How to Avoid the AI Perfection Trap

By Tomas Gorny, CEO and co-founder, Nextiva

AI progress is stalling. Inside today’s businesses, top employees are spending their days on work that doesn’t need them.

“Companies in most industries are investing heavily in artificial intelligence: 88% of companies reporting regular AI use,” four authors noted recently in The Harvard Business Review. “Yet many leaders report familiar frustrations. AI adoption stalls.”

Boston University notes the excuse that many organizations turn to. “When AI initiatives stall, organizations often reach for the same explanation: the tools were not advanced enough,” the Questrom School of Business reports.

I see this happening across many industries. I also see one of the biggest forces causing this problem -- and how to fix it.

AI Takes Onboarding

Think of it as a perfection trap. The more people hear about the potential of AI, the more successful they expect it to be in the short-term. They think it must either work wonders or be too dangerous to use. They don’t realize the more complex reality.

Bringing AI into your organization is similar to onboarding a fantastic new hire. It needs to be trained.

It’s a form of onboarding. AI requires context. It needs lots of data -- information about the organization, customers, goals, struggles, and much more. It also needs a clear scope. It must be taught what goals to pursue. All of this requires access to your organization’s entire database. But data silos often block AI from helping.

For example, take cloud communication platforms. In order to help an organization deliver the best possible customer experience (CX), an AI tool needs access to each customer’s entire journey across every touchpoint. When separate channels have data locked away, AI can’t get the job done.

It needs to know what each customer has said and experienced, whether via chatbots, email, apps, or phone calls with both human agents and AI receptionists. Only a unified customer experience management (UCXM) platform can ensure the tool has all this.

Nextiva’s survey, The Leader’s Guide to CX Trends, found that 86% of companies with multiple CX tools report having siloed data. But when your CX tools all operate on the same UCXM platform, this problem evaporates. Everything from generative AI to intelligent virtual agents to inbound and outbound call center software should operate on the same shared data.

Scaffolding and Accountability

Even then, your AI won’t be flawless straight out of the gate. It needs to learn from each interaction, and receive guidance from staff members on how to do better.

New research demonstrates that when certain structures are in place, organizations can advance their use of AI successfully despite struggles early on. Published in the journal Information, the study focuses on agentic AI. “Well-scaffolded control and accountability cues can support continued reliance even when performance is imperfect,” author Stefanos Balaskas explains.

He discusses several key elements. There should be visibility into how the AI works, which means avoiding black box solutions. The tool should include “approval gates,” which keep humans ultimately  in charge. And the system should “support error recovery,” with clear and simple ways to cancel actions and escalate tickets to human agents when things go wrong.

There's a “sweet spot” for AI, in which “mid-level autonomy,” with the AI acting as a co-assistant, “can excel beyond both low and high autonomy,” Balaskas writes. In this way, the tool can optimize its assistance without displacing the user’s decision rights.

The businesses that will win with AI aren’t waiting for a mythical moment when the technology handles everything. They’re deploying thoughtfully, measuring results, scaling what works, and fixing problems as they happen.

I have a lot of conversations with business leaders. The ones who are struggling the most share one trait: they’re waiting. Waiting for the technology to be “ready.” Waiting for a use case that’s guaranteed to work. Waiting for someone else to go first.

AI doesn’t get ready on its own. It gets “ready” to handle more and more tasks by running inside your business, learning your customers, and absorbing how your team operates.

It’s the same lesson that business leaders are generally aware of when it comes to human errors. For individual staffers, “The relentless pursuit of productivity can, ironically, be the fastest path to burnout,” Nextiva wrote in a blog post.

But when it comes to automation, some people are less prepared to accept failures along the way. As you adopt AI, there will be setbacks. Don’t let them put you off. Apply the same logic. Make each step a learning opportunity. Teach your AI, have patience, and then watch it succeed. In short, don’t let perfect be the enemy of good.