Turning AI Myths into Enterprise Gold

Overcome the hype and unrealistic expectations surrounding AI by recognizing today’s biggest AI myths and how to address them.

By Clive Swan | February 2020

Clive Swan: Turning AI Myths into Enterprise Gold

We’ve all heard news stories about how artificial intelligence (AI) is powering self-driving cars and computer programs that outplay humans in the game of Go. But the question we’re hearing from enterprise leaders is: How can AI transform our enterprise?

The answer is that AI is one of the most disruptive technologies we’ve seen in decades and that it’s leading to new business models and revenue opportunities. In fact, a recent McKinsey study shows that early adopters of enterprise AI could double their cash flow by 2030 and achieve higher long-term growth rates than their peers. Laggards, meanwhile, could see a 20 percent decline in cash flow over the same period, the study found.

Unfortunately, while business leaders are realizing how AI can help their enterprises thrive, many are stumbling on their way to implementing AI at scale. I’d argue that the biggest roadblocks to AI success are two factors we’re all quite familiar with: AI hype and the myths it creates.

To neutralize these barriers and smooth the journey to enterprise AI, IT leaders and business executives must understand the realities behind the five biggest myths surrounding AI.

Myth #1: Enterprise AI requires a build-it-yourself approach. Your business is unique and thus requires custom-tailored AI applications created by highly skilled, internal experts or expensive third-party data scientists and developers.

Reality: The best approach is balanced: You should both buy and build applications. The key is to buy commercial off-the-shelf solutions for general business processes and focus your internal data scientists on projects that will drive your competitive differentiation. For example, minimizing customer churn is a critical objective for all telecom providers. It’s likely that the very best AI solution for identifying which customers are ready to defect and which offers will convince them to stay will be a proprietary one, specific to that provider.

Myth #2: AI will quickly deliver game-changing business benefits. The idea of an AI moon shot inspires executives to believe in a transformative solution that immediately triples revenues, for example.

Reality: AI is not magic. Truth is, the path to AI success takes time and a focus on some unglamorous steps that organizations must take to make AI work. Organizations need a strategic framework to deliver interrelated solutions that compound AI’s benefits rather than rolling out a collection of disconnected solutions that realize a fraction of AI’s potential.

For example, one AI solution within an ERP platform can reduce the manual effort needed to process invoices arriving without PO numbers, by recommending an account code to complete the payment. That alone offers value, but the benefits are magnified when that first solution is combined with a related AI solution that identifies which suppliers will offer the greatest savings for early payment. Because more invoices are processed more quickly, there’s a larger pool of invoices—and suppliers—to consider for early payment, leading to greater savings.

Myth #3: With AI, we won’t need people. Enterprises can significantly reduce head count, because AI automatically completes tasks now being done by employees.

Reality: AI becomes most valuable when it augments, rather than replaces, human capabilities. In short, AI and people need each other.

That starts with AI removing grunt work, the routine tasks that don’t take advantage of a workforce’s skills and expertise. People are then freed to do more strategic, value-added activities. A side benefit may be that top talent becomes more motivated, productive, and committed to staying with the company.

For example, consumer products companies can capitalize on the pairing of AI and people to improve sales forecasts for upcoming quarters while avoiding costs associated with excess inventory or shortfalls when consumer demand is higher than expected. AI performs the basic work of combing through internal business information and third-party marketplace data to provide initial sales analyses. Sales and marketing professionals then refine the forecasts with their expertise and insights gained from speaking with key customers. The AI application, in turn, learns from each human interaction to continuously improve its output.

Myth #4: The more data, the better. Stuff as much information as possible into bigger and bigger data lakes, and AI analyses will thrive.

Reality: Data quality matters more than quantity. AI needs what I call smart data: information that is of high quality; up to date; complete; and cleansed of redundant, outdated, and erroneous entries.

AI-driven data engines are the key to trustworthy data. The best examples both cleanse and enrich internal corporate records before they’re consumed by AI solutions. A leading AI data engine continuously extracts company data from public digital properties and scours daily press articles to detect company signals such as “company has filed for bankruptcy” or “company is expanding internationally.”

One development company I’m familiar with took advantage of an AI data engine when it merged two customer relationship management databases. The engine first cleaned up conflicts in the combined customer records and then enriched the information with up-to-date revenue and head count data from public sources as well as company signals. That created a solid foundation to help the sales and marketing departments prioritize accounts by their profit potential. Sales reps are now spending less time researching those accounts and more time engaging with top prospects.

Myth #5: Enterprise AI needs only data and models to succeed. Armed with a solid AI model and a comprehensive storehouse of enriched data, enterprises can quickly move from small-scale pilot applications to full enterprise implementations.

Reality: The right models and data are a start, but it takes additional effort to scale applications for production environments. Models typically need initial tailoring and tuning to fit to the “shape” of a customer’s data. But, even worse, they also need that tuning on a recurring basis to ensure that the model remains effective. Historically, all such extensive maintenance has been an expensive manual task carried out by data scientists. And although that may have been an acceptable cost for a single, bespoke AI solution, it simply isn’t a scalable approach when an organization wants to deploy 100 models or when a cloud vendor is deploying 100 models to 10,000 customers.

Replacing that expensive manual maintenance effort with an automated solution is absolutely critical as AI becomes pervasive, with hundreds of solutions deployed across enterprise applications. The approach Oracle and some other organizations are taking is to apply machine learning to automate that maintenance. Without that, customers would end up paying an “AI tax” to maintain their solutions or accept a progressive reduction in their effectiveness.

Hype has been growing steadily as AI becomes a growing force for mainstream business operations. It can fuel unreasonable expectations among senior executives who don’t fully understand the underlying challenges of bringing a new and disruptive technology into production systems. But by directly addressing the biggest AI myths with a healthy dose of reality, IT leaders can guide organizations on a multiyear AI journey that delivers incremental benefits and long-term competitive advantage.

Clive Swan is senior vice president of applications development for Oracle Adaptive Intelligent Apps.

Photography by Bob Adler/Getty Images