Finding AI’s low-hanging fruit | VentureBeat

Delivering AI solutions from the test bed to production environments will probably be the key focus for the enterprise throughout the next year or longer. But organizations should be cautious not to push AI too far too fast, despite the pressure to keep up with the competition.

This often leads to two key problems. First, it pushes inadequate solutions into environments where they are quickly overwhelmed and this leads to failure, disillusionment and mistrust from the user base that ultimately inhibits adoption. The AI industry is not helping anything with its stream of promises that their solutions offer complete digital autonomy and transformative experiences.

Small victories are still victories 

In some circles, the idea of going smaller with AI is catching on. Instead of a complete forklift upgrade across the entire business process, it’s better to do the easy stuff first. That is, put AI to work in limited, non-critical areas and see how it performs before promoting it to bigger and better things. In this way, successes are more frequent, trust is more easily earned and AI can learn how to integrate with the world as it is before trying to improve it.

For many organizations, however, the question is where to find this low-hanging fruit. 

According to Joe Bush, editor of The Manufacturer, it’s all around us. Resource consumption, for one, can be monitored far more easily and effectively with an intelligent platform than with teams of operators. While he speaks to an industrial audience, the same need to minimize the use of electricity, water and other basic commodities exists in the enterprise. With the right sensor-driven data, AI can also assess workloads across the digital environment and even shift it around to ensure the work-machine balance remains optimal. And AI can also react to changing circumstances far quicker than manual operators and can streamline key processes like reporting, maintenance scheduling and supply.

Of course, it doesn’t hurt to have a plan in mind when deploying AI into production environments, since it is far more valuable working in tandem than isolation. Accenture’s Bhaskar Ghosh, Rajendra Prasad and Gayathri Pallail argued recently in The Harvard Business Review that instead of aiming for quick victories or grand strategic transformations, the wisest course right now is to concentrate on building capabilities that address problems that will recur in the future. This will require careful analysis of current capabilities and identification of any gaps that are creating failures. Then you can create a step-by-step approach to deploying AI so it achieves the small victories that will ultimately lead to the grand transformation.

Small and wide data

Some organizations are also starting to realize that throwing AI at big data and hoping for something magical to happen is not the way to go either. In fact, according to Rohan Sheth, associate vice president  of Infrastructure Solutions at colocation provider Yotta, AI will likely be less effective at crunching through massive volumes of data and more effective using lesser amounts of more precise data – what some are already calling small and wide data. To get there, though, the enterprise will have to improve its capabilities to analyze and condition data before it is fed into AI models, which, coincidentally, is another area in which AI can be of great utility.

The extent to which AI can support the enterprise depends very much on an organization’s “data maturity,” says Sumit Kumar Sharma, enterprise architect at In2IT Technologies. In a recent interview with ITWeb, he explained that there is no “one-size-fits-all” approach to AI because every organization’s needs and legacy environments are different. Depending on the way data is generated, consumed and retained, different flavors of AI will provide a unique set of services and these services will be better for some business models than others. For instance, a business-to-business (B2B) supplier would have more use for chatbots and natural language processing than a large analytical firm, which in turn would probably gravitate more toward machine learning and neural networking.

At this point, it might sound like AI is simply another technology looking for a solution and in a way it is. But there is one major difference between AI and past generations of technology: it can adapt and respond to new data and changing circumstances. This gives the enterprise a lot of leeway to try and fail with AI, as long as each failure leads to further understanding as to how to succeed in the future.

It may be tempting to push AI into the most important aspects of the business right away in order to reap the rewards of a fully transformed operating model, but it’s not ready for that yet. Just like any other employee, it has to start small and prove itself before it can be promoted to greater responsibilities.