Harvard Business Review · Article

How to Move from AI Experimentation to AI Transformation

Names the “micro-productivity trap”: everyone gets faster at tasks, and the P&L doesn't move.

Published
Reading time
About 15 min
Authors
Arjun Dutt, Gene Rapoport, Aaron “Ronnie” Chatterji, Gawesha Weeratunga, Harrison Satcher
Read the original

What is it about?

Written by partners at Bain and members of OpenAI's economic research team, the article argues that most companies past the pilot stage treat AI like plug-and-play SaaS: scattered use cases, individual speed-ups and existing processes left intact. Gains stall at the firm level because the surrounding workflow still runs on tacit knowledge, manual hand-offs and legacy systems.

The way out has four parts: narrow AI to a few strategic domains, rebuild whole cross-functional workflows around it, let frontline experts lead the redesign, and measure business outcomes such as win rate or conversion against a non-AI baseline, with the C-suite owning the targets.

Key numbers

  • An anonymized Fortune 1000 manufacturer surfaced 14 use cases in a week-long workshop and rebuilt only 2 of them end to end.
  • Early-stage quotes went from four to six hours of engineering time to a 20-minute estimate by non-designers.
  • Win rate on AI-supported quotes rose 10 percentage points within three months; the company says it is on track for about $30 million in additional profit.
  • Lowe's associate assistant is used in more than 1,700 stores; customer satisfaction is 200 basis points higher when associates use it.

Why read it

  • It is the best short explanation of why individual productivity gains don't show up in enterprise results.
  • The manufacturer case shows the value came from removing work (12 discarded use cases, engineers spared on unlikely bids), not just from speeding it up.
  • Lowe's evaluation method, where experts write prompts and validated answers and models are compared against them, is a reusable template.

Who should read it

  • CIOs with many AI pilots and no P&L story
  • COOs and business-unit leaders told to “use more AI”
  • Transformation and strategy teams

About the author: Aaron “Ronnie” Chatterji

Chief Economist, OpenAI; professor at Duke University (on leave)

OpenAI's first chief economist, appointed in October 2024. He began as a financial analyst at Goldman Sachs, became a professor at Duke's Fuqua School of Business, served as a senior economist on President Obama's Council of Economic Advisers, and later coordinated implementation of the $52 billion CHIPS and Science Act at the White House.

In 2020 he took a sabbatical to run for North Carolina State Treasurer, won the Democratic primary and lost the general election with 47% of the vote. He also co-wrote the 2018 HBR essay that introduced the idea of “CEO activism”.

Read it carefully: common misquotes

  • Bain and OpenAI sell into these situations, and there is no survey or control group behind the cases.
  • The $30 million is a projection, and the manufacturer is anonymized.
  • Lowe's “conversion more than doubles” compares users with non-users, which may simply reflect high-intent shoppers.

How it shapes the CIO's AI Playbook

Notes by Pavel Veselov, last reviewed . Summaries are my own; quotes and figures belong to the original authors.

All 11 sources in the CIO's AI Library