---
id: "concept-cost-of-errors"
type: "concept"
source_timestamps: ["§ Where and When to Use Generative AI"]
tags: ["risk-management", "decision-making", "ai-safety"]
related: ["framework-gen-ai-deployment", "concept-no-regrets-zone", "concept-human-first-zone"]
definition: "The severity of the financial, legal, reputational, or personal consequences that would occur if an AI system makes a mistake on a specific task."
source_url: "https://hbr.org/2025/11/the-gen-ai-playbook-for-organizations"
source_title: "The Gen AI Playbook for Organizations"
sources: ["agentic"]
sourceVaultSlug: "hbr-seg-agentic"
originDay: 6
articleStem: "hbr-cl-87-genai-playbook-orgs"
sourceUrl: "https://hbr.org/2025/11/the-gen-ai-playbook-for-organizations"
sourceTitle: "The Gen AI Playbook for Organizations"
---
# Cost of Errors

The **Cost of Errors** is one of the two foundational dimensions (the vertical axis) of the [[framework-gen-ai-deployment|Gen AI deployment framework]]. It shifts the organizational question away from *"Is the AI intelligent enough?"* to *"How severe are the consequences if the AI makes a mistake?"*

**High cost of errors** — an error would lead to serious financial loss, legal liability, reputational damage, or physical/emotional harm. Examples: misdiagnosing cancer, mishandling a vulnerable psychotherapy patient, or hiring a toxic executive. In these domains firms must keep humans firmly *in the loop* (see [[concept-quality-control-zone]]) or *at the center* of the decision (see [[concept-human-first-zone]]).

**Low cost of errors** — a mistake carries limited risk. Examples: a course-evaluation summary missing a minor nuance, or a preliminary resume screen overlooking a marginal candidate. These are ideal candidates for immediate, autonomous or semi-autonomous deployment (see [[concept-no-regrets-zone]] and [[concept-creative-catalyst-zone]]).

This axis maps directly onto emerging **risk-based AI governance** frameworks (e.g., the EU AI Act's risk tiers), which likewise treat the *impact of an error* — not raw model capability — as the primary determinant of how much human oversight a deployment needs. Cost of errors is assessed task-by-task after you [[action-deconstruct-jobs|deconstruct jobs into component tasks]].