A recent trend shows that while more workers are utilizing artificial intelligence tools than ever before, their confidence in these technologies is on the decline. According to a study by ManpowerGroup, workers’ confidence in AI dropped by 18% over the last year, even as adoption increased by 13%.
An analysis reveals that artificial intelligence chatbots, while designed to assist, can sometimes provide misleading or harmful information. According to a report from the Stanford University Institute for Human-Centered Artificial Intelligence, nearly 30% of users encountered harmful suggestions during interactions with AI chatbots.
AI-driven automation is transforming customer service operations, shifting from mere experimentation to essential scaling in large enterprises. According to Tom Eggemeier, CEO of Zendesk, the adoption of AI is yielding significant benefits for businesses. Eggemeier noted that over 50% of automated resolutions are leading to increased customer satisfaction and operational efficiency, emphasizing the need for organizations to rethink their workflows and data management in order to fully leverage AI capabilities.
Why do we care?
We’ve got a market that’s buying AI faster than it trusts AI. Eighteen-point confidence drop while adoption climbs thirteen percent. That’s not a paradox—that’s the hangover from two years of hype meeting reality.
And look, the Stanford “30% harmful suggestions” number is doing a lot of work in this narrative. We don’t know which chatbots, we don’t know what “harmful” means, and we definitely don’t know if that applies to enterprise implementations with actual guardrails. But here’s why it matters anyway: your clients don’t know that either. They see that headline, they experience their own AI frustrations, and they start questioning every recommendation you’ve made.
Now flip to the Zendesk CEO claiming 50% automated resolutions are driving satisfaction. Of course he’s saying that—his company’s valuation depends on it. But even if we take him at face value, notice what he’s actually admitting: you need to “rethink workflows and data management” to make this work. That’s the real cost. That’s the part nobody budgeted for.
So here’s the trap MSPs fall into: they see declining confidence and think “slow down on AI.” Wrong move. The confidence gap isn’t a reason to retreat—it’s the market telling you where the services opportunity lives. The distance between “deployed” and “trusted” is billable work. It’s four distinct risks MSPs are now responsible for: operational risk when outputs are wrong, financial risk as pricing and models change, governance risk when systems can’t be audited, and reputational risk when clients blame “your AI” for bad outcomes.
And the deployments that actually work: they’re narrow use cases, constrained data, human review, and clear failure paths when the model gets it wrong.
Because when that chatbot gives harmful advice—and it will—someone’s taking the call. If you’re just the reseller, you’re taking the blame without the margin. If you’re the governance layer, you’re the one who caught it before it reached the client.