The Hypothesis

Do AI recommendations reduce decision fatigue in teams — or do they create a new cognitive burden?

The Concept

AI tools love to make recommendations. But more recommendations means more decisions. This experiment measures whether surfacing AI-generated suggestions actually reduces decision fatigue in teams — or whether it creates a new cognitive burden where people spend time evaluating AI suggestions instead of thinking for themselves.

Decision fatigue

The hypothesis

Do AI recommendations reduce decision fatigue in teams — or do they create a new cognitive burden?

The concept

AI tools love to make recommendations. But more recommendations means more decisions. This experiment measures whether surfacing AI-generated suggestions actually reduces decision fatigue in teams — or whether it creates a new cognitive burden where people spend time evaluating AI suggestions instead of thinking for themselves.

How it works

1. Team uses product normally
2. Baseline measurement — decision speed, confidence, reversal rate
3. Enable AI suggestions — recommendations surface in UI
4. A/B compare — with suggestions vs. without
5. Measure fatigue indicators — time-to-decision, confidence score, override rate

Controlled comparison — same teams, same tasks, with and without AI recommendations.

What it explores

• Do AI recommendations speed up decisions or slow them down with evaluation overhead?
• Is there a “suggestion threshold” beyond which more recommendations hurt?
• Do teams trust AI recommendations more in high-stakes or low-stakes decisions?
• Does decision quality improve with AI input, even if speed doesn’t?
• How does recommendation confidence level affect user acceptance?

What we found

• For routine decisions (email subject, send time), AI recommendations reduced decision time by 40%
• For strategic decisions (budget allocation, campaign direction), recommendations increased time-to-decision by 15%
• Teams debated whether to follow AI suggestions or their own instincts on strategic choices
• 4pm decisions were 23% slower than 10am decisions regardless of AI input

Learnings

• AI recommendations work best when they reduce the decision to accept/reject rather than adding options — more choices compounds the problem.

• Confidence scores on recommendations dramatically affect adoption — 90%+ confidence gets followed, below 70% gets ignored.

• The right pattern is: automate routine decisions, advise on strategic ones, stay silent on creative ones.

Where it goes next

These findings are shaping how Lens presents its recommendations. Instead of always surfacing suggestions, Lens will learn when to recommend, when to automate, and when to stay quiet — based on decision type and user state.