Confidence Is Running Ahead of Results

74% said AI has made their organization more productive. But for most of them, the improvement measured under 25%.

Upwork's research group surveyed business leaders about how they're actually using AI at work. Two of the answers, read next to each other, tell you more than the rest of the report does. 62% said they'd be confident handing an AI agent high-stakes work — the kind where a mistake costs real money. 74% said AI has made their organization more productive. But for most of them, the improvement measured under 25%. Those two findings don't contradict each other. Both can be true at once. But side by side they describe something worth noticing: belief in these tools is running well ahead of what anyone has actually measured from them. The technical part — you can skip this box Upwork Research Institute, Q1 2026 Business Leader Landscape survey. 750 U.S. business leaders across business services, healthcare, manufacturing, retail, and software/technology. Small and mid-sized business findings draw on 195 respondents from companies with 10–99 employees. Published June 10, 2026. What that actually means It's a real survey with a reasonable sample, not a press release with a number stapled to it. The small-business findings come from about 195 companies roughly the size we work with — 10 to 100 people. Big enough to take seriously. Small enough that no single percentage should be treated as gospel. One thing worth knowing before you read any of it: the survey was run by Upwork, a company whose business is connecting you with human freelancers. That doesn't make the numbers wrong, and we've no reason to think anything was cooked. But everyone publishing research about the future of work has a position on the future of work. Read accordingly. Including anything we write. Where the gap costs money Another number from the same survey: 32% described AI agents as mission-critical to their company's strategy. Set that against most respondents measuring under a 25% improvement, and you have a meaningful group of companies who've made something central to their plans while getting a modest return from it. That's the expensive shape. Not because a 20% gain is bad — it's real, and we'd take it — but because "mission-critical" implies a level of dependence that the measured results don't yet support. The businesses that get burned here won't be the ones who tried AI. They'll be the ones who built a process around it before checking what it was actually delivering. What we'd actually do The companies getting genuine value out of this right now are not the ones handing over big decisions. They're the ones pointing it at the tedious middle of the workday. Typing up notes. First drafts of things somebody will rewrite anyway. Sorting through incoming email. Pulling specific information out of long documents. High volume, low stakes, and — this is the part that matters — easy for a person to check in seconds. That last point is the whole test. If you can't quickly tell whether the output is right, it's the wrong job to start with. One more finding worth flagging: 27% named data security and compliance as their biggest barrier to adopting AI. That's the correct instinct, and we'd encourage it. Before pointing any AI tool at your business data, find out where that data actually goes, whether it's used to train the vendor's models, and whether that's compatible with what you've promised your own clients. It's a shorter conversation than people expect, and considerably cheaper to have beforehand than afterward. Start with boring. Boring is measurable, and measurable is how you find out whether the confidence is earned.