Over the last few years, we’ve watched companies use AI with very different results. Some initiatives have strengthened operations. Others have created expensive lessons.
In 2021, Zillow shut down its iBuying business after recording more than $500 million in losses tied to the unit. The company had built machine-learning models to estimate home values and automatically generate purchase offers at scale. The approach worked under certain market conditions. But when housing dynamics shifted, Zillow ended up owning thousands of homes it couldn’t resell profitably.
The system didn’t fail in a dramatic technical sense.
It simply didn’t hold up under real financial pressure.
In 2024, Air Canada lost a ruling after its website chatbot gave a passenger incorrect information about a bereavement-related refund or discount. The airline argued that the chatbot should not be treated as an authoritative source. The tribunal disagreed. If a system operates on your platform, you are responsible for what it communicates.
The software was functional.
The accountability structure wasn’t fully thought through.
Now look at two examples where AI has become part of day-to-day operations.
Netflix uses its recommendation engine to shape what people watch and how long they stay on the platform. Most viewing activity is driven by those recommendations. That matters because engagement and retention directly affect subscription revenue. The system is not impressive because it sounds advanced. It’s valuable because it influences churn and lifetime value.
UPS offers a similar story, but in a different context. Its ORION route optimization system guides how drivers move through their day. By optimizing routes across its network, UPS has reduced fuel usage, lowered emissions, and saved hundreds of millions of dollars annually. There’s no spotlight on the technology itself. The measurement is practical: fewer miles driven, lower operating costs.
Same period.
Similar category of technology.
Very different outcomes.
The difference isn’t how sophisticated the models are. It’s whether they are tightly connected to business realities.
The Question That Often Gets Overlooked
When AI projects begin, conversations usually focus on the technical side.
Is the model accurate?
Is it better than the previous version?
Is the infrastructure modern?
Those are reasonable questions. But they don’t answer the one that matters most.
Are we actually better off than before?
Are we making fewer costly mistakes?
Are decisions more reliable?
Are costs lower in ways that show up in financial results?
Is risk more controlled, especially when conditions change?
If those answers are unclear, the AI may be active, but it isn’t necessarily adding value.
What These Examples Tell Us
Zillow’s experience shows what can happen when predictive systems scale faster than financial safeguards. The issue wasn’t that machine learning is flawed. It was that exposure increased under volatility, and the business couldn’t absorb the downside.
Air Canada’s case shows that automation does not remove responsibility. Deploying a system without clearly defining ownership of its outputs creates risk, not efficiency.
Netflix and UPS, on the other hand, tied their systems directly to outcomes leadership already tracks closely: retention, engagement, cost per delivery, fuel usage. If those numbers move in the right direction, the AI is working. If they don’t, it’s re-examined.
There’s nothing dramatic about that. It’s disciplined management.
A Practical Way to Think About It
If you’re evaluating an AI initiative, a few grounded questions help cut through the noise:
- Are decisions better, not just faster?
- Has one measurable source of waste or error actually declined?
- Does the system reduce downside risk, especially under stress?
- Is it clear who owns the consequences when something goes wrong?
- If you turned it off tomorrow, would performance noticeably suffer?
If you can answer those confidently, you likely have something real. If not, it may still be early or misaligned.
AI does not need attention to prove its value.
When it works well, it becomes part of the operating fabric. It supports decisions quietly. It improves margins gradually. It stabilizes processes over time.
That is usually how you know it’s doing its job.
A Note from BRUKD Consultancy
At BRUKD, we work with organizations that want to think carefully before placing AI into core operations. The goal isn’t rapid adoption. It’s clarity understanding where AI strengthens decision-making and where human authority needs to remain clearly defined.
If you want a structured way to assess your AI readiness, you can start here:
https://brukdconsultancy.com/assessment.html
If you would prefer to talk through your situation directly, you can reach us here:
https://brukdconsultancy.com/contact.html