A friend who runs recruitment at a 300-person company called me last month, slightly proud, slightly worried. They'd just finished their first quarter using a new AI sourcing stack. Funnel volume up 4x. Recruiter activity through the roof. Time-to-screen down by half. Leadership was thrilled.

Two questions in I knew they were in trouble.

What's your offer acceptance? Hovering around 60%, give or take.

What's your 12-month retention on the hires you made this year? Honestly, we don't really track that.

So the AI is firing on all cylinders, the dashboard looks great, and you have no idea whether any of it is producing better hires. Right?

Long pause on the call.

This is the quadrant I think about more than any other. I've taken to calling it Building on Sand, because that's what it is. A team using AI aggressively, with metrics that look good on the surface, sitting on top of an operation that wasn't reliable to begin with. The AI didn't break the operation. It just made the breaks invisible by making everything fast.

Here's the thing that took me a while to internalise: bad hiring decisions don't show up in your dashboard for nine to eighteen months. The person you hired in March who looked great on paper, sailed through your AI-screened pipeline, got an offer in five days because the system was so efficient — they're going to be a performance issue in October. Maybe November. By the time you can see it, you've already made forty more decisions using the same broken process. The bill arrives later, all at once, and it's enormous.

I keep thinking about this in terms of food. Imagine your kitchen has rats. You buy a really nice oven. Faster, more efficient, better temperature control. The meals come out beautifully. Faster than before. You're cooking three times as much in the same hours. The dinner parties go great.

The rats are still there. The kitchen has just become a more productive rat habitat.

This is the dynamic. And here's why most people in this quadrant don't see it.

The metrics that AI improves are the ones at the top of the funnel. Speed, volume, throughput, recruiter activity. The metrics that would tell you the operation is broken are the ones at the bottom: quality of hire, retention, manager satisfaction, offer acceptance trends over time. Almost nobody tracks the bottom-of-funnel metrics with the same discipline they apply to the top. So the AI lights up the parts of the dashboard you watch, and the parts you don't watch are where the damage is happening.

I told my friend on the call: pause the AI for sixty days. Audit your last twenty hires. Find the ones that aren't working out. Trace each one back through your process and find the moment where, if a human had been paying full attention instead of trusting the workflow, they would have caught it.

You'll find a pattern. Everyone does. It's usually the same two or three failure modes repeated. Once you can name them, you can put guardrails in. Then you can turn the AI back on.

Will they actually do it? Probably not. The cost of admitting you're in this quadrant is high. Leadership has been told the AI investment is paying off. The funnel metrics are good. Stopping to ask "but are the decisions good" is a deeply unwelcome conversation.

Which is exactly why most companies in Building on Sand stay there until the retention data forces the question, by which point a lot of damage has been done.

If you read this and felt a flicker of recognition, that's the signal. Don't ignore it.