It’s a conversation I’ve had many times:
Leader:: “We’re time-poor. Can’t we just automate that?” Me: “We can. But should we?”
From auto-granting contractor site access, to automatically closing permits, or letting AI “see” hazards, the pressure to automate every people step is real. And it’s being amplified by powerful AI marketing promising speed, cost savings, and risk reduction.
But here’s the core message... “just because you can automate something doesn’t mean you should” - especially in safety-critical work.
The Case for Caution: Where Automation Can Introduce Risk.
There’s no question that automation can make operations more efficient. But when we push automation too far, especially into areas requiring judgement, oversight or ethical decision-making, we can create dangerous blind spots. Let’s look at why...
Automation Can Bypass Critical Checks.
Imagine a contractor finishes an online induction and the system automatically grants them the role to sign into a site. Sounds efficient, right?
But who checked that:
- Licences are valid
- Insurances are current
- They’re not on a safety incident blacklist
Or consider automatically closing permits at the end of a shift... who verified that no-one was still in a confined space or still working on site?
These aren’t hypotheticals... regulators like OSHA regularly cite confined-space deaths where final human checks were skipped or assumed to be “handled by the system.” In addition, there is the Therac-25 case, where software removed safety interlocks, which led to six patient deaths, all because human oversight was designed out.
The Hardest Steps to Automate Are Often the Ones That Shouldn’t Be.
The steps leaders most want to automate are usually the ones that are technically the hardest to automate – precisely because they require human intervention or oversight.
Why? Because coding the rules to cover every possible combination of scenario is:
- Exceptionally complex
- Time-consuming
- Expensive
In fact, the effort often makes the solution prohibitively expensive, meaning it’s never rolled out. The result is that instead of improving processes, businesses remain stuck with no enabling technology at all – losing the chance to capture any of the benefits that well-designed, balanced technology could deliver.
The takeaway is clear... Smart automation focuses on supporting human decision-making, not trying to replace it altogether.
Predictive Analytics Needs Strong Foundations.
Another common pitfall is jumping into predictive analytics and AI insights before your organisation has the data maturity to support them.
Incident reports, hazard logs, and maintenance records are often:
- Heavily manual
- Incomplete
- Error-prone
Layering predictive models over shaky data doesn’t give you actionable insights, it gives you false confidence. Harvard Business Review puts it bluntly: “If your data is bad, your machine-learning tools are useless.” Deloitte reinforces that vein, commenting that without disciplined baseline data, predictive systems will not deliver meaningful value.
Let’s Not Overstate the Risks.
At this point, you might think I’m anti-automation... I’m not. There’s a powerful counter-argument we need to acknowledge to make this discussion balanced and credible.
Humans Are Error-Prone, Too.
It’s easy to criticise automation failures. But let’s not pretend human decision-making is flawless.
People:
- Get tired
- Get distracted
- Carry biases
- Make inconsistent or emotional choices
McKinsey research shows that well-implemented automation can reduce human error rates by up to 80%, especially in routine, repetitive tasks. In fact, automating rule-based steps can rapidly improve safety by ensuring consistency and reliability.
Speed and Scale Matter.
In fast-moving industries, waiting for human approvals can slow operations dangerously.
- Time-sensitive shutdowns delayed by permit bottlenecks
- Urgent contractor onboarding held up by manual checks
- Hazard alerts stuck in review while conditions change on the ground
AI systems process vast amounts of data quickly and more accurately than humans can, making it possible to scale risk management in ways manual systems never could.
AI Is Getting Better at Handling Imperfect Data.
Yes, predictive models rely on good input data – but today’s machine learning systems are increasingly able to:
- Handle incomplete datasets
- Identify outliers and errors
- Learn and improve over time
Sometimes, automation isn’t just a shortcut – it’s a necessary evolution to manage the complexity and volume of modern operations.
Regulations Are Evolving to Embrace Automation.
While many regulators still stress human oversight, we’re also seeing growing acceptance of automated systems – especially when they’re well-designed and rigorously monitored.
The Smarter Path: A Maturity-Based Balance.
So where does this leave us? Not in a simplistic “automation good” vs “automation bad” debate. The most successful organisations take a balanced, maturity-driven approach.
Early stages: Focus on digitising capture, validating data, and surfacing dashboards for human review.
Mid-maturity: Introduce automation for rule-based steps, but keep manual approvals for high-risk actions.
Advanced stages: Leverage predictive models, with human override points.
Optimised operations: Move to closed-loop automation only where data quality, system maturity, and risk context make it safe and appropriate.
Deloitte calls this journey “Data Mastery”, showing that mature firms are three times more likely to extract real value from analytics.
The Unifii Perspective: Augment, Don’t Replace.
At Unifii, we design systems that:
- Automate the heavy lifting: data capture, validation, alerts
- Surface the right information to the right people, at the right time
- Enable leaders to make informed decisions confidently and efficiently
It’s not about chasing the shiniest AI feature. It’s about building trust, accountability, and adaptability into the heart of your operations.
Five Questions Every Leader Should Ask Before Automating.
- Could a wrong decision here hurt someone or the business?
- Is our underlying data complete, accurate, and timely enough to trust?
- Will removing this check degrade skills or accountability?
- What’s the recovery plan if the automation fails?
- Does this automation align with our current maturity or are we jumping ahead?
If you can’t confidently answer these, it’s a sign to slow down, reflect, and design more thoughtfully.
From Compliance to Confidence.
Automation and AI are powerful tools – but they’re not magic wands. In safety-critical work, the strongest organisations take a maturity-aligned approach:
This is how you move beyond ticking boxes and truly build operational excellence.
Until next time, stay safe, lead well and remember that the best systems are the ones that make humans smarter – not the ones that replace them.
Cheers,
Paul
