Imagine you are sitting in your Monday review and the latest AI productivity dashboard appears on the screen. Response times are down. More customer conversations are being logged. Meeting summaries are being generated automatically, and use of the company’s approved AI tools is well ahead of target. Almost every indicator is green, so you should be feeling rather pleased.

Then you ask what has changed for the business. Customer retention is flat. Sales conversion has barely moved. Decisions are not being made any faster. The dashboard says people are doing more, but you cannot quite see where the additional value is showing up.

As you probe further, a few patterns emerge. Customer service teams are answering the simplest queries first because response time is visible. Employees are routing routine tasks through the approved AI tool because adoption is being tracked. Meetings sound more polished now that everyone knows a transcript can be analysed later.

So, this week, let’s revisit an old management idea that has suddenly become much more important. What happens when the Hawthorne Effect meets AI? As more of our work becomes visible, searchable and measurable, will people become better at their jobs, or simply better at producing the signals that the system recognises?

 

A Useful Idea Hidden Inside a Messy Experiment

The Hawthorne studies took place at a Western Electric factory during the 1920s and 1930s. The version many of us learnt in business school was wonderfully neat. Researchers changed the lighting and other working conditions, productivity improved, and the conclusion was that employees perform better when they know somebody is paying attention.

The underlying research was far less obliging. Later scholars questioned whether there was a single, consistent Hawthorne Effect at all, suggesting that the productivity gains could be explained by several other factors: changes in supervision, group dynamics, seasonal effects, novelty, and the meaning employees attached to the experiment. Management folklore took a complicated experiment and gave it a much cleaner ending.

Rethinking the Hawthorne studies actually makes them more interesting, not less. The enduring lesson isn’t that observation automatically improves performance. Perhaps the deeper insight is this: observation isn’t neutral, nor does it operate in a vacuum. People don’t respond simply to being watched. They respond to what the watching represents. Does being observed make them feel valued? Supported? Controlled? Threatened? What’s more, observation doesn’t happen in isolation. When companies introduce new forms of monitoring, they often change the wider environment in which people work.

Imagine a manager introduces an AI productivity dashboard. After six months, output improves. A simplistic Hawthorne interpretation would say: “Output improved because employees knew they were being watched.”

A more rigorous interpretation would ask:

• Did the dashboard make priorities clearer?
• Were the metrics used for support or evaluation?
• Did managers become more engaged during this time?
• Did employees receive better feedback?
• Were additional resources provided?

The improvement in performance may come from the whole intervention, not observation alone. That distinction becomes critical as workplace monitoring expands in the age of AI.

The Watcher Has Changed

At Hawthorne, observation had boundaries. Workers knew that an experiment was taking place, and eventually the researchers would leave. In your organisation, the observer may now be woven into email, calendars, CRM systems, customer calls, meeting transcripts, collaboration tools and the AI assistants people use throughout the day.

AI also sees a very different kind of workplace. Traditional systems could count calls, hours or transactions. AI can analyse language, compare conversations, identify patterns across thousands of interactions and make inferences about what may happen next. Activity that once required human judgment can now be translated into a score that looks remarkably objective.

There is another important difference. Generative AI can be both the colleague helping you do the work and the system reporting how you did it. The same assistant can draft the email, summarise the meeting, recommend the next action and create a record of your behaviour. The coach and the scorekeeper can sit inside the same tool.

This creates a powerful feedback loop. AI observes how employees behave. Employees discover what AI notices and adjust their behaviour. The system then learns from that altered behaviour, while leaders may treat the resulting pattern as an objective account of work. Each turn of the loop can take the organisation a little further away from what actually creates value.

The psychological effect may be even more significant. A manager does not need to be watching you at a particular moment. You only need to believe that almost any digital trace could be interpreted later. The observer begins to accompany people

Four Possible Effects of AI Observation

With the growing scope and scale of AI tracking, multiple possibilities are in play:

1. Improved Outcomes.

AI measurement might genuinely enhance performance, with many organisations reporting higher productivity during AI pilots. The danger is assuming these gains come solely from the technology. High-profile AI initiatives often coincide with greater leadership attention, clearer goals, additional resources, sharper feedback and a strong sense of purpose.

This raises an important question: Are AI pilots demonstrating the value of AI – or the combined value of several factors? It is rarely obvious which part of the intervention deserves the credit. Once the pilot ends, some gains may fade even if AI monitoring continues as before.

2. Performing for Metrics.

Once employees learn what’s being measured, they may begin managing the evidence. This is where AI comes up against Goodhart’s Law: When a measure becomes a target, it ceases to be a good measure. Response times become quicker without resolving the issue. More interactions are logged while relationships fail to deepen. Approved AI tools are frequently used even if they add little value.

AI can now quantify many activities, but some of the most vital contributions at work – sound judgement, mentorship, building trust, challenging poor decisions – remain difficult to measure. If organisations mistake what can be captured for what creates value, employees will invest more effort in improving the metrics rather than the underlying outcomes.

3. Reduced Candour.

Observation doesn’t always improve behaviour. It can also narrow it. When meetings are recorded, messages tracked and every interaction leaves a data trail, employees naturally become more cautious. They may become less willing to challenge assumptions, sit with uncertainty or develop bold ideas.

The paradox is that many organisations adopt AI to improve innovation, yet a culture of relentless evaluation can lead to conformity. Visible productivity may improve while curiosity, honesty and healthy debate decline.

4. Better Learning.

When used as a mirror by employees, AI observation can become a powerful learning assistant. It can identify coaching opportunities, highlight recurring patterns and support reflections on how meetings, presentations or negotiations unfolded. This kind of feedback can help people understand their own work more deeply and support continuous development.

Core Issues: Purpose, Trust & Transparency

Not all AI observation serves the same purpose. It may be geared towards:

• Learning. Providing employees with performance insights to shape and accelerate their development.
• Coordination. Supporting teams by identifying bottlenecks, improving workflows and allocating resources.
• Control. Helping the organisation uphold standards, monitor compliance and hold people accountable.

All three types of observation are legitimate in certain contexts. The problem begins when one form is disguised as another. When a “development tool” is primarily used for evaluation. When a “teamwork dashboard” turns into surveillance. That’s when trust begins to decline.

The introduction of AI measurement is never just a technology decision. It’s a management message. Employees will naturally ask:

• What is the intention behind this initiative?
• Who will be able to access this data?
• How will this information be used?
• What decisions will it drive?
• Can I challenge the conclusions?

These questions aren’t signs of resistance to technology. They are fundamental questions about fairness, trust and privacy. Research shows that the way AI monitoring is framed, governed and explained plays a critical role in its impact. A study published by the National Bureau of Economic Research found that digital monitoring did not improve performance by itself. Rather, how transparently it was communicated by management shaped the employee response.

Research also shows that when supervisors use monitoring information for developmental feedback, employees are more likely to maintain trust and perform well. When the same information is used mainly for control and evaluation, relationships can deteriorate.

The Leadership Challenge

Before introducing any AI-enabled measurement system, leaders should consider the following five actions:

1. Define the Decision.

Start by asking: What specific decision will this data help us make?

“Improving productivity” isn’t a decision. Reducing customer delays, tracing the causes of quality defects, or detecting recurring workflow bottlenecks are much more specific. The clearer the decision, the easier it becomes to determine what should be measured – and what should be left out.

 2. Anticipate the Behaviour.

Ask: How might a capable employee improve this metric without improving the underlying outcome?

Every metric creates an incentive, which in turn influences behaviour. Faster response times or higher AI adoption may look impressive while adding little value. The best time to think through these behaviours is before implementation.

 3. Look Beyond the Dashboard.

Ask: What valuable work might become less visible because it isn’t being measured?

As AI metrics refocus priorities, certain contributions might recede into the background – despite being vital for high-performing organisations. Leaders should be careful not to unintentionally signal that only visible work matters.

 4. Create Value for Employees.

Ask: How will the people generating this data benefit from it?

Employees are more likely to trust measurement when they benefit from it as well. A system that empowers people to improve their own performance feels fundamentally different from one simply designed for managerial oversight.

 5. Keep Judgement Open.

Ask: Can employees question the interpretations produced by AI?

AI assessments can appear authoritative even when they are incomplete or mistaken. Employees should never be evaluated solely on conclusions they cannot inspect, challenge or contextualise.

The original Hawthorne studies raised an intriguing question: How does workplace behaviour change when employees know they are being watched?

The challenge for leaders in the AI era is much more complex. As AI observation becomes continuous and embedded in work, how can we ensure that people become better at what they do – not simply better at being measured? How can we optimise not only what is visible but what creates value?

The impact of AI-enabled observation will be shaped by the intention behind it, by what we choose to prioritise. The way these systems are designed and implemented will reflect the organisation’s true beliefs around trust, accountability, learning, honesty, innovation and much more.

What message are you and your organisation sending?