Four Best Practices for Successful AI Projects

Our neighbor is a brew master for a large brewery, and he wants to retire, but the company doesn’t want him to leave. The problem is that no one else and nothing else can do his job. It begs the question: why can’t artificial intelligence (AI) obtain the necessary recipes and knowledge from him, and then do the job?

Perhaps the answer can be found in an April, 2021, survey of 700 IT professionals and C-level executives conducted by Juniper Networks, The survey showed that while 95% of survey respondents felt that their companies would benefit by embedding AI in their operations, only 6% had actually implemented AI operationally.

There is clear hesitancy, yet there are also examples where organizations have been highly successful in their AI implementations.

What have we learned about successful AI best practices that can speed AI adoption?

1. Address the AI distrust issue head-on

It’s always important to remember that AI is a tool, not an entire operational replacement. Yes, AI can automate certain operational areas of a company, such as monitoring for environmental conditions or running a manufacturing assembly line, but AI is at its best when it interacts with human beings who have unique knowledge of what they do, and who act as the final decision makers in any process.

Employees can be fearful of losing their jobs when they hear about AI, so a good approach is to engage with employees upfront in the design of their work processes and how the AI will fit into those processes. If employees design and buy into revised new work processes and understand the role that AI will play, there will be less fear.

In the management area, the fear is that AI will take over decision making. Several years ago, I met with the CIO of a payment processor who had tested automated disaster recovery (DR) and failover in his data center. He understood that AI could preempt problems and provide automation, but he could not accept that the ultimate decision to roll into a DR would be taken away from him. The solution was for the AI DR detection to be embedded in the data center, but for the decision to initiate a DR to be left to the CIO.

2. Properly prepare data for AI

An AI system is only as good as its data. Few organizations have fully developed data fabrics that contain all the connections and relationships of an organization’s data, no matter the data type or data source -- and few have data that has been thoroughly vetted and prepared to assure its accuracy.

In 2020, the ACLU tested Amazon's facial recognition software on professional athletes in New England. One in six players were wrongly distinguished, and 25 were identified as criminals.

In the UK, the government learned that 16,000 COVID-19 cases had gone unreported. The problem? Line and column limitations in an Excel spreadsheet that left some data out.

AI was involved in both use cases, and the results were disappointing -- but was the AI at fault or did the problem originate with the data?

3. Design strong use cases that deliver impactive benefits

Continental, a German automotive supplier, utilized AI to predict the optimal points for tire changes on commercial fleets. The AI enabled Continental to optimize its tire stock, increase up-time, and reduce maintenance costs. General Motors, in a collaboration with Autodesk, applied AI to development of a new seat belt bracket. The new bracket design created a product that was 40% lighter and 20% stronger than the original one.

Both are cases where companies got immediate bottom-line results from AI. Success stories like these inspire management and boards and make them want to invest more.

4. Never forget about customer satisfaction

In the Seattle metro area, there is still a network of surviving neighborhood hardware stores that are run by proprietors who know everything you need to learn about garden fertilizers, building a cabinet, or getting the right kind of epoxy or nail. Customers are willing to pay more for the convenience of just stopping by, but also for the know-how and service that these “old hands” offer.

AI systems need to do better at customer service.

Almost everyone can relate to the frustration of getting caught in the branches of an automated telephone tree. In some cases, you will be taken through as many as half a dozen layers of robotic questions before you can get to a real human agent who is capable of dealing with a complex question that you have. In other cases, you just get disconnected when you ask for a human.

What great AI system designers do is enlist personnel who are truly skilled in customer service. These individuals play integral roles in AI system design. The strategy works because of the human factors required when engineering and designing the optimal interface between man and machine. This is what customer service experts bring to the table, and they are not skills typically found among data scientists or IT professionals.

What to Read Next:

IT Pro Salary Survey: What You Earned in 2020
Why You Need a Data Fabric, Not Just IT Architecture
How to Get a Stronger Grip on Contract and Vendor Management
ServiceNow’s CIO Sees Strategic Fit for Low-Code Development