HBR says AI-era performance metrics reward the wrong people. 10 fixes for mid-market People teams.

Playbook angle: In a performance management process, metrics only help if they are tied to the scorecard managers actually use. Review each fix below against your current review criteria and keep only the measures you can evidence.

Here is a conversation I keep having with People leaders in Tallinn, Stockholm and London. Their teams adopted AI tools faster than anyone planned. Output went up. Then review season arrived, and the managers could not tell who had actually done good work and who had simply produced a lot of it.

Harvard Business Review names the problem clearly. In “Performance Management Needs New Metrics in the AI Era” (Randy Bean, Erik Strauss and Randeep Singh, HBR, 6 July 2026), the authors argue that companies are measuring people who work with AI using the same old yardsticks, and that those yardsticks now point the wrong way.

What HBR is actually saying

In plain language:

• The old metrics now mislead. Companies still score people on productivity, goal completion and efficiency. The authors call the result a paradox: employees who rely heavily on AI “may appear highly productive,” while those who verify assumptions, challenge outputs or correct errors “might appear less efficient precisely when they’re adding the most value.”
• Spend is high, proof is thin. They cite the Data and AI Leadership survey (Bean with Tom Davenport): 91% of organisations were increasing AI investment, yet just 18% said they were getting a high degree of measurable business value from it.
• Performance management was already struggling. They point to a 2024 Gallup survey in which just 20% of employees strongly agreed their reviews were fair or transparent.
• AI is brilliant, then suddenly isn’t. In a study of more than 750 knowledge workers they cite, AI made people more than 25% faster on tasks within its range. On a task just outside it, people using AI were 19% less likely to get the right answer. If you only reward speed, you train people to push work past that edge and hope.
• Measure three things, not one. They propose a three-layer framework: the person’s contribution, the AI system or agent, and the combined human-AI result. Each layer gets a small number of metrics, reviewed monthly or quarterly, and tied to visible decisions on coaching, staffing, promotion and tool governance.
• Start with one workflow. Their advice is to pick one process where AI is already changing the work, redesign the metrics before the rating form, keep development conversations separate from pay decisions, and name who is accountable for the AI in that workflow.

Why mid-market teams feel this first

Large enterprises have analytics teams to argue about measurement frameworks. A 150–800 person scale-up in Estonia, the Nordics or the UK usually has a People team of two to six, managers with eight or more reports each, and AI tools that arrived bottom-up, team by team. So the review form still asks “did you hit your goals?” and “how productive were you?”, and the honest answer is “yes, and a chatbot helped,” which tells you very little.

The good news: most of what HBR asks for at the person level is not exotic. It is clear goals, a written standard for good judgement, and evidence that builds up all year. That is ordinary performance management done properly, and it is where the fixes below start.

10 fixes for measuring performance in the AI era

1. Grade outcomes and judgement, not volume. Rewrite your scorecard so “more output” never scores on its own. Ask what changed for the customer, the team or the business, and how well the person judged when to trust the tool. In Revolut People: structured scorecards with criteria you set, from pre-built templates or your own custom cards.
2. Write “good judgement” into the bar for each level. HBR’s person-level measures centre on spotting when AI is out of its depth and documenting why you overrode it. Put that into the expectations for each seniority level so managers score against a standard, not a hunch. In Revolut People: standardised skills and values frameworks, so everyone is assessed on the same criteria across teams and countries.
3. Make goals about results, and keep them live. If a goal can be hit by generating more stuff, AI will hit it. Set goals on outcomes, and track progress during the year rather than at review time. In Revolut People: company → team → individual goals with real-time tracking, synced with tools like Jira, Linear, Notion and Salesforce.
4. Capture the catches when they happen. The moment someone spots a wrong AI answer before it reaches a client is exactly the evidence HBR says gets lost. Make it normal to log it as feedback that week, not to remember it in December. In Revolut People: continuous and peer-to-peer feedback, linked to your company values and visible at review time.
5. Use 1:1s to ask where AI helped and where it didn’t. A standing 1:1 question (“what did the tool get wrong this month, and how did you spot it?”) turns coaching into a record of judgement. In Revolut People: AI-powered 1:1 summaries that highlight key talking points, so the notes build up instead of restarting from zero.
6. Reward people who make the whole team better. HBR suggests measuring whether someone raises team throughput, not just their own: improving a workflow, documenting it, sharing it. Name that in your criteria and ask peers about it. In Revolut People: 360 and peer feedback alongside manager input in the review.
7. Review more often than once a year. HBR recommends reviewing each layer monthly or quarterly. AI tools change quickly; a twelve-month gap means you are scoring last year’s way of working. In Revolut People: review templates and cycles you can run on the cadence you choose, with access to previous cycles for the full picture.
8. Keep coaching separate from pay. This is HBR’s sharpest warning: if the same AI-generated signal is used to coach and to pay, employees will treat it as surveillance. Use the year-round data to coach; make pay a distinct, explained decision. In Revolut People: feedback and 1:1s run all year, while promotions, performance improvement plans and salary adjustments come out of the cycle as separate post-review plans with a named owner. (The separation is your policy choice; the software just keeps the steps distinct.)
9. Calibrate on evidence, not on who shipped most. Calibration is where output-heavy stars get waved through. Put goal progress, feedback and 1:1 history next to each rating, and challenge any rating that rests on volume alone. In Revolut People: calibration with scorecards pre-filled by AI from goal progress, 1:1 discussions and performance across the period, which managers then review and own.
10. Pilot one workflow and name an owner for the AI in it. Pick one process (sales proposals, support, finance reporting), decide who is accountable for the tool’s behaviour, and agree how you will tell if the pilot worked. No HR platform does the governance part for you. In Revolut People: analytics on goal attainment, review cycles and performance trends, so you can see whether the pilot team’s results actually moved.

Includes a price for your headcount + 3 fixes for your review cycle.

Where Revolut People fits, and where it doesn’t

Revolut People started as the in-house system Revolut uses to run performance across its own workforce. It is built for the person layer of HBR’s framework: making human contribution visible all year.

• Goals: cascaded, real-time goals synced with the tools teams already use.
• Reviews and calibration: structured scorecards that put evidence ahead of opinion, 360 feedback, calibration and previous cycles.
• 1:1s and continuous feedback: AI-powered 1:1 summaries and peer feedback linked to values.
• Outcomes: post-review plans for promotions, PIPs and salary adjustments, so every person gets a clear outcome.
• Analytics: goal attainment, review cycles, performance trends and engagement, so you can spot patterns rather than anecdotes.

To be straight about the limits: Revolut People measures people, not AI agents. HBR’s second layer (agent error rates, escalation and audit trails) belongs with whoever owns those tools. What a performance platform can do is make sure the human judgement HBR wants to reward is actually written down somewhere you can find it. Method context is in the playbook summary, Publish a talent bar by seniority and Gartner on calibration.

Bottom Line

AI has made output cheap, so output is no longer a good measure of performance. HBR’s point is that the people catching AI’s mistakes are often your most valuable, and your current metrics may be marking them down. You don’t need a giant new framework to fix that. Set outcome goals, write good judgement into the bar for each level, capture feedback and 1:1 notes all year, keep coaching apart from pay, and calibrate on evidence. Start with one team before your next cycle.

Free tools: estimate what your review cycle costs with the performance review cost calculator, or see if your HRIS is holding performance back with the HRIS switch checker.

FAQ

What are AI-era performance metrics?

Measures that judge how well people work with AI, not just how much they produce. HBR’s July 2026 framework looks at three layers: the person’s contribution (for example, spotting when AI is wrong and helping the team work better), the AI system or agent, and the combined human-AI result.

Should we track how much employees use AI?

Usage on its own is not performance. HBR warns that rewarding speed and output can reward over-reliance on AI. Where it suggests tracking adoption, it is at team level and paired with how deeply the tools are used. Focus individual reviews on outcomes and judgement.

Should AI write our performance reviews?

AI can save managers time by summarising goal progress, feedback and 1:1 notes, but a person should own every rating. HBR’s advice to keep development and pay decisions separate also means being transparent about what data feeds which decision.

Does Revolut People measure AI tools or agents?

No. Revolut People is a people performance platform: goals, reviews, calibration, 1:1s, feedback and analytics. It helps capture human contribution all year; measuring AI agents themselves sits with whoever owns those tools.

Is Revolut People available in Estonia, the Nordics and the UK?

Yes. Revolut People is available to businesses with a registered entity in the UK, the EEA or the US, which covers Estonia and the Nordic countries.