The Incentive Nobody Audits

Every company has a values statement. Most of them include words like innovation, collaboration, or agility. Many include all three. The values are on the website. They are on the wall in the meeting room. They were the subject of a workshop in 2023.

Every company also has an incentive system. The incentive system includes quarterly targets, performance reviews, bonus structures, promotion criteria, and the unwritten rules about what actually gets rewarded and what gets punished. The incentive system is not on the wall. It is in the spreadsheets, the calibration meetings, and the corridor conversations that nobody minutes.

The gap between the values statement and the incentive system is the single most reliable predictor of AI adoption failure.

This is the thing nobody audits. Not because it’s invisible — because auditing it requires naming a contradiction that everyone knows and nobody wants to formalise.

The Gap

A company says it values innovation. Its bonus structure rewards output volume. A team member who spends two hours experimenting with the new AI tool instead of processing the next batch of invoices has innovated — and has also produced two fewer hours of measurable output. The incentive system notices the output gap. The values statement notices nothing, because values statements do not have measurement mechanisms.

The rational response — and I use “rational” in the economic sense, meaning consistent with the actual incentive structure — is to not innovate. Process the invoices. Hit the target. Keep the bonus. Use the AI tool during lunch, if at all.

This is not resistance to change. This is accurate interpretation of the incentive environment. The team member is not failing to adopt. They are succeeding at optimisation — optimising for the signals that actually carry consequences.

The Three Incentive Misalignments

In the companies I’ve worked with at Bluewaves, three specific incentive misalignments kill AI adoption before the technology even has a chance to prove itself.

Misalignment 1: Individual Metrics vs Collaborative Tools

AI tools frequently create value by enabling collaboration — sharing knowledge across teams, surfacing patterns that span departments, enabling one person’s insight to amplify another person’s work. The value is collective. The value is emergent. The value resists attribution to a single individual.

Individual performance metrics cannot capture this. If my quarterly review evaluates me on the number of customer tickets I close, and the AI tool helps me close tickets faster — that helps me. But if the AI tool also helps me create a knowledge base entry that helps five colleagues close similar tickets — the value accrued to the team, not to me. My metrics don’t improve. My five colleagues’ metrics improve. I invested the effort. They received the benefit.

In an incentive system built on individual attribution, collaborative value is an externality — a benefit that accrues to the system but is not captured by the individual’s performance measure. Externalities are, by definition, under-produced. People do not invest effort in outcomes they are not rewarded for.

The fix is not eliminating individual metrics. It is adding collaborative metrics that capture the collective value AI tools produce. How many knowledge base entries did you contribute? How often was your input used by others? How many cross-team interactions did the tool facilitate? These are measurable. They are rarely measured.

Misalignment 2: Error Avoidance vs Experimentation

AI tools require experimentation. The first ten queries are learning queries — calibrating what the tool can and cannot do, discovering its strengths and limitations, developing an intuition for when to use it and when not to. Experimentation produces some errors. This is not a malfunction. It is the learning process.

If the incentive system penalises errors — and most do, explicitly or implicitly — experimentation carries risk. The risk is not catastrophic. Nobody gets fired for a bad chatbot query. But the risk is reputational: the colleague who sees a mediocre AI output on your screen, the manager who notices the quality dip in your first week of tool adoption, the quarterly review where “explored the new AI tool” is not a recognized accomplishment but “maintained quality standards” is.

Edmondson’s research on psychological safety is relevant here, but the incentive layer is distinct from the safety layer. You can have a psychologically safe team — one where speaking up is genuinely welcomed — and still have an incentive system that penalises the very experiments that adoption requires. The team feels safe to try. The incentive system punishes trying. The result is a team that knows it’s safe to experiment but has rational reasons not to.

The fix: create an explicit experimentation allowance. Not a vague “we encourage exploration.” A specific, measurable protection: 10% of each team member’s time during the adoption period is designated for tool experimentation, and this time is excluded from output-based performance metrics. The allowance must be documented in the performance review criteria, not just communicated verbally. Verbal encouragement without structural protection is just noise.

Misalignment 3: Speed Metrics vs Learning Curves

AI tools make work faster — eventually. During the adoption period, they make work slower. The learning curve is real. Looking up how to phrase a query, interpreting an unfamiliar output, verifying the tool’s response against your own knowledge — all of this takes time. Time that, in a speed-optimised incentive system, registers as a performance dip.

Most incentive systems measure speed directly or indirectly: calls handled per hour, tickets closed per day, reports produced per week. During the adoption period, these numbers drop. The drop is temporary. The drop is the cost of investment. But the incentive system does not distinguish between “performance drop due to incompetence” and “performance drop due to learning investment.” Both look the same in the spreadsheet.

A team member who observes this dynamic makes a calculation: the cost of the performance dip (visible, immediate, measured) versus the benefit of tool fluency (invisible, deferred, unmeasured). The calculation almost always favours abandoning the tool and returning to the existing process.

The fix: suspend speed-based metrics during a defined adoption period, or establish a separate performance baseline for the adoption period that accounts for the expected learning dip. “We expect your throughput to decrease by 15% during the first two weeks. This is budgeted.” The specificity matters. A vague reassurance — “don’t worry about your numbers” — is not credible when the performance review is in eight weeks.

Why Nobody Audits This

The gap between stated values and actual incentives is known. In every organisation I’ve worked with, people at every level can describe the gap with precision. The procurement officer knows that “innovation” is a value and “invoices processed per day” is a metric. The team lead knows that “collaboration” is a value and “individual ticket closure rate” is a bonus criterion. The HR director knows that the performance review template rewards output, not learning.

Nobody audits the gap because auditing it requires naming it. And naming it creates accountability. If the gap is documented — if someone writes down “our stated value is innovation and our bonus structure rewards output volume” — then someone must decide: change the values or change the incentives.

Both options are uncomfortable. Changing the values feels like abandoning principles. Changing the incentives feels like disrupting a system that works (or at least functions). So the gap persists, unnamed and unresolved, and every initiative that depends on the values (AI adoption, collaboration platforms, learning programmes) underperforms because the incentive system is working against it.

This is the core move I return to in my work: name the gap between what is stated and what is rewarded. The gap is not a failure of communication. It is a structural feature of organisations that have not aligned their measurement systems with their aspiration systems.

The Karasek Connection

Robert Karasek’s demand-control model adds a layer that connects incentive misalignment to stress. Karasek showed that the most harmful work configuration is high demands plus low control. When the incentive system creates high demands (hit your numbers) and the AI adoption process reduces control (use this unfamiliar tool in an unfamiliar way), the combination produces job strain — the configuration most associated with chronic stress, disengagement, and turnover.

The incentive misalignment amplifies the strain. The team member faces competing demands that cannot both be met: maintain output (incentive demand) and learn the new tool (adoption demand). The control is low on both axes: the output targets are non-negotiable, and the tool adoption is mandatory. The result is not resistance to change. It is a stress response to an impossible configuration.

The intervention is not motivational. You cannot motivate your way through a structural contradiction. The intervention is structural: resolve the conflicting demands by adjusting the incentive system to accommodate the adoption investment. This is a design decision, not a leadership decision. It requires changing the spreadsheet, not the speech.

The Audit

Here is what an incentive audit for AI adoption looks like. It takes one to two days. It costs nothing but honesty.

Step 1: List the actual performance metrics. Not the aspirational ones. The ones that appear in performance reviews, that determine bonuses, that influence promotion decisions. Be specific: “tickets closed per day,” “revenue generated per quarter,” “projects completed on time.” Include informal metrics — the things that are measured by attention rather than spreadsheets. “Being seen as productive” is an informal metric. “Being seen as a team player” is an informal metric. Both carry real consequences.

Step 2: List the behaviours AI adoption requires. Experimentation with the tool. Tolerance of the learning curve. Knowledge sharing across the team. Error reporting when the tool produces incorrect output. Investment of time in learning that does not produce immediate output.

Step 3: Map the conflicts. For each adoption behaviour, identify whether the actual performance metrics reward it, ignore it, or punish it. Use three categories: aligned (the metric rewards the behaviour), neutral (the metric is unaffected by the behaviour), or misaligned (the metric penalises the behaviour).

The map will show where the incentive system supports adoption and where it undermines it. In my experience, most organisations find that 30–50% of their actual metrics are misaligned with the behaviours AI adoption requires.

Step 4: Decide. For each misalignment, one of three responses: change the metric (adjust the incentive to align with adoption), protect the behaviour (create an explicit carve-out that shields the adoption behaviour from the misaligned metric), or accept the misalignment (acknowledge that this specific adoption behaviour will be under-produced and adjust adoption expectations accordingly).

The fourth response — do nothing and hope — is the default in most organisations. It is also the reason most AI deployments stall at the adoption stage.

The Seasonal Pattern

I want to name something I’ve observed across the companies Bluewaves works with, because it connects to timing and because it explains a pattern that frustrates many IT leaders.

Q4 AI deployments fail at a higher rate than Q1 or Q2 deployments. The technology is the same. The training is the same. The use case is the same. The difference is the incentive environment.

In Q4, annual targets are approaching. The gap between actual and target is either closing (pressure to maintain) or widening (pressure to catch up). Either way, the incentive system is at peak intensity. Every minute spent learning a new tool is a minute not spent closing the gap. Every experiment that reduces output is a luxury the Q4 calendar cannot afford.

In Q1, the targets have reset. The pressure is at its annual minimum. The budget for the new year is confirmed. There is cognitive slack — not much, but some. The same tool deployed in January, to the same team, with the same training, performs better in adoption metrics than the same tool deployed in October. The difference is the incentive calendar.

This is not a revelation. It is an observation that most deployment timelines ignore because the technology readiness date is treated as the deployment date, regardless of the incentive environment. The tool is ready, so we deploy. The team is not ready — structurally, in its incentive architecture, physiologically — but the tool is ready.

Readiness is not a technology attribute. It is an environmental attribute. The environment includes the incentive system, and the incentive system has seasons.

The Manager’s Role

I want to name the manager’s position specifically, because it is the hardest position in this dynamic.

The manager is caught between the incentive system (which they enforce) and the adoption requirement (which they champion). They must tell the team “hit your numbers” and also tell the team “take time to learn the new tool.” Both instructions come from their mouth. The team hears both. The team follows the one that has consequences.

The manager who says “I want you to explore the AI tool” and then asks on Friday “why were your numbers down this week?” has sent two messages. The second one cancelled the first. The cancellation was not intentional. It was the incentive system speaking through the manager — and the incentive system speaks louder than encouragement.

The fix is structural, not motivational. The manager needs the incentive system to be adjusted before they can credibly champion adoption. “I want you to explore the AI tool, and your target has been reduced by 15% for the next two weeks to create space for that exploration” is a credible message. “I want you to explore the AI tool, and also hit the same targets as last month” is not. The team knows the difference. The team always knows the difference.

The Integration

I hold two things without resolving them.

First: incentive systems exist for reasons. They drive behaviour. They create accountability. They make performance visible and measurable. Dismantling them in the name of AI adoption would be irresponsible and unnecessary.

Second: incentive systems are not neutral. They express what the organisation actually values, regardless of what the values statement says. When the incentive system penalises the behaviours that adoption requires, adoption fails. This is not a personnel failure or a technology failure. It is a design failure in the incentive architecture.

Both things are true. The work is in the space between them — adjusting the incentive system enough to accommodate adoption without dismantling the accountability that the system provides.

This is design work, not policy work. It requires specificity: which metrics, during which period, with which protections. It requires measurement: did the adjustment produce the adoption behaviour? Did it produce unintended consequences? It requires iteration: the first adjustment will be imperfect. Adjust again.

The incentive nobody audits is the incentive that determines whether your AI deployment succeeds or fails. Not the technology. Not the training. Not the leadership communication. The spreadsheet that determines what gets rewarded and what gets punished.

Audit the incentive. The adoption follows the incentive. It always has.

The technology is ready. The training is designed. The business case is sound. The leadership is committed. None of these matter if the incentive system — the actual mechanism that determines what people do on a Tuesday afternoon — is working against adoption.

The incentive is the infrastructure. Audit the infrastructure. Fix the infrastructure. The adoption follows.

It always has.