Your company is probably spending more on AI than ever. Your productivity numbers don't show it.
A new Atlanta Federal Reserve study found that roughly 90% of executives believe AI has not yet boosted productivity at their companies. Business leaders and investors are starting to acknowledge what the data already knew: there's a deepening gap between AI investment and the productivity gains everyone expected to follow.
That finding is uncomfortable. It's also completely fixable — but only if you're honest about what's actually causing it.
The tools aren't failing. The deployment pattern is.
Most operators who aren't seeing results are using AI the same way: they're dropping it into an existing workflow as a faster version of what they were already doing. They plug ChatGPT into their email drafting routine. They use Midjourney to speed up social graphics. They run meeting summaries through an AI notetaker and then... keep running the same bloated meeting schedule. Every one of those uses saves a few minutes. None of them moves a metric.
The businesses that are seeing real gains aren't doing that. They're doing something that sounds simple but is actually a harder mental shift: they're redesigning the workflow itself around what AI changes, not just speeding up the old one. That means asking a different question. Not "how do I do this faster?" but "given what AI can now do reliably, does this workflow need to exist in its current form at all?"
Here's a concrete version of what that looks like in practice. Say you run a professional services business and you're spending 6–8 hours a week building client-facing reports — pulling data, writing summaries, formatting deliverables. The AI-tool-as-speed-boost move is to use Claude or ChatGPT to draft the prose faster. You save maybe 2 hours. The workflow redesign move is to recognize that AI can now own the entire generation step if you give it a structured template and clean input data — and your job becomes reviewing and interpreting the output instead of producing it. That's not 2 hours saved. That's the entire task category reassigned.
The distinction matters because the first approach produces marginal efficiency. The second produces a structural change in what your team's time gets spent on. Only one of those shows up in your productivity numbers.
So why aren't more businesses doing the second thing? Because it requires you to look at a workflow and decide it needs to change, not just accelerate. That involves some risk, some rethinking, and the willingness to accept that a process you've been running for years isn't optimal just because it's familiar. Most teams take the easier path: add the tool, keep the process, report back later that nothing changed.
The fix, practically speaking, comes down to three steps. First, audit your time by category, not by task. You're not looking for "where can I use AI?" You're looking for "where do I spend 4-plus hours a week on work that produces a repeatable output?" Those categories are your targets. Second, for each target, define what "done" looks like before AI touches it — the format, the criteria, the pass/fail line. Third, test one category fully before moving to the next. Not a pilot across five processes. One. Run it for 30 days, measure the before and after, and treat the result as evidence, not anecdote.
The 90% figure from the Atlanta Fed study isn't a verdict on AI. It's a verdict on implementation defaults. Buying tools is easy. Redesigning work is harder. The gap between those two things is exactly where the ROI is hiding.
Stop waiting for the productivity to show up on its own. It won't. The return lives in the redesign, not the installation.