Does employee monitoring improve behaviour? The research

Larry Ellison wants cameras everywhere. At an Oracle analyst meeting, he said AI would soon process every feed we produce - dashcams, doorbells, police body cams, security systems - and that this would be a good thing: “citizens will be on their best behavior, since we’re constantly recording and reporting everything that is going on.”

You don’t have to run a country to recognise the pitch. The same logic gets made to you about your own business: more visibility into what your team is doing will make them behave better. AI has made that kind of employee monitoring cheap in a way it never was before - every call transcribed and scored, every screen tracked, every message searchable. The pitch is always the same shape: more visibility means better behaviour. Fewer mistakes, more accountability, tighter execution. It sounds obviously true. It is only true in a narrow, specific way, and the gap between that narrow truth and the broad claim is exactly where a lot of monitoring decisions go wrong.

Every growing company hits this question eventually: how much do you instrument, and how much do you trust? Informal oversight stops working at some point, and someone proposes a system to close the gap. The instinct to reach for more visibility is not wrong. But it usually arrives with an assumption baked in that the research does not support. The tool doing the watching has changed. The psychology of being watched hasn’t.

Where it works

Watching people does change behaviour - that part holds up. A 40-year meta-analysis of 161 CCTV studies found roughly a 13% drop in crime in monitored areas compared to unmonitored ones. But look at where that effect lives. It’s strongest in car parks. It’s weakest against anything premeditated or high-stakes. It depends on the camera being actively monitored, not just present, and it grows stronger the longer the system runs, taking years to build its full deterrent reputation before it starts to fade.

It only works on one kind of behaviour: opportunistic, low-commitment stuff, in a space people pass through, watched by someone who can act on what they see. It is not “surveillance makes people behave better,” full stop. It’s “surveillance suppresses a narrow category of behaviour, under specific conditions, for people who have no ongoing relationship with the space being watched.”

Britain is the test case. London alone has more than 130,000 public cameras, built up over decades. When researchers compared camera counts against crime levels across 150 major cities worldwide, they found barely any correlation. More cameras hasn’t reliably meant less crime - which fits: most of us stopped consciously registering the cameras years ago, and a camera that’s stopped registering as a camera can’t be doing much deterring.

Where it breaks down

Your team is not a car park. They have an ongoing relationship with the space being watched, and that’s exactly where the research reverses.

The clearest data on sustained monitoring of the same people over time comes from workplace surveillance studies, and it points the opposite direction from Ellison’s claim. Heavy observational monitoring erodes trust. It raises stress and lowers engagement. When people know monitoring data will be used for control - performance review, discipline - they don’t tighten up, they disengage, and counterproductive behaviour goes up, not down. One in nine workers in a recent US survey said they’d quit a job specifically over excessive monitoring. Candidates rate heavily monitored employers as less ethical before they’ve even taken the job.

The mechanism is not complicated once you see it. Good behaviour at work mostly isn’t an opportunistic snap decision the way shoplifting is. It’s a function of whether someone believes the system around them is fair, whether they trust their manager, and whether they’re bought into the outcome. Monitoring can suppress the visible symptoms of a trust problem for a while. It does nothing for the trust problem itself, and over time it makes it worse, because the message received is “we don’t believe you’ll do this without being watched” - which is a hard message to build discretionary effort on top of.

What this means for how you run things

Ellison’s question is “will this make people behave better.” That’s the wrong question. The real one is: what am I trying to stop - and is it small and opportunistic enough that just being seen will stop it?

I’ve tracked time before - not company-wide, just for teams deploying campaigns for clients. It was necessary for a plain operational reason: that kind of work is sold and staffed by time, and if you don’t know where the hours go, you can’t cost an engagement properly, catch scope creep before it eats the margin, or plan capacity as you take on the next account. That’s a narrow, specific problem. Tracking time solved it.

What made it work was being upfront about it before it started, not after someone asked why. The team was told plainly what was being tracked and why - to cost the work correctly and plan who was free for what, not to check up on anyone. It was said just as plainly what the data would never be used for: it was never the input for a performance conversation, and nobody’s time-tracking numbers were ever compared against anyone else’s.

None of that works without trust already in the room. Introduce the same tracking into a team that doesn’t already trust its managers, and it stops being a costing tool and becomes surveillance - a reason to perform for the system instead of doing the work. The tool didn’t create the trust. It depended on it being there first.

I never needed the data to spot someone underperforming, either. If someone was struggling, I already knew - from the work itself, long before a timesheet would have told me anything. Most managers close enough to the work already know this. The data answered a costing and capacity question, not a competence question. Use it for the second one and you’ve built exactly what Ellison is selling.

That’s what visibility is good for - a narrow, named problem, communicated openly, inside a team that already trusts you.

It’s a poor fit for most of what worries a CEO about how their team behaves - whether people are engaged, whether they’re making good judgment calls when no one’s checking, whether they’d flag a problem early instead of hiding it until it’s expensive. None of that is opportunistic behaviour you can suppress with a camera. It’s trust behaviour, and trust behaviour runs on the opposite input: less control, more clarity about what’s expected and why, and evidence that the system is fair even when nobody’s watching.

Ellison is describing a world where the answer to “how do people behave well” is always more recording. Most of what determines whether your team behaves well when you’re not in the room was never going to be solved by recording more of the room.

If you’re weighing a monitoring or oversight decision right now and want to work through what it’s really solving for, that’s the kind of question an advisory session is built for. Book a 20-minute call →

FAQ

Does CCTV actually reduce crime? Modestly, and only in specific conditions. A 40-year meta-analysis of 161 CCTV studies found roughly a 13% drop in crime in monitored areas - strongest in car parks, weakest against anything premeditated, and dependent on the camera being actively monitored rather than just present.

Does monitoring employees improve their performance? The research says the opposite for sustained monitoring. It erodes trust, raises stress, lowers engagement, and increases counterproductive behaviour when the data is used for control - performance review or discipline - rather than a specific operational purpose.

When does workplace monitoring actually work? When it solves one narrow, named problem - like costing or capacity planning - is explained openly before it starts, is explicitly ruled out as an input for performance decisions, and is introduced into a team that already trusts its managers. Without those conditions, it stops being a monitoring tool and becomes surveillance.