Building AI systems at scale demands diverse thinking. Here is what that looks like in practice.
At first glance, the fact that roughly 77% of AI developers are male looks like a diversity issue. In practice, it is something far more fundamental: it is a product quality issue.
When the majority of people building AI systems share similar backgrounds, experiences, and assumptions, those systems tend to reflect an insular world view. The result is not intentional bias but something more subtly corrosive: blind spots in development that only become visible once products are deployed at scale.
That matters in any AI application, of course, but it’s even more of an issue when those systems are making safety-critical decisions in the real world.
Geotab’s platform processes data from millions of vehicles in more than 160 countries. The AI models underpinning that platform reduce risk, improve driver safety, and optimise fleet performance across vastly different operating environments. A model that performs well for one set of drivers but fails for another is more than an abstract concern about diversity and inclusion – it is a product defect.
Taking a diverse approach
This is the context in which we build AI systems at scale. The focus should not be on representation for its own sake but on ensuring that systems reflect the complexity of the real world. That is not a value statement – it is how you build better products.
Achieving it starts with recognising the source of the blind spots. Training data can never truly be complete: patterns that appear robust for one fleet type, or across one driving culture, can break down elsewhere. Edge cases – unusual, infrequent, and context-specific – are where AI systems are most likely to fail, yet those are often the scenarios that matter most in safety-critical environments.
Ironing out those failures depends on the people building and testing the models – specifically, their ability to question assumptions, challenge conclusions, and spot what others might miss. That is where team composition itself becomes a product variable.
It’s important to have the right people present when design decisions are made, long before a model is deployed. Teams that bring a broader, more diverse range of perspectives – technical experience, cultural backgrounds, cognitive processing – are more likely to interrogate data differently, to ask a wider range of questions, and to recognise when a model might be performing well overall but underperforming for specific sets of users. If problems are framed in an echo chamber, blind spots become baked in.
My team embeds that thinking into how Geotab’s systems are developed and deployed. Our ‘Responsible AI Commitment’ includes mechanisms such as explicit fairness metrics, red teaming, and bias auditing to pressure-test models in development.
Geotab places as much emphasis on who is in the room as it does on the processes those teams follow. The aim is not to meet quotas or satisfy external scrutiny, but to build teams that see problems from multiple angles – and, crucially, to challenge each other before assumptions become product behaviour.
The Business Case
AI systems that generalize well across different users, regions, or operating conditions reduce risk, generate trust, and ultimately drive growth. It creates systems that are more resilient, more adaptable, and better suited to global deployment. For a company operating at Geotab’s scale, that is not a theoretical advantage. It is a competitive one.
The AI industry now has a choice: it can continue to build systems in relatively narrow contexts, addressing issues only after they emerge in the field; or it can take a more deliberate approach – investing in teams, perspectives, and “broad lens” engineering processes that highlight blind spots early and build robustness from the ground up.
At Geotab, we have made our choice: to build AI that works for everyone by building better products that perform reliably in the complexity of the real world.
And in a world where AI increasingly shapes how businesses operate and how people stay safe, that is not just good practice. It is an essential goal.
