Beyond Incrementalism: why most AI projects produce a digital version of a bad process

The tools work. The studies prove it. So why do most AI initiatives fail to change anything?

That question sits at the center of Beyond Incrementalism: Navigating Organizational Change, Behavioral Inertia, and Human Fear in AI-Driven Innovation, a paper I co-authored with Ken Russell, Joe Wilck, Scott Gerard, John Hoag, Becky Scott, John Harrison, and Andrew Sohn through the Barton College Data Analytics Consortium.

The paper draws on sixteen sessions of the consortium, a group of educators and practitioners from higher education, data science consulting, and industrial engineering, read alongside the research on change management and AI productivity. Its argument is simple and uncomfortable: the biggest bottleneck is organizational. Human behavior resists change, and that resistance outweighs any budget or technical hurdle.

The ideas worth stealing

• The incrementalism trap. Organizations swing between tiny tweaks and moonshots. The real value lives in the middle, where most people cannot imagine things.
• Nine sources of resistance. The paper sorts them into three families: threat based, principled, and experiential. Not all pushback is fear, and treating it all the same backfires.
• The fluency trap. Polished AI output looks so confident that people stop checking it.
• Data literacy comes first. Many organizations are not ready for AI because they were never ready for their own data.

A three step path forward

1. Fix the process first. If you would not be proud of the manual version, do not automate it.
2. Commit to scaling, not downsizing. People will not help build a tool they believe will replace them.
3. Run low stakes, high visibility pilots. Seeing is believing, and small wins in public do more than big promises in private.

This is the same pattern I see in service, sales, and retention every day. The hardest work lives in the habits, the fears, and the processes nobody has questioned in years.