AI Best Practices: Steps to Success for Business Leaders and Practitioners

Although artificial intelligence has permeated the technology landscape, companies are still struggling to understand how to integrate AI processes to meet business goals. AI’s transformative possibilities can only be unlocked when business decision makers and AI practitioners understand how it can enhance business processes.

To help, CompTIA’s AI Advisory Council created AI in Business: Top Considerations for Before Implementing AI. The guide is intended to help both business leadership and AI practitioners think through how best to use AI to drive operating efficiency and profitability. It offers a snapshot of major AI technologies and implementation barriers, then details 15 key questions businesses should ask as they embark on an AI journey, based on council members’ experience and expertise.

Why AI and Why Now

Rama Akkiraju, IBM fellow and one of the contributing council members, says the success stories coming out of multiple industries are motivating AI exploration by more organizations.

There’s good reason to do so, according to fellow council member Lloyd Danzig, chairman and founder at International Consortium for the Ethical Development of Artificial Intelligence. He says more efficient algorithms paired with advances in computational power open the door to solve problems that were unsolvable until now. “Businesses can now leverage data to make better predictions, operate more efficiently, and drive a superior customer experience,” he said.

So far, that has been most evident in customer service operations, especially with chatbots. IT operations systems management has been another area showing positive traction, said Akkiraju. On the other hand, the healthcare industry has produced several applications that haven’t scaled because areas including oncology or radiology may be too broad and complex for AI thus far. Instead, small startups have found more consistent success with AI solving more granular problems.

Related content: AI Terminology: A Glossary for Beginners

First Steps and Next Steps

Any attempt at AI implementation requires asking the right questions. After reading AI in Business, Danzig said the best next steps will vary according to business types and needs. However, those steps should be much clearer after reading the guide and answering the questions it poses.

“Of course, every business and use case is different,” said Danzig. “The Best Practices document is a starting point on top of which context-specific learning and information gathering can be applied,”

Akkiraju said some readers will be able to turn to an internal data team to begin addressing the questions. Others, perhaps smaller companies, may look to engage consultants to do the assessment. Everyone, however, needs to address one crucial success factor covered only briefly in the guide: expectations.

“Where I see the most mistakes being made in implementing AI-infused solutions is expectation setting and expectation mismatch. Overpromising can lead to severe disappoints.  AI vendors have to be honest and upfront about what’s possible, what’s not possible and the time it will likely take to derive value,” said Akkiraju. “What’s possible on day one and month five could be different. A well-designed AI solution can learn from user feedback and improve over time.” When, expectations are not properly set, companies might think that they’ll start out with a perfect system that will automate an entire business process. In many cases AI systems need time and iterations of user feedback and learning in each company’s specific environment to get to the needed levels of accuracy and to deliver value.

That goes for AI practitioners and AI vendors, too, who Akkiraju says often panic and change course too soon when early results don’t match expectations.

“The builders of AI products need to take a long-term view to solve harder problems with AI. While it is important to adjust the course and refine use cases based on the feedback, making frequent product roadmap changes doesn’t help solve hard problems. Solving hard problems need investments, support, and focus for extended periods of time”. she said.

Start Small, Build on That

Akkiraju said the key criterion for an early win is a small start. A time-boxed proof-of-concept  can yield important insights. Either an in-house data science team or external vendor can run a small concept that makes big strides.

“Experiment and explore in small use cases, better understand the strengths and areas that need further improvement and have a roadmap for how those areas will be addressed by vendor or company itself and expand on initial successes,” she advises. “Wherever that path has been followed, I have seen great success.”