business models to deploy AI effectively. “Effective CIOs work in a partnership model with their C-suite peers to ensure the development of a holistic AI strategy and associated success in deploying the technology at scale,” she says.