Companies are feeling pressure from competitors to implement artificial intelligence even when they are not sure it can deliver on its promises to help the business thrive.

I remember a mid-sized manufacturing company that adopted an AI-enabled customer service tool. The promise was that the tool could solve the customer’s problem without human intervention.

But, instead, the tool frustrated the customers because the chatbot couldn’t answer their questions and didn’t give many options for resolution by a human being and so people turned to a competitor with a reputation for excellent customer service. It turned out that the customer service team was an early detector of the problems, but by the time the company ditched the tool and went back to humans, customers had moved to competitors—a result that took months to show up in the sales data. The company changed its strategy and let humans respond to customer concerns early in the upset cycle and used AI for data analysis such as reviewing all of the complaints recorded by humans to identify patterns or easily resolve technical problems.    

This real-life example underscores frustration of CEOs over implementation and return on investment with AI. It’s understandable but also highlights the critical importance for the C-Suite to establish norms and practices for the ethical implementation of AI solutions, defining when human work or interaction is required and when the tasks can be outsourced to an AI tool      

That task takes on added importance with the absence of guardrails about AI use. While several states have passed legislation, the lack of a coherent national strategy for AI risk mitigation puts more responsibility on companies and consumers.

When it comes to identifying, procuring and establishing the best AI system for a company, chief executives don’t need to be tech wizards. They do need to make sure the foundational ethical task of balancing effectiveness with efficiency is maintained.

Organizations that recognize those challenges and ask strategic questions will be able to both ensure the best tools are in place to accomplish the task and that human wisdom and experience make sure the systems run effectively.

Jumping before thinking

But before leaping into an AI purchase, executives need to remind themselves that their workforce’s creativity and implicit knowledge are invaluable.

In their rush to adopt new technology, some companies override their values and select an AI solution that is ill-fitted for their organization. For example, a company might have a commitment to hiring from a diverse pool of applicants to find well-qualified people who will also live out the organization’s values. An AI tool that reviews resumes and then conducts the first interview may both overlook candidates who would be a good match and alienate candidates who want a company that values them as humans. For instance, dozens of large language models are used for generative AI, including Open AI’s ChatGPT and Anthropic’s Claude. Companies cannot simply choose one and assume it will work for them. And when they factor in a myriad of industries, from professional services to technology to retail—the models must serve their needs.

For example, entrepreneurial enterprises might value experimentation and want a system that allows for creativity; a law firm might value the opposite, wanting firm protocols and standardization that limit freelancing and mistakes. 

Over decades of doing leadership development work, I’ve noticed the breakdown happens when the rush to buy new technology overrides thoughtful strategy decisions. A recent MIT study noted that only 5 percent of AI pilots deliver meaningful impact.

Ask key questions

Chief executives should consider these five key questions as they evaluate AI solutions:

• What are we trying to accomplish? Are we creating something new, which is a task for humans, or analyzing existing data, which is a good use for AI?
• How do we build trust with the new AI system? Is there a clear record of all changes to the system so we can correct any errors, or is the audit trail invisible, making it impossible to determine where the problem lies?
• How do we ensure employees feel included in the process? Have employees been asked to contribute their implicit knowledge, thereby strengthening the programming and supporting buy-in?
• What are the risks? Do employees know how to ensure that organizational data is kept confidential? Have risks for hacking been minimized?
• What should we be anticipating? Given that many AI initiatives fail because of a lack of training and trust, what do we need to do to maximize the success of the venture?

As leaders answer these questions, the decision-making process supports a successful AI selection.

One of the most difficult tasks for organizations is mapping the work employees do and how they do it. That’s hard for mature companies because many of those processes aren’t written down. But smartly figuring out where AI can be employed and where it can’t be is critical. And that system analysis takes time.

After that, executives need to engage the IT department. The CEO can ask the design and implementation team whether the tool meets C-Suite objectives. They can also ask what implementation strategies will support an effective adoption of the tool. Finally, the IT team will also be able to discern between the appropriate use of internal data sets and large language models (such as ChatGPT and Claude). With these kinds of questions answered, the executive can more confidently sign-off on a purchase. Importantly, the programmer needs to understand the company’s values and how leadership intends to incorporate them into a technological solution. At this critical first stage, C-Suite leaders and programmers must be aligned.

Next, the department purchasing the AI system needs to identify early adopters who experiment with the AI system. These users can offer feedback on any challenges they run across, such as difficulty of use or repeated errors, and benefits, such as helping with routine tasks and analyzing data spread over several departments. Getting broad buy-in and overcoming naysayers make this one of the hardest aspects of AI implementation.

Be vigilant about use

One reason perpetual training is advised to help employees troubleshoot issues is to ensure they are maximizing the use of AI tools. Otherwise, there is a real chance that tools will gather dust, or frustrated employees will develop quiet workarounds to the technology.

Through it all, organizations need to make sure they always have a human in the loop, a technical way of saying human beings still need to check citations, web links and other statements for accuracy. Studies confirm that AI can produce errors, generating fake links, providing inaccurate document summaries, giving wrong answers and making up citations for facts. While the risk of error is lower with internally generated content, leadership should require the document’s creator to review the output for accuracy and to ensure it aligns with the organization’s values. 

C-Suite leaders who insist on rigorous and routine examination of their AI processes are the ones who will lead their teams to success.