Intuition and gut feel still matter in the age of AI

Executives have been taught to fear bias in artificial intelligence. From recruiting algorithms that discriminate to predictive models that reinforce inequality, “bias” has become shorthand for systemic failure. Entire risk management programs now focus on finding and eliminating it.

But there’s another side to the story. In many business contexts, the human biases we dismiss as flaws — gut feel, pattern recognition, instinct — are not only useful but essential. They represent embedded expertise, hard-won intuition and domain-specific foresight. And when used wisely alongside AI systems, these biases can accelerate innovation, safeguard against blind spots and unlock strategic advantage.

The challenge is learning to tell the difference between destructive bias and constructive bias — and building organizations that can use the latter to guide AI rather than fight it.

When instinct beats the model

I remember sitting in a product review where the data said we were ready to launch. The models looked clean, the projections were green. But something in my gut told me otherwise. I pushed for a pause. Weeks later, in early consumer testing, my fear was validated — the product would have failed spectacularly. That moment cemented for me that bias, in the form of gut feel, isn’t always a bug in the system. Sometimes, it’s the safeguard.

Of course, I’ve seen the reverse too — teams that put blind faith in the data and overruled instinct. One reformulation project went forward because “the numbers said it would work,” only to collapse in factory trials when the texture turned gritty at scale. That’s the cost of ignoring experienced bias.

The science behind good bias

Researchers have long argued that bias isn’t always harmful:

• Bounded rationality: Herbert Simon introduced the idea that humans make decisions under constraints of time and information, using heuristics — rules of thumb — to reach “satisficing” solutions rather than perfect ones. This foundation has been extensively discussed in modern decision theory.
• Fast and frugal heuristics: Gerd Gigerenzer and colleagues argue that in uncertain environments, simple heuristics often outperform complex models. Intuitive judgment is not caprice, but adaptive reasoning.
• Automation bias caution: In human-AI systems, there is a well-documented tendency for people to over-rely on algorithmic recommendations, sometimes ignoring contradictory evidence. Studies in domains from health care to aviation show that automation bias can lead to serious errors.

Even in AI research, constructive bias is central. Machine learning often uses inductive priors — assumptions about structure that guide learning. Expert bias, when sound, can act as that prior, steering AI toward plausible or ethical solutions.

The bias compass

To make this distinction practical, I use what I call the bias compass:

• Constructive, forward-looking bias: Gut feel about future risks, emerging trends or untested formulations. Guides AI into unexplored but high-potential territory.
• Constructive, backward-looking bias: Legacy heuristics that no longer fit today’s context — for example, assuming old consumer preferences still apply.
• Destructive, forward-looking bias: Over-caution. Useful in food safety or financial compliance, but dangerous if it blocks innovation outright.
• Destructive, backward-looking bias: Prejudice, favoritism and systemic inequity. These biases damage trust, erode performance and must be actively eliminated.

The compass helps leaders distinguish when bias is serving as foresight versus when it’s acting as a blindfold.

Cross-industry applications

1. Consumer packaged goods (CPG): Navigating complexity

In CPG, intuition often saves millions. In one case, a reformulated SKU looked flawless in R&D and in AI-optimized simulations. But one scientist flagged a nagging concern: “This will clog the factory lines.” Leadership listened. Sure enough, early line tests confirmed the issue. That gut call kept us from a multimillion-dollar retooling crisis.

On the flip side, I’ve seen backward-looking bias cripple companies. A well-known brand refused to shift away from a legacy SKU, convinced its “classic” status guaranteed loyalty. Within two years, they lost double-digit share to agile competitors who leaned into plant-based innovation.

2. Financial services: Guarding against tail risk

Risk models excel at processing historical data, but markets are shaped by rare, high-impact events. I once saw a risk model downplay geopolitical shocks as “noise.” A senior manager overruled it, overweighting tail risk scenarios. That bias paid off — their portfolio absorbed far less damage than peers when the shock materialized.

But I’ve also seen destructive bias. On one trading desk, confirmation bias kept analysts clinging to a position long after evidence shifted. The AI models had already flagged the decline. The loss ran into nine figures.

3. Technology: Balancing engagement with trust

In tech, bias often shows up in product instincts. I once worked on a machine-learning-powered junk email detection system where a new feature was predicted to dramatically improve filtering. The data looked perfect. But the UX team flagged a gut concern — it would create false positives for legitimate customer messages. They were right. Had we launched, the backlash would have been immediate.

Status quo bias is just as costly. I’ve seen teams resist deploying AI-driven workflow automation simply because “this is how we’ve always done it.” That bias delayed efficiency gains by years.

Building organizations that harness good bias

How can leaders institutionalize the productive side of bias while filtering out the destructive?

1. Name it. Create language in your organization to distinguish constructive vs. destructive bias.
2. Capture expert intuition. Build human priors into AI systems — for example, encode R&D constraints or trader “red flag” conditions into modeling engines.
3. Design human-AI workflows. Don’t expect AI to replace judgment; build processes where experts can override, redirect or refine outputs.
4. Audit backward-looking bias. Regularly test for legacy heuristics or prejudices that no longer serve.
5. Reward foresight. Promote employees who make gut calls that anticipate reality — not just mirror data.

What leaders should do Monday morning

Executives don’t need another abstract lecture on bias. They need actionable steps:

• Ask: Does this bias point us forward (toward future risks/opportunities) or backward (anchored in past assumptions)?
• Balance: Use bias to set hypotheses, then let AI test and validate them.
• Institutionalize: Capture tacit expertise before it walks out the door. Incorporate it into your AI workflows.
• Challenge: Don’t let constructive bias turn into dogma. Continuously recalibrate with fresh data.

The strategic takeaway

The hardest leadership call I ever made was a release decision. The data said go, but my instinct — and experience with past launches — said the risks weren’t fully captured. I pulled the plug. It was unpopular at the time, but it prevented a reputational hit we’d still be cleaning up today.

For me, the rule is simple: If I don’t trust the source of the data or understand how it found its way into the processing engine, I lean on gut feel. Bias becomes my safety net against blind faith in black-box outputs.

AI is extraordinary at exploiting the past. But competitive advantage comes from anticipating the future. Human bias — when constructive and forward-looking — offers exactly that.

Whether you’re formulating beverages, managing portfolios or scaling new platforms, your gut feel is not a liability. It’s an underutilized asset. The task for leaders is not to eliminate bias altogether, but to harness it — using the bias compass as a guide — to innovate faster, manage risk smarter and earn the trust of consumers, regulators and markets alike.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?