Why AI Projects Fail on the Frontline

Understanding why AI projects fail has become one of the more valuable exercises for any enterprise putting technology in the hands of frontline workers. On the Frontline Mobility Edge podcast, BlueFletch COO Brett Cooper spoke with Mark Rogers, Director of Market Intelligence at Zebra Technologies. They dug into the pattern behind stalled AI efforts in retail, warehousing, and healthcare, and what separates the projects that stick from the ones that quietly get shelved.

Watch the full episode for the complete discussion.

Why Do So Many Enterprise AI Projects Get Abandoned?

A few years ago, a Zebra presentation predicted that 30% of AI projects would be abandoned by 2025. Rogers thinks the real number now feels higher, and the industry data is starting to agree. He pointed to a newer projection: by 2028, roughly 30% of AI projects will be abandoned because of what analysts call silent employee indifference. In plain terms, the workforce quietly stops backing an initiative.

The starting point is often simpler than a failed model. On a recent customer visit, Rogers asked a room of enterprises a basic question: who has an AI strategy? Not a single hand went up. “There’s seemingly a strategy or a confusion on where to start that seems to be ubiquitous,” he said. When the strategy is unclear, and the ground keeps shifting underfoot, abandonment becomes the predictable result. That gap between ambition and execution is a big part of why AI projects fail.

Prefer to listen? Catch this episode on your favorite podcast app.

What Is Really Blocking Frontline AI Adoption?

Rogers kept returning to a few root causes for why AI projects fail. The technology is rarely the hard part. The organization around it usually is.

Siloed data and weak governance

Most stalled projects trace back to data. Research Zebra conducted with Oxford Economics found that two out of three retailers still operate with siloed data. Without clean, governed, connected data underneath, even a promising AI use case has nothing solid to run on. Governance is the foundational layer most teams have not addressed, and it is a big part of why warehouses are slow to adopt AI despite real appetite for it.

Shadow IT and the “better tool” problem

The second blocker is a familiar one wearing a new outfit. When a leading AI tool is noticeably better than the copilot a company already pays for, employees reach for the better tool on their own. Rogers compared it directly to healthcare, where clinicians have long used unsanctioned transcription apps to move faster. The instinct is understandable, because people want to be effective. Left ungoverned, though, it scatters sensitive work across tools no one is watching, which is exactly where identity and access controls on shared devices earn their keep.

Unverified sources and the credibility gap

The third issue is trust in the output. Rogers noted that his team relies on accredited, licensed research sources, including firms like Gartner. A general AI tool, by contrast, can answer a market-sizing question in ten seconds by pulling from whatever is publicly available, which is often far less credible. Cooper shared a personal example: he asked an AI for tax guidance, then asked it to cite IRS sources, and it admitted the approach was wrong. His fix was to add guardrails that force the tool to cite sources for anything factual. That is governance in miniature.

Where Is On-Device AI Actually Paying Off?

Not every AI story is a cautionary one. Rogers is genuinely optimistic about on-device AI, where a co-processor lets a device run vision, audio, or transcription models locally instead of round-tripping everything to the cloud. Zebra devices such as the TC53e and EM45 already point in this direction.

The economics are part of the appeal. Pushing AI to the edge trims network and token costs that otherwise stay hidden until they hit the bottom line. The workforce case is just as strong. Rogers pointed to retail turnover running near 60%. Even with that churn, a frontline associate has real influence over revenue and the shopping experience, so any on-device capability that removes friction is worth a close look. According to Rogers, Zebra now describes itself as more of a hardware company than ever, as physical devices show up everywhere from self-checkout lanes to store shelves. For teams weighing that investment, where to start with AI and automation is often a harder question than which model to run.

What Do Vision, RFID, and Robots Mean for Frontline Roles?

Cooper and Rogers closed on where the hardware is heading, and the answer was refreshingly unhyped.

Vision and RFID, matched to the customer

At recent NRF and MODEX shows, vision tools drew a lot of attention, and one customer even talked about dropping RFID to go all in on cameras. Rogers sees vision as part of the future state, but he resists a single answer. The right mix depends on tier and use case. Larger operations may run full multimodal setups that combine barcode, RFID, and vision, while much of the mid-market is still automating its first manual workflow. Smaller companies can sometimes leapfrog larger ones precisely because they carry less legacy baggage.

The human-in-the-loop future

On robotics, Rogers stayed grounded. He compared it to riding in a self-driving car that still has humans monitoring from a control room. One projection he cited: by 2030, 20% of managers will have at least one robot in their org structure. The frontline job of the future may look less like moving boxes and more like supervising a mix of people and machines. He was quick to note, though, that labor and regulatory pushback could shift the timeline.

Frequently Asked Questions