Setting KPIs That Drive Performance

Rina, the HR lead at a 420-person payments company in Nairobi, inherited a KPI sheet with 18 measures for each manager. Nobody could remember them. Sales managers watched monthly revenue, support managers watched ticket volume, and engineering managers watched story points. The CEO wanted profitable growth, but the company was rewarding speed, busyness, and local optimization.

The fix was not a prettier spreadsheet. Rina cut each role to five measures, split leading indicators from lagging results, and made every department show how its KPIs supported the company OKRs. Within two quarters, managers were having sharper conversations because the scorecard finally described the work that mattered.

Start with the business question

Lead with the answer: a good KPI answers a management question. "Revenue" answers "Are we selling enough?" "Qualified pipeline created" answers "Will we sell enough next quarter?" "First-contact resolution" answers "Are customers getting useful help quickly?"

If a KPI does not change a decision, remove it. HR teams often track training hours, performance-review completion, and engagement scores because those numbers are easy to gather. Those measures can help, but only if they change an action: which managers need coaching, which roles need better onboarding, which teams need workload intervention.

Use this test in every KPI workshop: "If this number moves by 20%, what will we do differently next week?" If the room is silent, the metric is decoration.

Use the leading, lagging, input, output framework

SMART goals are useful for grammar, but they do not tell you whether a KPI will drive behavior. A stronger framework is two-by-two:

You need both. If you only track lagging outcomes, managers find out too late. If you only track inputs, people can look busy while the business misses the point.

For a customer support manager, "tickets closed" is an input measure. It can be useful for capacity planning, but it can also reward rushed answers. Pair it with first-contact resolution, customer satisfaction, reopened tickets, and response-time compliance.

For an engineering manager, "story points completed" is a weak input because teams estimate differently and can inflate points. Better leading indicators include cycle time, escaped defects, deployment frequency, and aging pull requests. Better outcomes include uptime, customer-impacting incidents, and product adoption for shipped work.

Keep the scorecard small

The useful range is three to seven KPIs per role. Three is enough for a narrow role. Seven is the ceiling for a complex leadership role. Anything beyond that becomes a reporting inventory, not a performance tool.

Use this sequence:

For example, a 12-person sales team may use: new qualified pipeline, closed revenue, win rate, average sales cycle, forecast accuracy, and customer handoff quality. That is six measures. Adding "number of emails sent" will probably make the team spam prospects. Adding "LinkedIn posts" will distract from selling unless social selling is a deliberate strategy.

Build KPIs by function

KPIs should reflect the work, not a generic HR template. A company can use a shared design logic while giving each function measures that fit.

Sales

Use a balanced set: qualified pipeline created, win rate, closed revenue, forecast accuracy, average deal cycle, and expansion or renewal quality where relevant. Do not let sales live on revenue alone. A seller can close bad-fit customers who churn in three months and still look like a hero.

Engineering

Use delivery, quality, and reliability together. Good examples are cycle time, deployment frequency, escaped defects, incident count, service-level objective attainment, and meaningful product outcomes. Be careful with individual engineer KPIs. Software is team work; individual output measures can damage collaboration quickly.

Customer support

Use first response time, first-contact resolution, customer satisfaction, reopened tickets, backlog age, and escalation quality. Ticket volume alone is a trap because complex cases need time and judgment.

Marketing

Use qualified pipeline influenced, conversion rates by channel, cost per qualified lead, sales acceptance rate, content-assisted revenue, and brand or audience measures where the strategy requires them. Avoid vanity metrics unless they predict a downstream result.

HR

Use hiring quality, time-to-fill for critical roles, regrettable turnover, manager capability measures, engagement action closure, payroll accuracy, and policy resolution time. For HRBPs, connect metrics to the client group: attrition in critical roles, succession coverage, employee-relations cycle time, and performance distribution quality.

Cascade from company OKRs without copying them

Company OKRs describe strategic bets. KPIs describe operational health. They should connect, but they should not be clones.

If the company objective is "Improve enterprise customer retention," support may own renewal-risk response time, product may own adoption of key features, customer success may own health-score recovery, finance may own billing-dispute resolution, and HR may own retention of customer-facing managers.

That is cascading. Copying the same "enterprise retention" metric to every department is lazy. It makes everyone accountable in theory and nobody accountable in practice.

Use quarterly KPI reviews even if compensation is annual. Business priorities move faster than annual forms. Review the measure, not just the number: is it still predictive, still fair, and still under the team's influence?

Watch for KPI traps

Goodhart's Law is the classic warning: when a measure becomes a target, it can stop being a good measure. In HR language, when people know a number controls praise, pay, or job security, they will optimize for the number.

Common traps:

• Gaming: support agents close tickets early to hit speed targets.
• Sandbagging: managers negotiate easy targets so they can exceed them.
• Local optimization: one team improves its score by creating work for another team.
• False precision: ratings use two decimal places even though evidence is weak.
• Perverse incentives: recruiters are rewarded for hires made, not hires who succeed.

The practical fix is a countermeasure. If speed is a KPI, add quality. If volume is a KPI, add conversion. If cost is a KPI, add service level. If manager ratings are a KPI, calibrate them.

Make KPI conversations human

KPIs are decision tools, not weapons. A manager should be able to say, "This number is off. What is happening, what have you tried, and what support do you need?" before saying, "You missed target."

For a first quarter under a new KPI, treat the measure as a learning baseline unless the role already had clear historical data. A 90-person company creating its first customer-success KPI should not punish managers for a target that was guessed in a conference room.

Set targets with ranges, not wishful thinking

Targets should come from evidence. Use last year's performance, current capacity, benchmark data, business commitments, and known constraints. If a support team closed 4,000 tickets last quarter with 12 agents, do not set a 6,000-ticket target after hiring one trainee unless automation, product fixes, or process changes make that realistic.

Use three target levels:

• Floor: the minimum acceptable result.
• Target: the result the plan is built around.
• Stretch: the result that would be excellent but should not be assumed.

For example, a talent acquisition team might use 35 hires as floor, 45 as target, and 55 as stretch for a quarter. The quality countermeasure could be 90-day new-hire success above 85%. Without that countermeasure, recruiters may optimize for speed and acceptance instead of fit.

Targets should be renegotiated when assumptions change materially. If a country launch is delayed, hiring targets tied to that launch should change. If a product outage creates 40% more support volume, response-time targets may need temporary adjustment. Flexibility is not weakness when the facts change.

Review KPIs in a real operating meeting

KPIs die when they live only in HR software. Put them into the meeting rhythm.

For a department head, the weekly KPI review should answer:

• What changed since last week?
• Which leading indicator is off track?
• What decision do we need today?
• Who owns the next action?
• What do we need to stop doing?

For an employee one-to-one, the KPI review should be narrower:

"Your first-contact resolution is strong, but reopened tickets increased from 8% to 14%. Let us review three examples and see whether the issue is knowledge, handoff quality, or ticket complexity."

That conversation is better than "your KPI is red." Red is a signal. Management is the response.

Retire KPIs that no longer help

KPI systems get cluttered because nobody wants to remove last year's favorite measure. Retire KPIs deliberately. A metric should be removed or demoted when it no longer changes decisions, when the work has stabilized, when people are gaming it, or when a better measure becomes available.

For example, a customer success team may track onboarding checklist completion during a messy rollout. Once completion is consistently above 98%, the KPI can become an audit item while the team focuses on time to first value and customer adoption.

Add a quarterly question to leadership reviews:

"Which KPI should we stop treating as a performance measure?"

That question protects attention. It also tells employees the organization is serious about measurement quality, not just measurement volume.

Coach managers to discuss misses

When a KPI is missed, managers should separate cause from accountability. A miss caused by poor effort, weak judgment, or ignored warnings is different from a miss caused by changed assumptions, under-resourcing, or a dependency outside the employee's control.

Use a four-question review:

This keeps KPI conversations fair. It also prevents the opposite failure: excusing every miss as context until the measure has no meaning.

Key takeaways

• Use KPIs to focus decisions, not to describe every task.
• Replace the old SMART-only approach with leading, lagging, input, and output thinking.
• Keep each role to three to seven KPIs.
• Build function-specific scorecards that include quality and risk controls.
• Cascade from company OKRs by translating strategy into department-owned work.
• Audit KPIs for gaming, sandbagging, and local optimization every quarter.