The expectation gap that sinks many projects
Gartner found in April 2026: only 28 percent of AI use cases in IT infrastructure and operations actually succeed and meet ROI expectations, 20 percent fail outright. The MIT GenAI Divide report reaches a similar conclusion: 95 percent of enterprise AI pilots fail to deliver measurable business impact. A recurring factor behind these numbers isn't technical shortfall, but an expectation gap that forms early in the project – one that project leads can either widen or close through how they communicate.
Two translation mistakes that both cause harm
The first mistake is overpromising: presenting an AI capability as certain and unlimited without naming known constraints. That creates short-term excitement, but the moment the limit becomes visible, it costs more trust than honest restraint would have cost from the start. The second, less-discussed mistake is underselling: broadly talking down AI capabilities to keep expectations low from the outset. That prevents sensible investment in genuinely promising use cases.
What an honest framing actually includes
An honest framing names proven strengths and known limits in the same breath, not separately or only on request. It states uncertainty explicitly where it exists, instead of hiding it behind confident language – a pattern that already came up in the module on reading AI outputs critically, applied here to your own communication toward others.
Practice section: a phrasing template for stakeholder updates
A simple template helps apply this balance consistently: "[Capability] works reliably for [concrete, already-tested use case]. For [related but untested use case], there's no solid track record yet." This phrasing promises nothing untested, but also doesn't withhold a real strength – and works equally well in status reports, kickoffs, and decision memos.