pilot-momentum-flywheel

Source Metadata

Original cangjie-skill frontmatter from the distillation run:

id: f21
title: "试点势能飞轮 — 先求成功再求价值"
book: "AI for Everyone"
author: "Andrew Ng（吴恩达）"
source: p22
tags: [transformation, strategy, momentum, sequencing]
related_skills:
  - slug: ai-transformation-playbook, relation: depends-on
  - slug: start-small-find-partner, relation: composes-with
  - slug: cross-functional-brainstorming, relation: depends-on

R — Reading (quote from the original text)

"The primary consideration for the initial project or projects is success, rather than necessarily being the most valuable... picking projects with a higher probability of success can set the flywheel effect in motion" — p22

I — Interpretation (restate in your own words)

This is acounterintuitivestrategy: your first AI projectshould notbe the most valuable one — it should bethe most likely to succeedone.

Why?

• Early success buildscredibilityandorganizational confidence
• Success attracts more internal customers to come collaborate with you
• forminga flywheel effect: success → credibility → more projects → greater success
• Once the flywheel is spinning, then gradually take on higher-value projects

Google Brain case: speech recognition (low risk, non-core business) → success → other teams noticed → Maps project → greater success → eventually Ads (highest value). It took several years for the flywheel to fully spin up.

A1 — Past Application (past application cases)

• Google Brain: For the first project, choose speech recognition — not Google's most core business, but with high technical feasibility. After success, the second project was Maps, and only the third was Ads (highest value). The flywheel gradually spun up: success → credibility → more internal customers → bigger projects → greater success. The whole process took several years.

A2 — Future Trigger (future trigger conditions)

Trigger scenarios：

• The team is struggling with "what to choose as the first AI project"
• There are multiple AI ideas, and it's unclear which to do first
• The organization is just starting to build AI capabilities and needs to decide the starting point

Language signals：

• "What should the first AI project be?"
• "What AI should we do first?"
• "Which one should we choose as the first project?"
• "Pilot project"
• "Where do we start?"

Key distinction: When you need to choose from multiple candidate projectsthe first oneto execute, use this framework.

E — Execution (execution steps)

1. List candidate projects: Gather all possible AI project ideas.
2. Score by probability of success: For each project, assess "the likelihood of making it succeed" (rather than "how much it would be worth if it succeeds"). Consider factors: whether data is available, whether the technology is mature, whether team capabilities match.
3. Choose the project with the highest probability of success: Don't choose the most valuable one; choose the one most likely to succeed.
4. Execute and publicize the success: Complete the project with high quality and proactively publicize the results throughout the company.
5. Attract the next project: After success, other business lines will proactively come to you — at this point you can choose a more valuable project.

Completion criteria: The first AI project is successfully delivered, and at least one new internal customer proactively comes to collaborate.

B — Boundary (Usage Boundaries)

Inapplicable Scenarios：

• The organization already has AI credibility (it no longer needs to "prove itself" and can directly pursue high-value projects)
• All candidate projects have similar probabilities of success (in this case, choose the one with the highest value)
• Urgent business needs (cannot wait for the flywheel to start turning)

Failure Modes：

• Choosing an "easy but irrelevant" project: it succeeds, but no one cares, so a flywheel cannot form
• Not publicizing after success: the flywheel needs to be "seen" to start turning
• The definition of "success" is unclear: the project is completed, but everyone does not consider it a success

Author's Blind Spots：

• Andrew Ng does not clearly define the criteria for "success"—success must be visible and quantifiable. It is recommended to clarify before launch: what counts as "success" (e.g., accuracy targets met, deployed and running, business metrics improved). Otherwise, once the project is completed, people may disagree about whether it was "successful," and the flywheel cannot start. In addition, "pursuing success first" does not mean "choosing a worthless project"—a successful project still needs to have some business value; it simply does not pursue the "greatest" value.

Related skills

• ai-transformation-playbook — depends-on: The pilot potential flywheel is the core strategy of the first step in the five-step AI transformation guide and needs to be understood within the overall transformation roadmap.
• start-small-find-partner — composes-with: The strategy of first finding partners and starting with small projects is consistent with the flywheel concept—building credibility through small successes and attracting more collaboration.
• cross-functional-brainstorming — depends-on: Before selecting a pilot project, cross-departmental brainstorming is needed to identify candidate projects; this is the prerequisite step for starting the flywheel.