Why AI adoption lags: Blame the clunky UI

Companies are pouring millions into enterprise AI, but inside the office, employees often ditch the shiny new tools and quietly retreat to spreadsheets.

According to Petr Baudis, CTO and cofounder of document processing platform Rossum, the culprit isn't necessarily the AI model itself—it is usually terrible user interface design.

The hidden bottleneck in enterprise adoption

When evaluating AI platforms, decision-makers are frequently blinded by sleek aesthetics. But in practice, if a tool is clunky, slow, or cluttered with endless dropdowns, workers simply refuse to use it.

Baudis argues that an AI tool's practical capability boils down to functionality and simplicity. If the UI is overwhelming, adoption drops, and recurring usage flatlines.

Data backs this up. A survey by BCG and MIT Sloan Management Review found that when employees derive personal value from AI, the organization is 5.9 times more likely to capture business value. Furthermore, those employees are 3.4 times more likely to report higher job satisfaction.

For real adoption, the interface has to make daily routines easier, not harder.

Morton Salt's 10-second turnaround

Consider Morton Salt, North America's leading salt producer. The company previously spent countless manual hours sorting through purchase orders arriving in wildly inconsistent formats.

Rossum deployed a dedicated AI engine tailored specifically to Morton's purchase orders. Because the interface centered on fast human validation rather than confusing complexity, the system slashed document processing time by up to 95%, bringing average handling down to just 10 seconds per order.

To replicate that success, Baudis recommends keeping interfaces cohesive and accessible across skill levels. He also highlights the need for transparent metrics, dedicated feedback channels like Slack, and ongoing employee training as new features roll out.

Why it matters

We think this highlights a critical blind spot in the generative AI gold rush.

For enterprise buyers, raw model intelligence is quickly becoming table stakes. The real return on investment happens at the interaction layer: if your team needs a training manual just to prompt an agent or review extracted data, that tool will become shelfware.

For software builders, the competitive moat is shifting from model size to workflow empathy. A mediocre model wrapped in an effortless, frictionless UI will beat a frontier model trapped behind a cluttered dashboard every single time.

However, a cleaner UI isn't a silver bullet on its own. As noted in the findings, software works only when leadership sets clear productivity metrics and opens a feedback loop. Technology can save 95% of someone's time, but only if the front door isn't locked behind bad design.