Deleting the Outliers and The Statistical Trap Impacting Corporate AI Transformation

By Suzanne (Sues) Tonks

In quantitative statistics, outliers are an inconvenience. They skew the mean, distort the regression line, and muddy the waters of statistical significance. To achieve a clean, predictable dataset, the standard practice is often simple. Delete them. In a lab, this creates elegance. In an organisation, it creates a trap.

Years ago, I stood on a TEDx stage to deliver a talk on a concept that had long troubled me from both a psychological perspective and as a corporate advisor. It's the cultural obsession with silencing the anomaly. Before diving into deep psychological research, my career was anchored in commercial analytics, Public Relations, and high-stakes issues management for some of Australia's largest organisations and global brands.

In those pressure-cooker environments, I was often the one brought in to do the polishing. To smooth over the corporate messaging, manage the reputational risk, and align everyone to a unified, predictable script.

I’ve sat in the boardrooms of both low-performing and exceptionally high-performing senior executive teams. The difference between them rarely comes down to raw intelligence. It comes down to how they handle variance. Low-performing teams delete their outliers to preserve a false sense of comfort. High-performing teams lean into the friction.

The AI Tsunami and the Danger of the "Mean"

This data-cleaning mindset has reached a critical, dangerous flashpoint. We are currently facing an emerging tsunami of change driven by Generative AI.

The corporate temptation right now is to use AI to optimise for the middle of the bell curve. To automate standard processes, predict normative behaviours, and drive rapid efficiency.

In doing so, organisations are inadvertently building massive systems based purely on neurotypical or normative outputs.

But predictability is the enemy of innovation, and blind efficiency is the precursor to systemic failure.

To navigate a world disrupted by generative AI, organisations don't just need fast algorithms. They need atypical brains. They need the outliers who can look at a machine’s output and ask the uncomfortable, non-linear questions that the algorithm is blind to.

Focusing on where the answers hide (or at least, the really good questions)

In my current qualitative PhD research focusing on identity formation, anchored in Erik Erikson’s psychosocial developmental stages, I deliberately chose qualitative methodology over quantitative metrics. Why? Because you cannot understand the deep mechanics of identity or transformation by looking at the statistical mean. You have to listen to the marginalised, outlying experiences.

Erikson theorised that identity is forged through conflict and resolution. He called it Identity vs. Role Confusion. The same is true for corporate AI strategy.

A mature, secure organisational identity cannot be automated. It is forged through the healthy friction of human cognitive diversity.

Yet, as leaders race to implement AI, they face a distinct behavioural trap. They only listen to the listenable.

We have trained corporate environments to value highly polished, conforming communication. But the person in the room who can spot the downstream human impact of AI use, or the subtle ethical flaw in an algorithm or output, is often an outlier. They might be a neurodivergent thinker or a challenger whose communication style doesn't fit into a neat corporate script.

When a leader dismisses that unpolished skepticism, they are doing the organisational equivalent of deleting an outlier from a dataset. They are cleaning the data at the expense of safety, ethics, and truth.

How do we Measure the Discerning Leadership Required?

AI requires more than technical literacy. It requires ethical agency, system fluency, augmented command and emotional intelligence. It demands leaders who are cognitively flexible enough to manage human-machine collaboration without losing their moral discernment.

The qualitative insights coming out of our discussions with clients and leadership community at Limestone Group have highlighted a stark reality. There are some razor sharp minds out there. However, most senior teams are unequipped for the psychological and systemic complexities of this shift. While we are currently compiling this data for a peer-reviewed journal article, the immediate need for organisations to act. And now.

Out of a demand for leaders to understand exactly where they stand in this changing AI landscape, the AI Management Positioning Inventory, the AIMPI was developed. This diagnostic screener doesn't measure technical coding skills. It measures the core management functions relating to AI including ethical agency, to identity a leaders position and ability to look down the line at the human impacts of technological AI transformation.

It helps current and future leaders self-reflect and identify whether the leadership culture is designed to listen to the critical outliers who protect the organisation. Or if leaders are simply polishing the way toward a crisis.

Have your say and know your position. I invite you to contact us directly via the contact page or through the leadership lab to be sent access to the AIMPI Assessment Screener.

It is certainly food for thought. And we need our thinking leaders on this.

Warmest,

Sues