Look Dumb or Get Disrupted
Accept visible early flaws while learning fast enough to survive disruption.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 5
- Confidence
- 93%
An incumbent confronting disruptive technology faces two asymmetric risks: launching too early can expose embarrassing failures and weaken trust, while moving too slowly can allow a competitor to define the new category. The model recommends neither blind acceleration nor passive waiting. Instead, identify where the technology already creates user value, deploy it selectively, collect real-world feedback, and expand as accuracy improves. Leaders should explicitly compare reputational damage from visible mistakes with the strategic cost of lost users, data, learning, and market position. The output is a staged adoption strategy that tolerates reversible embarrassment while protecting high-risk use cases. Progress depends on whether deployment generates actionable learning and whether each iteration improves quality enough to justify broader exposure.
Origin
Extracted from Marketing Against The Grain, where Kieran Flanagan used Google's AI Overviews launch to explain the dilemma facing incumbents that must deploy immature technology without surrendering their market to AI-native challengers.
Core principles
- 01Incumbents must protect existing trust while building what could replace them.
- 02Real-world usage generates feedback that private testing cannot fully reproduce.
- 03Waiting for a disruptive technology to mature can surrender the emerging category.
- 04A flawed launch is tolerable only when learning produces measurable improvement.
- 05Adoption should advance selectively rather than treating every use case equally.
How to run it
- 1
Define the disruption threat
Specify how the emerging technology could replace part of the current customer experience, business model, or distribution advantage. Identify credible competitors already pursuing that path.
Pro tip Describe the threatened customer job rather than focusing only on named competitors.
Watch out Do not dismiss a challenger merely because its current product is small or unreliable.
- 2
Compare the two failure modes
Estimate the trust and reputational cost of launching prematurely, then compare it with the market share, data, and learning lost by waiting.
Pro tip Separate reversible embarrassment from damage that would be difficult to repair.
Watch out A vague fear of disruption can become an excuse for exposing users to unacceptable risks.
- 3
Choose bounded use cases
Deploy first where the technology already delivers meaningful value and errors are detectable or reversible. Withhold it from cases requiring consistently high accuracy.
Pro tip Use explicit quality thresholds for each use-case category.
Watch out Do not force the technology into every customer interaction merely to signal speed.
- 4
Launch for learning
Release the capability to enough real users to reveal failure patterns, edge cases, and unmet expectations. Capture structured feedback and behavioral signals.
Pro tip Instrument the experience before launch so learning is observable.
Watch out Exposure without a feedback loop creates reputational cost without strategic benefit.
- 5
Refine and expand selectively
Correct recurring failures and broaden deployment only where evidence shows improved value and trust. Continue comparing progress with the competitive threat.
Pro tip Treat rollout scope as a variable that can expand or contract by use case.
Watch out Do not interpret isolated positive anecdotes as proof that the experience is ready everywhere.
In the wild
A dominant search company introduces generated summaries only for query types where synthesis helps users. It suppresses summaries for ambiguous or safety-sensitive questions, logs recurring failures, and expands coverage as evaluation scores and user satisfaction improve.
→ The company gains deployment data and product learning without making the immature feature universal.
A large software provider fears an AI-native rival will redefine customer support. Rather than replacing every agent, it first deploys AI for internal draft generation, measures correction rates, and later exposes automation directly for low-risk account questions.
→ The provider develops operational knowledge while preserving human review for consequential cases.
Common mistakes
Waiting for perfection
Delaying all deployment until the technology is flawless can surrender users, data, and category leadership to faster competitors.
Launching everywhere at once
Universal exposure magnifies predictable failures and ignores substantial differences in accuracy requirements between use cases.
Mistaking attention for learning
A controversial launch provides little strategic value unless failures are captured, analyzed, and converted into product improvements.
Is it for you?
Best for
It is best for market leaders facing a fast-improving technology with credible new competitors and substantial first-mover learning advantages.
Not ideal for
It is not ideal when product errors could cause irreversible physical, legal, or financial harm.
From the transcript
“they can either steamroll ahead and continue to put AI front and center to actually get more feedback from users to be able to refine…”
“or they canot do that and face being disrupted”
“they can either integrate AI into their search and win that battle or they get incredibly heavily disrupted by these other players”
From the episode
Look Dumb or Get Disrupted: Google's Risky AI Overviews Launch