Artificial Intelligence Done Right
Artificial intelligence (AI) has captured the imagination of a wide variety of businesses. I have this image of CEOs in boardrooms around the globe declaring, “We must have AI! Our competitors use AI! We can’t be left behind!” There might be some table-pounding associated with this scenario. There will certainly be corporate minions scurrying around to fulfill the AI dreams of their CEO.
As Mindbreeze’s CEO Daniel Fallmann assured me in a recent phone conversation, this is not the correct way to go about implementing an AI project. You don’t just go out and buy a box filled with AI that you plug into your computer. There are a lot of moving parts to consider. I asked Fallmann if he recommended starting small rather than trying to take on every possible area of a company where AI could be useful. He corrected me slightly. “I’d say starting focused would be the better way to put it.” In large corporations, for example, Mindbreeze might start with customer service. But it could equally be the maintenance department or research & development.
The idea is to get a 360-degree view of the department’s activities. “Talk with the head of the department,” advised Fallmann, “not with IT.” Look at what information is being collected and the KPIs (Key Performance Indicators) as they are today. What needs to be changed to make that information actionable? How can AI bring down the workload of the call center or the streamlining of no-defect parts maintenance? In Fallmann’s opinion, it’s not just about saving money, but also about creating a “Wow effect.”
Measuring the result of an AI project isn’t simply an ROI (Return on Investment) exercise. It also needs success criteria that measure the ultimate impact of business process transformation. He gave the example of the U.S. Food & Drug Agency (FDA) and its usage of Mindbreeze to identify inconsistencies in the submissions from pharmaceutical companies. The FDA wants to avoid a situation where a pharma company’s submission is rejected when a very similar one from a different company is approved.
Fallmann has another concern—dirty data, or as he terms it, “data garbage.” If fed dirty data, a machine learning (ML) application will return lots of false positives, leading to bad decisions. You need 85-90% accuracy, he thinks. Concentrate on the business case and whether the existing data is inaccurate, incomplete, inconsistent, or otherwise compromised, it will be difficult to obtain relevant results.
In Mindbreeze’s guidelines for businesses, Fallmann advises to start by identifying and determining the use case. Look for opportunities to pinpoint pain points that can be improved through a sustainable AI implementation, but don’t try to tackle too many at once. Then turn to defining the success criteria, which will revolve around business needs, data sources and quality, semantic relationship and extraction, and an ROI calculation. Next is hands-on testing with company data, which may need to be expanded and probably requires some cleansing.
A critical component of the Mindbreeze guidelines is involving the users. In a world defined by continuous change management, feedback from employees will vastly increase your possibilities for success. After all, they know how they do their jobs better than an outsider and can be instrumental in training the data set. A final step in the guidelines is validating the ROI calculation. Moving from a test phase to a departmental project and then to the entire company gives you the chance to measure the improvement in efficiency, employee satisfaction, and overall profitability.
In the final analysis, Fallmann told me that, to do AI right, “you need execution, not just strategy.” When top management calls for an AI strategy, Mindbreeze wants to have a concrete plan for executing on the strategy, making real, measurable, and sustainable improvements that lead to a competitive advantage.
Companies and Suppliers Mentioned