Columbia Business School professors: What the Balogun red card can teach us about AI and judgment

The article uses a World Cup VAR/AI contrast (offside goal decided instantly vs. a later foul call that drew debate) to argue that AI will automate easy, measurable decisions but will concentrate human judgment on contested, higher-stakes calls. It notes that 44% of executives say they would override AI-based decisions, implying leaders must design clear decision-responsibility and trust protocols to avoid both automation bias and algorithm aversion.

Analysis

This is not a catalyst for model vendors so much as a signal that value migrates to workflow control, auditability, and exception handling. Once the easy calls are automated, the scarce economic asset is the layer that records why a decision was made, who approved it, and how liability is managed. That structurally favors enterprise platforms with embedded workflow and permissions, while commoditizing standalone “AI can decide for you” pitches.

The second-order winner set is broader than the article implies: identity, logging, data lineage, and policy enforcement should see more budget attachment as firms deploy AI in regulated processes. That is supportive for MSFT, NOW, SNOW, PANW, and CRWD over a 6-18 month budget cycle, because they sell the control plane around AI rather than the output. By contrast, BPO-style labor replacement stories and thin app-layer names face margin pressure if customers decide the machine can do the rote part but still needs expensive human escalation.

Contrarian view: the market may be overestimating the near-term revenue of “agentic” AI and underestimating the spend required to make it governable. If enterprises become more cautious, adoption could slow for pure-play AI names even as broader software budgets reallocate toward compliance and workflow. The falsifier is visible in earnings: if AI deployments scale without higher demand for security, observability, or approvals, this thesis is too cautious.

Key Decisions for Investors

• No standalone trade in TSTS until there is proof of product-level exposure to AI governance or decision workflow revenue; treat it as a watch item, not a position, over the next 1-3 months.
• Overweight MSFT and NOW on pullbacks as a 6-18 month basket: they monetize AI through workflow control and enterprise stickiness, with better downside protection than pure-play AI apps.
• Pair long PANW/CRWD vs short a basket of high-multiple AI app names over 1-3 months; thesis is that governance, identity, and logging spend expands faster than discretionary app-layer enthusiasm. Falsify if security budgets decelerate or AI adoption proves lower-friction than expected.
• If AI sentiment spikes again, consider fading C3.ai (AI) into strength rather than chasing it; the risk/reward is favorable only if bookings and operating leverage inflect, not on narrative alone.