IBM Research Study Explains A Lot About Successful Enterprise AI Strategies

Over the last decade, IBM has made significant investments in AI. IBM Watson burst on the scene in 2010, followed by various DIY (do it yourself) infrastructure designed to enable AI software such as its Power Systems, OpenPOWER, and OpenCAPI. The company’s PowerAI software toolkit was another milestone, enabling enterprises with specific deep learning frameworks in mind to get into AI and ML quickly. The last several years have seen the rollout of POWER9, an architecture built from the ground up specifically for AI and ML, and IBM unveiled some big Power Systems customer wins at THINK 2019

Recently the company commissioned a piece of research on the DNA of a successful enterprise AI strategy, from, a firm whose managing partner Michael Gale wrote a highly successful book on digital transformation. The survey comprised of interviews with 566 executives and department leaders from companies with over 500 employees—45% of which were global enterprises. With many businesses currently looking to capitalize on the potential benefits of the emerging technology trend of AI, I see this as a valuable resource to make sure businesses aren’t diving headfirst without looking. Let’s take a closer look at the findings.

Keep your friends close, your AI closer

The major takeaway from the research, in my mind, was the strong correlation uncovered between the companies who possess dedicated on-prem AI capabilities and positive financials and performance. The research illustrated this point with some telling numbers—businesses who have highly dedicated, on-prem AI abilities are 500% less likely to get a negative outcome or fail than other solutions, with 245% better performance in product and service design. Additionally, the survey found that on-prem AI is at least 50% less likely to suffer from low performance. The highest change in revenue reported in the survey after implanting on-prem AI, was a whopping 208%. The highest change in earnings, for that matter, was 63%, and the highest change in margin rates was an impressive 77%.

Another finding of the survey is that AI is still very much in the experimental early stages within the enterprise. The survey found that over half of the executives polled were experimenting with AI or utilizing it on a limited basis around their organizations. Additionally, it found that roughly 1 in 7 are still in the planning stages, and a rather small 13.8% say they are actually fully committed to utilizing AI in their business. At this moment, only 42% of those surveyed currently run their AI on-prem and in the private cloud, and over 40% of those surveyed reported their AI forays are generating either neutral, negative, or tiny gains (under 5%). There’s clearly a lot of room to grow in terms of getting AI to work for business. Research like this plays an important role in guiding organizations on what is working and what isn’t, and from the looks of it, on-prem AI seems to be part of the answer. 

“Most clients I speak with believe in the promise of AI, so it’s still surprising to see that most companies are not actually seeing results from their AI initiatives. The ones that are seeing tangible, measurable results do one thing differently – they keep the strategy, data, technology, people and processes close to the core and controlled. They refuse to outsource that work,” Kimberly Storin, VP, IBM Cognitive Systems. “Keeping it close to the core requires new rules for infrastructure, but just as importantly, new rules for organizational collaboration and culture.”

In October, Kimberly led an industry conference presentation discussing this topic with an all-star panel of AI experts including Margaret Dawson of Red Hat, Hillery Hunter of IBM and Hilary Mason previously of Cloudera. The panel went deep into the idea that it’s easy for mediocrity and excellence in AI to look the same early on. The difference is that in the mediocre AI strategy the investments are being made in a naïve way, with disparate projects, tooling and infrastructure. The excellent AI strategies are the ones where the company “centralizes those decisions and investments and leverage smart defaults to make unified decisions,” according to Hilary Mason. In addition, data access can be better managed by creating better collaboration models and breaking down organizational silos. On premise AI strategies lead to better overall governance and control, as well as enable organizations to strategically approach scaling AI projects.

An interesting tidbit from the research was that, during this experimental phase, there is no one particular kind of AI that seems to be dominating in the enterprise—only 2% of enterprise AI is focused solely on either machine learning, deep learning, natural language, or visual learning. Rather, businesses seem to be experimenting with a combination of all these technologies. 69% say they are utilizing machine learning, deep learning, and visual AI, while 15% say they are utilizing all four. While this can probably be partially attributed to the fact that AI is new and organizations aren’t sure what applications will drive the most value, I think it’s likely that these organizations will discover that a combination of AI technologies actually is the answer. 

The research also identified several key challenges cited by respondents in terms of implementing AI successfully within their organizations. These include the lack of right technology specifications (more CPU and GPU power, more productivity) and employee skillsets, the potential for shifting priorities within the organization, the need for technologists to better understand the business challenges faced in transformation, and the management of the personnel needed to build out AI initiatives. 

Wrapping up

All in all, I see this as some valuable industry insight into the current state of enterprise AI. Businesses are falling over themselves to get on the AI bandwagon, but on-prem AI in particular seems to be giving businesses a real boost to their bottom line. Implementing AI in your business is a significant undertaking and value-adding is not a guarantee. This is good information for any business considering an AI transformation to have in front of them.