Episode 559: Why AI Projects Get Worse Before They Get Better

Play audio-only episode | Play on YouTube | Play on Spotify

Episode Summary

David Luria, author of Flatten the AI J Curve: Your Unfair Advantage in the Race to Enterprise Adoption, brings a practical conversation about what happens when an AI pilot moves into real operational use. Luria, a vice president and business manager at T. Rowe Price and a Six Sigma Black Belt, explains the AI J curve: productivity can fall before it improves, especially as teams learn new workflows, run old and new processes in parallel, and adjust to changing responsibilities. He argues that project managers should set expectations before implementation begins, keep early scope narrow, and use short stepping-stone projects to learn quickly. He also explains why culture matters. AI often removes tasks that people associate with their expertise, so adoption requires more than access to a chatbot. Leaders and project managers need to create room for experimentation, communicate what will change, and help people understand how their roles can shift toward higher-value work. Luria calls for business ownership close to the work, supported by leadership, IT, and risk teams, rather than treating AI transformation as a technology handoff.

The conversation then turns to how organizations move beyond "toy mode," where employees use AI individually without a clear connection to business value. Luria defines value as utility divided by cost and urges teams to rethink processes before automating them. His warning against "paving the cow path" centers on a simple question: should the process exist at all? Instead of measuring success mainly through estimated hours saved, he recommends looking closely at quality, including errors, inconsistency, delays, unnecessary complexity, cycle time, and variability. He also stresses that AI systems behave probabilistically, so a successful demonstration does not prove production reliability. Teams need repeatability testing, controlled prompts, and disciplined collaboration with IT as business users gain the ability to build agents and software-like capabilities themselves.

Finally, the discussion returns to the difficult point the organization knew was coming: the productivity dip. When stakeholder confidence starts to erode, Luria recommends returning to the expectations established at the start, showing what the team has learned, and speaking directly with concerned stakeholders. One-on-one conversations can surface information that never appears in a group meeting and may reveal the adjustment the project needs. Cornelius and David also examine how to talk about capacity, how project managers can connect business and technology, and why traditional key performance indicators may need to change. The final takeaway is less technical: Luria points to the idea of letting go. As AI changes knowledge work, project managers need to understand people's pain points, frame solutions in terms that matter to them, and help teams move forward without clinging to tasks simply because "this is how we have always done it."

In This Episode, You Will Learn

• The AI J curve - Why productivity can fall after a promising pilot and how early expectation setting helps sponsors understand the dip.
• How to start small without thinking small - How stepping-stone projects, business ownership, and imagination help teams learn before they scale.
• How to move beyond toy mode - Why broad chatbot access does not automatically create business value and how teams can connect AI use to measurable outcomes.
• How to redesign processes before automating them - Why project managers should question existing workflows, avoid paving the cow path, and ask whether a process should exist at all.
• How to build reliability and stakeholder confidence - How repeatability testing, IT partnership, better metrics, and direct stakeholder conversations help keep AI initiatives on track.

My course · 14.5 hours · 14.5 PDUs

Artificial Intelligence for Project Managers

I built this course to make you AI literate, not just better at typing prompts into ChatGPT. You'll learn how AI fits into the way you actually run projects.

See the course on Udemy

Resources Mentioned

• Flatten the AI J Curve: Your Unfair Advantage in the Race to Enterprise Adoption - David Luria's book on shortening the productivity dip and improving enterprise AI adoption. https://flattenthej.com/
• The J Curve Insider Audio - Additional audio material related to the J curve. Visit the page and enter pmpodcast for access. https://flattenthej.com/insider

Quotes from This Episode

• "It gets worse before it gets better." - David Luria
• "Putting something in AI through an AI process that's a crap process, you're just putting lipstick on a pig." - David Luria
• "Talk to people and find out their pain points. And if you have a solution, they will listen." - David Luria

Connect with David Luria

• LinkedIn: https://www.linkedin.com/in/davidluriaprofile/
• Website: https://flattenthej.com/

Time-Stamped Show Notes

• [00:31] - Why successful AI pilots can run into a productivity dip when organizations move into real operational change.
• [03:14] - The AI J curve: why performance gets worse before it gets better and how teams can shorten the dip.
• [04:47] - Month zero: setting expectations, preparing the culture, and creating room for experimentation and failure.
• [14:40] - Ownership and the business AI disruptor: connecting leadership support with frontline business knowledge.
• [20:37] - Think big, start small: using imagination and stepping-stone projects instead of automating old assumptions.
• [24:29] - Moving beyond toy mode: connecting AI activity to utility, cost, quality, and measurable business value.
• [32:09] - Fix the process first: avoiding the cow path and asking whether a process should be eliminated before it is automated.
• [37:00] - Managing operational risk: partnering with IT, testing repeatability, and creating more controlled prompting practices.
• [47:19] - Month four arrives: protecting stakeholder confidence, revisiting expectations, and using better measures of project value.
• [55:17] - Take action: letting go of old tasks, understanding people's pain points, and helping teams adapt to AI-driven change.