It started on a ski slope in Switzerland.
I asked an AI assistant to act as an experienced ski guide—designing a multi-day itinerary for a trip with my son. The brief was simple: optimize for great skiing based on weather conditions, with good lunch spots along the way.
What came back was impressive. Detailed routes, logical sequencing, thoughtful recommendations. It looked like the work of someone who knew the mountain intimately. – But on closer inspection, the cracks appeared.
Runs were stitched together from different parts of the mountain in ways that didn’t quite connect. The map looked convincing—but contained clear inaccuracies. It was only through careful review that the flaws became obvious.
That experience reinforced something we are seeing more and more at Meridian Veritas: AI outputs can look right—without actually being right.
The Hidden Risk
For mid-sized businesses, AI offers a step-change in capability. It enables smaller teams to move faster, operate smarter, and compete more effectively.
But there is a subtle risk.
AI systems can introduce small, hard-to-detect errors. And when those outputs are fed into other AI tools or workflows, the errors compound. One step builds on another, creating results that are polished—but flawed.
This is particularly relevant as businesses begin to adopt multi-step AI processes or agent-based workflows.
The danger is not obvious failure. It is quiet inaccuracy.
Why Human Oversight Is Essential
This is where Human in the Loop (HITL) becomes critical.
It means embedding human judgment at key points in AI-driven processes—where context, experience, and accountability matter most.
For mid-sized businesses, this is not about slowing things down. It is about ensuring that speed does not come at the expense of quality.
Even the most advanced AI systems—whether general-purpose or trained on proprietary data—operate within limits. They cannot fully grasp the nuances of your business, your customers, or your market dynamics.
Your people can.
The Compounding Effect
As AI tools become more connected, the risk increases.
One model generates an insight. Another refines it. A third turns it into a recommendation. If the starting point is slightly wrong, the end result can be significantly off—yet still appear credible.
Over time, identifying the root cause becomes increasingly difficult.
This is why a more deliberate approach is needed: controlled autonomy, where AI operates at speed, but within clearly defined boundaries and with targeted human oversight.
Getting It Right
Mid-sized businesses do not need complex transformation programs to manage this. They need focus and discipline:
- Build workflows with clear human checkpoints
- Train teams to challenge and validate AI outputs
- Focus oversight on high-impact decisions
- Use AI to augment expertise, not replace it
Done well, this builds trust—and trust enables speed.
Final Reflection
That ski itinerary generated by my expert AI ski Guide looked perfect—until it wasn’t.
AI is a powerful tool. But without human judgment, it can lead you confidently in the wrong direction.
For mid-sized businesses, the advantage will not come from AI alone—but from how effectively they keep humans in the loop.