A new report on opportunities to create business value with artificial intelligence (AI) contains a data point that speaks to the long road ahead: By the end of 2022, one in four large companies, or 25%, is expected to have moved beyond the pilot phase to operationalize its AI work. The remaining 75% are piloting or considering AI projects.
While that 25% figure number is modest, it’s up significantly from the 9% that had operationalized their AI work as of 2020, and just 5% in 2018.
The data, contained in a report called “How to Create Business Value with AI” from IBM’s Institute for Business Value (IBV), is based on discussions with more than 35 organizations with AI implementations — and includes a dozen case studies on AI usage across as many industries. The report sets out to debunk common myths surrounding AI and, in so doing, creates a guide to using it effectively, particularly on business processes where it can have a tangible impact.
One of the report’s core recommendations is that “C-suite and other leaders not buy into some of the myths surrounding it, such as ‘AI shortcuts don’t work’…Instead, they need to make decisions grounded in AI reality.”
Build on a proven foundation.
The IBV report cites the benefits of leveraging off-the-shelf foundational and pre-trained models that can provide a cost-effective and expeditious starting point with AI projects. One important factor that plays into this recommendation: Companies have struggled to take advantage of work previously done by data scientists to train data sets, as each business problem was approached with a new AI model.
Now, shortcuts are emerging; these are pre-trained models analogous to off-the-shelf software that can be installed and used rapidly. The approach is designed to help organizations accelerate their work without having to generate completely new data sets for every application; instead, they can leverage knowledge gathered from solving one problem to help solve related problems.
Examples of such pre-trained models include Google’s BERT and OpenAI’s GPT-3. These types of models have three key benefits: They improve the economics of AI by amortizing costs across multiple use cases, they improve results by delivering greater accuracy from larger data sets, and they bring new capabilities to bear.
Shortcuts have also gained traction in commercial products, such as RapidMiner (recently acquired by Altair), whose latest product release includes a “fully automated AI” capability that generates models based on business expertise alone, targeting non-coders and non-data scientists. RapidMiner described this as part of its objective to democratize AI.
Illustrating the benefits of off-the-shelf models, Boston Scientific spent $50,000 while leveraging open-source AI models to address its goal of automating the inspection of stents in the medical products field, according to IBM’s report. Through this and other measures, Boston Scientific was able to capture $5 million in direct savings while achieving higher accuracy in inspections than it had previously.
Focus beyond cost savings.
Examples abound of companies looking to apply AI to automate or perform functions that become costly when they are performed repetitively by humans and there’s little to no human value-add. Through this lens, AI can be accurately viewed as having the potential to reduce costs and that, of course, is a good outcome.
But the report notes that cost reduction is not the sweet spot for AI applications. Indeed, leading organizations are focused on growing their business and achieving competitive differentiation through the use of AI. As the authors expressed, those companies at the AI forefront are focused on customer-centric, top-line growth.
By way of example, IFFCO-Tokio, an India-based insurance joint venture, deployed a form of AI to evaluate images of damaged cars and classify the models, parts damaged, and type of damage after accidents occurred.
AI was able to determine whether parts could be repaired or required replacement. Further, it provides a cost estimate while keeping a human assessor involved. While settlement costs dropped by 40%, the customer “acceptance” ratio improved from 30% to 65%, resulting in increases in customer satisfaction, retention, and acquisition.
Recognize that one size does not fit all.
How to Create Business Value also includes something of a cautionary note, calling out the myth that AI is a one-size-fits-all proposition, or that AI can and should be considered in virtually any application or use case to drive business results. Not so, the report’s authors say.
Before embarking on an AI initiative, the first order of business is to determine whether AI enablement can serve a larger strategic initiative or address a core business problem. In short, there needs to be a fit for the purpose of the AI initiative.
Tailored uses of AI can solve distinct business problems — across geographies, and industries. But the report’s case studies show that the right approach often becomes clearer after the right data set to solve the problem is chosen. Laying the correct groundwork is a critical element of success — Boston Scientific and IFFCO-Tokio help illustrate the point clearly.
For more exclusive coverage of innovative cloud companies, check out Cloud Wars Horizon here:
