---
id: "concept-ai-duplication-contradiction"
type: "concept"
source_timestamps: ["§ Effect #2: Duplication and Contradiction"]
source_url: "https://hbr.org/2025/09/dont-let-ai-reinforce-organizational-silos"
source_title: "Don't Let AI Reinforce Organizational Silos"
tags: ["data-silos", "model-conflict", "risk-management"]
related: ["concept-purpose-first-approach", "entity-western-pacific", "contrarian-universal-data-set", "question-resolving-model-contradictions"]
definition: "The phenomenon where isolated departmental AI models, trained on different data sets, produce conflicting strategic recommendations regarding the same business entities."
sources: ["tail2"]
sourceVaultSlug: "hbr-seg-tail2"
originDay: 2
articleStem: "hbr-tail-130-ai-reinforce-silos"
sourceUrl: "https://hbr.org/2025/09/dont-let-ai-reinforce-organizational-silos"
sourceTitle: "Don’t Let AI Reinforce Organizational Silos"
---
# AI Duplication and Contradiction

**Definition:** The phenomenon where isolated departmental AI models, trained on different data sets, produce conflicting strategic recommendations regarding the same business entities.

When departments operate in AI silos, they frequently use separate, non-overlapping data sets and models to evaluate similar business entities (like customers). This leads to conflicting conclusions that threaten unified business strategy.

The authors' case is [[entity-western-pacific]], a multinational bank where the finance department's risk-management AI (using traditional credit scores) flagged a customer segment as high-risk and to be avoided, while — simultaneously — the marketing department's AI (using digital behavior and social data) flagged the *exact same segment* as a prime target for acquisition. The organization received contradictory directives from its own fragmented intelligence systems, producing internal tension and strategic paralysis.

This is Effect #2 of siloed AI adoption. The authors' prescribed fix is strategic, not technical: shift to a [[concept-purpose-first-approach]] rather than mashing everything into a universal data lake — see the contrarian note [[contrarian-universal-data-set]] and the quote [[quote-purpose-not-process]]. The immediate tie-break governance mechanism during transition is left open — see [[question-resolving-model-contradictions]].