---
id: "action-controlled-experiments"
type: "action-item"
source_timestamps: ["§ What to Do Instead"]
tags: ["measurement", "experimentation"]
related: ["concept-narrow-deep-use-cases", "framework-effective-ai-implementation"]
speakers: ["Thomas H. Davenport", "Laks Srinivasan"]
action: "Conduct controlled A/B experiments on narrow, deep use cases to measure actual AI productivity impacts."
outcome: "Accurate, data-driven understanding of AI's economic value and actual workforce requirements."
sources: ["execution"]
sourceVaultSlug: "hbr-seg-execution"
originDay: 8
articleStem: "hbr-foci-62-layoffs-ai-potential-not-performance"
sourceUrl: "https://hbr.org/2026/01/companies-are-laying-off-workers-because-of-ais-potential-not-its-performance"
sourceTitle: "Companies Are Laying Off Workers Because of AI’s Potential—Not Its Performance"
---
# Conduct Controlled AI Experiments

**Action:** Conduct rigorous, controlled A/B experiments on [[concept-narrow-deep-use-cases]] to measure actual AI productivity impacts *before* making headcount decisions.

**How:** Identify narrow and deep use cases (e.g., programming, customer service) that involve a limited number of jobs. Run workflows *with* and *without* AI in parallel and compare, so the productivity delta is isolated and attributable.

**Outcome:** Accurate, data-driven understanding of AI's economic value and true workforce requirements — the direct antidote to [[concept-ai-economic-value-measurement]] difficulty and to [[concept-anticipatory-ai-layoffs]].

Step 1 of [[framework-effective-ai-implementation]]. **Enrichment:** BCG explicitly recommends rigorous A/B testing for AI agents and workflows rather than superficial deployment.


## Related across articles
- [[concept-experimentation-trap]]
- [[concept-pilot-theater]]
- [[question-defining-ai-roi]]