11.4 Objectives, key results, and key performance indicators¶

Overview and motivation¶

Two goal-setting instruments dominate how large organizations turn strategy into measurable work: OKRs and KPIs. They are complementary, so this chapter treats them together and shows where each belongs.

Objectives and Key Results (OKRs) is a framework that connects an organization's strategy to the concrete, measurable outcomes each team commits to achieve in a defined period. It was developed at Intel by Andrew Grove and popularized after John Doerr introduced it at Google. An OKR describes the change you are driving right now.

A key performance indicator (KPI) is a metric your organization deliberately chooses to track because it reflects the ongoing health of something that matters. The word key matters here: a KPI is not any metric you can collect, but the small set you have decided are worth steering by. A KPI describes the steady state you are protecting.

Chapter 11.1 (the discovery pipeline) introduces both briefly. This chapter treats them in depth: how to write them, run their cadence, grade them, choose them, and protect them from distortion.

For large teams, the discipline these instruments impose is worth as much as the instruments themselves. In enterprises, dozens of teams pursue local goals that quietly diverge from strategy, and one well-chosen KPI can align hundreds of people. In government, multi-year programs commit public funds against legislated mandates, and "we delivered the modules in the statement of work" is no defense if wait times or fraud never improved. OKRs force intent to be stated as outcomes the public can audit, while a badly chosen KPI (a call center measured on calls closed per hour) can quietly corrupt an organization. This chapter is about getting those choices, and the guardrails around them, right.

Key principles¶

• Objectives inspire; key results measure. Direction plus proof.
• Key results are outcomes, not tasks. Measure the change you cause, not the work you do.
• OKRs for the change; KPIs for the health.
• Fewer is better. A few objectives, a few key results each, a small hierarchy of KPIs. Focus is the mechanism.
• Align, don't dictate. Leadership sets direction; teams propose how they contribute.
• Cadence over ceremony. Set, review mid-cycle, grade honestly, repeat.
• Measurable, actionable, and owned. Every KPI has a definition, a source of truth, an owner, and a lever that moves it.
• Lead where you can, confirm where you must. Prefer leading indicators; use lagging ones to verify.
• Assume every metric will be gamed. Design against Goodhart's Law with ratios, cohorts, and paired guardrails.
• Keep OKRs out of pay and ratings. They then get sandbagged and gamed.

Recommendations¶

Write objectives that are qualitative and directional¶

An Objective is a qualitative, inspirational statement of a desired end state in plain language a team can rally around, short and free of numbers. "Delight new customers in their first week" is an objective; "Increase activation by 20%" is a key result in disguise. Aim for three to five per level; more is a to-do list, not a focus.

Make every key result a measurable outcome, not an output¶

This is the most violated rule. Key Results are the few measurable outcomes that together prove the objective was achieved. An output is what the team produces (a shipped feature); an outcome is the change it causes (higher retention): an output can be 100% done while the outcome stays flat. Write two to four per objective, each with a metric, baseline, target, and source of truth (chapter 7.4). A key result you can complete by working hard rather than by moving a number is a task, not an OKR.

Align by cascading down and proposing up¶

Alignment runs both ways. In cascading (top-down) alignment, leadership publishes a few company objectives and each layer derives objectives that ladder up. In bottom-up proposal, teams closest to the work draft their own and negotiate upward. Do both; roughly half of OKRs should originate bottom-up. The goal is alignment, not dictation: handing teams their numbers destroys the ownership that makes OKRs motivating, and OKRs become the language a portfolio uses to express intended outcomes (chapter 10.1).

Run a disciplined cadence: set, review, grade¶

Run OKRs on a rhythm, commonly quarterly for teams inside an annual company frame. Setting drafts, negotiates, and publishes, so every team sees every other's. Mid-cycle review (check-ins plus a formal mid-point) tracks confidence and surfaces blockers while there is time to act; a key result trending red triggers a governance decision (chapter 1.5), not an end-of-quarter surprise. Grading scores each key result honestly on a 0.0 to 1.0 scale (the fraction of target achieved); a retrospective then decides whether the objective continues, changes, or retires.

Distinguish aspirational from committed OKRs¶

Separate two kinds, because they grade differently. A committed OKR is a promise: the team expects 1.0, and missing it signals a real problem; use these for must-hit obligations like a compliance deadline. An aspirational (or "stretch") OKR is a deliberately ambitious target where reaching 0.6 to 0.7 is success: graded at 1.0 it was set too low, at 0.2 it was fantasy with no plan. Label each type so nobody mistakes an ambitious 0.6 for a failed one.

Keep OKRs separate from performance reviews and compensation¶

OKRs are for direction and learning, not a bonus formula. The moment a key result determines someone's rating or pay, people sandbag (set targets they know they can beat, killing ambition) and game the metric (optimize the number while the goal decays). This is Goodhart's Law: when a measure becomes a target, it stops being a good measure. Grade OKRs for learning; evaluate people through the separate mechanisms of chapter 1.3. The two conversations can inform each other, but must not be the same one.

Distinguish KPIs from OKRs, and let a metric play both roles¶

A KPI measures ongoing health you sustain regardless of what you are changing: availability, conversion rate, cost per transaction. A key result expresses a target movement in a metric over a bounded time. A metric can play both roles: a KPI you move this cycle becomes a key result; the rest stay guardrails beside your OKRs. The rule: keep tracking it after the objective is met and it is a KPI; retire it with the objective and it was a key result. In short, OKRs for the change, KPIs for the health.

Define what makes a KPI "good"¶

A good KPI passes four tests. It is aligned: it traces to a stated goal. It is measurable: one precise definition, a named source of truth (the authoritative system of record), a unit, and a collection method, so two people computing it separately agree. It is actionable: the owning team has levers that move it. It is owned: one accountable person or team owns its trend, definition, and data quality. A metric that fails any test is a candidate for deletion.

Classify metrics as leading, lagging, and input/output/outcome¶

Two lenses keep a set balanced. Timing: a leading indicator is predictive and movable now (trial sign-ups); a lagging indicator is confirmatory and slow (annual revenue, churn); leading indicators let you steer before lagging ones deliver the verdict. Chain of value: an input metric measures effort spent (hours), an output metric measures what the system produced (features shipped), and an outcome metric measures the change you actually wanted (revenue retained). Teams gravitate to inputs and outputs because they are easy to count, but the value lives in outcomes.

Design metrics that resist gaming (Goodhart's Law)¶

Assume Goodhart's Law (a measure that becomes a target stops being a good measure), and design against it.

• Prefer ratios and rates over raw counts. "Tickets closed" rewards volume; "percentage resolved on first contact" rewards resolution. Raw counts are gameable by doing more of something worthless.
• Use cohorts. A cohort is a group defined by a shared starting point (all users who signed up in March), so a declining trend cannot hide inside a flattering aggregate.
• Pair every incentivized KPI with a guardrail. A guardrail metric is a paired counter-metric that must not degrade while you push the primary one (call handle time with satisfaction), making cheating visibly expensive.

Reject vanity metrics; require actionability¶

A vanity metric reliably rises, looks impressive, and changes no decision: cumulative registered users, page views, lines of code. The tells: it only goes up, it is an absolute count rather than a rate, and it has no answer to "what would we do differently if this doubled?" An actionable metric ties to a specific behavior you can influence and a decision it would change. Before adopting any KPI, ask what action a good or bad reading would trigger; if none, discard it.

Build a KPI hierarchy and choose a north-star metric¶

Do not track KPIs as a flat list. Arrange them as a KPI tree (or metric tree): a top metric broken into the metrics that mathematically or causally drive it, level by level, down to the operational measures teams own, so you know which lower metric to investigate when a top one moves. At the top, name a single north-star metric: the measure that best captures core value delivered to customers (for a marketplace, completed transactions). It aligns effort and stops departments optimizing conflicting local numbers, but only if it measures value rather than vanity and guardrails balance it.

Set baselines, targets, and thresholds¶

A KPI without a reference point is just a number. Give it three: a baseline (the current or historical value, so change is meaningful), a target (the value you intend to reach, and by when), and thresholds that trigger action (a warning threshold prompts attention, a critical threshold prompts intervention). Ground targets in evidence (past trend, benchmark, or capacity) rather than round-number optimism, and record the reasoning so a miss can be interpreted.

Visualize honestly on dashboards¶

A KPI dashboard should let a reader grasp status and trend in seconds without being misled. Show trend over time, not a snapshot; start value axes at zero unless there is a stated reason not to, because truncated axes exaggerate change; show variation and uncertainty rather than false precision; annotate context (deploys, incidents, seasonality); and avoid chart tricks that flatter the number (dual axes that imply correlation, cherry-picked ranges, 3-D effects). These practices connect to analytics and business intelligence (chapter 7.3) and product analytics (chapter 7.4).

Instrument operational KPIs: SLIs/SLOs and flow/DORA¶

Operational health has established KPI vocabularies. From site reliability engineering (chapter 9.1): a service level indicator (SLI) is a measured signal of service health (latency, success rate); a service level objective (SLO) is the target range for an SLI (99.9% of requests succeed); and the error budget is the allowed shortfall (the 0.1% you may spend on risk before you stop and stabilize). From the delivery pipeline (chapter 11.2): flow metrics track work through the system (flow time, throughput, work-in-progress, flow efficiency), and the four DORA metrics (from the DevOps Research and Assessment program) pair speed and stability: deployment frequency, lead time for changes, change failure rate, and failed-deployment recovery time. Both set speed beside stability, so no team can optimize velocity by sacrificing reliability.

Meet public-sector performance-reporting duties¶

Government adds a distinct requirement: KPIs are often published performance measures reported to legislatures, oversight bodies, and the public, sometimes under statute. Treat these with extra rigor: a stable definition held constant across reporting periods (so trends are comparable), a documented methodology and data source, and honesty about limitations. Because published measures create strong incentives, they are especially prone to Goodhart distortion (a target to reduce a waiting list met by redefining who counts as waiting), so pair each with guardrails and audit the definition itself, not just the number.

Trade-offs: pros and cons¶

Two tensions govern this chapter. Ambition versus safety: stretch goals create breakthroughs, but only where a 0.6 is celebrated, not punished, which is why coupling OKRs to compensation is corrosive. Focus versus coverage, sharpened by motivation versus distortion: enough measures to see the whole picture, few enough to act, ambitious enough to improve, guarded enough that ambition does not corrupt the measure. Resolve both by labeling OKR type, grading for learning, keeping ratings elsewhere, and holding a small set of owned, outcome-weighted KPIs with guardrails.

Examples¶

Startup. A six-person productivity-app startup is celebrating a rising "total registered users" chart until a board member asks whether anyone actually keeps using the product. They swap that vanity count for one aspirational quarterly OKR: the objective "New users feel the magic in week one," with key results for 7-day activation (from 25% to 45%) and week-4 cohort retention, each cohorted by signup week so a strong launch cannot hide churn underneath. They pick 7-day activation as their north-star metric, pair it with a guardrail (support tickets per active user) so they cannot juice activation with a pushy onboarding flow, and grade the OKR at 0.6 as an honest success, keeping it well away from anyone's equity or pay.

Enterprise. A global logistics company runs OKRs quarterly under annual company objectives. One is "Become the carrier customers trust for time-critical shipments." A regional operations team proposes, bottom-up, "Make same-day delivery dependable in our metros," with aspirational key results: raise on-time same-day delivery from 82% to 95%, halve complaints per thousand shipments, and hold cost per parcel flat. The company adopts on-time delivery rate as its north-star metric and builds a KPI tree (pickup punctuality, hub dwell time, last-mile success), each owned by a regional lead with a baseline, target, and warning and critical thresholds, and each speed KPI paired with a guardrail (on-time rate with damage rate). Platform SLOs (99.95% tracking-API availability, chapter 9.1) and DORA metrics (chapter 11.2) sit on the engineering dashboard. At mid-quarter the on-time key result trends to 0.4, triggering a governance decision (chapter 1.5) to reallocate two teams; the tree pinpoints one hub's dwell time. At quarter end the objective grades 0.6 (a success for a stretch goal), the learning reshapes next quarter's portfolio (chapter 10.1), and no one's bonus moved because of the 0.6.

Government. A state unemployment-insurance agency modernizing its claims system sets an annual objective: "Help eligible residents receive benefits quickly and correctly." Committed key results (they carry statutory weight) include paying 90% of valid first claims within 21 days, up from 63%, and holding the improper-payment rate below the federal threshold. Aspirational key results push further: raise self-service filing from 55% to 80% and cut the median call-center wait from 40 to 10 minutes. For its published measures, the agency reports median days from claim to first payment (an outcome, cohorted by claim month so a good quarter cannot mask a worsening backlog) rather than the vanity count of claims processed, paired with guardrails (payment accuracy, appeal-overturn rate) so speed cannot be bought with errors. Definitions are frozen across years and documented publicly, and an independent audit checks the definition itself, confirming that "waiting" claimants are not quietly reclassified. Frontline SLIs (portal uptime, call-answer rate) feed the operational dashboard. Because success is defined as claimant outcomes rather than delivered modules, the agency shows its legislature measurable public value, with OKRs graded quarterly for learning, separate from civil-service ratings.

Business case: motivations, ROI, and TCO¶

The return on both instruments comes from focus and alignment, avoiding the largest hidden cost in large organizations: many teams working hard, on time, on things that do not advance strategy. When intended outcomes are explicit and visible, duplicated effort surfaces, contradictory goals get reconciled before they collide, and low-value work loses its cover. A large organization's most expensive resource is the aligned attention of its people. The dominant ROI is redirected capacity.

The cost of the wrong KPI is not the dashboard. It is quarters of effort optimizing a number while the real outcome erodes, plus the cost of unwinding the gamed behavior. A call center that spent a year minimizing handle time while maximizing repeat contacts shows a negative ROI produced entirely by a well-intentioned metric.

The total cost of ownership (TCO) is modest but real. The OKR cadence consumes management time, and writing good outcome-based key results takes a few cycles to learn. Each KPI carries ongoing cost: pipelines, dashboards, review meetings, and mental load, which argues for a small set where every KPI earns its keep. Two failure modes destroy the ROI: OKR theater (goals written at quarter start, filed, never reviewed or graded), which pays the full ceremony cost for none of the benefit; and coupling OKRs to compensation, whose sandbagged and gamed metrics erode the outcomes the framework meant to improve. Keep the cadence light, the counts low, and the grading honest, and both instruments repay their overhead many times over.

Anti-patterns and pitfalls¶

• Key results that are tasks: "launch the app" instead of "raise mobile retention to X%."
• Too many OKRs: a dozen per team is a backlog, not a focus.
• OKR theater: set once, never reviewed, never graded.
• Cascade-only alignment: numbers handed down with no bottom-up proposal, killing ownership.
• Sandbagging and stretch goals mistaken for commitments: targets set low to guarantee a 1.0, or a healthy 0.6 punished until teams stop reaching.
• OKRs tied to pay or ratings: the direct cause of sandbagging and gaming (Goodhart's Law).
• Metric fixation: behavior optimizes the number, not the outcome. Pair with a guardrail.
• Vanity metrics: cumulative counts that only rise and change no decision.
• Unowned or ill-defined KPIs: no accountable person, or two teams computing "active user" differently and arguing instead of managing.
• Dashboard sprawl and dishonest charts: gauges with no hierarchy or north star; truncated or dual axes and cherry-picked windows that manufacture a story.
• Grading dishonestly: inflating scores discards the framework's only real output: the truth.

Maturity model¶

• Level 1, Initial: Goals are ad hoc or feature lists; success means "we shipped it." Metrics are mostly vanity counts with no baselines, targets, owners, or shared definitions. No cadence; dashboards show snapshots without trend.
• Level 2, Managed: Some teams run OKRs, but key results are often disguised tasks, cadence is inconsistent, and OKRs leak into performance reviews. KPIs exist for some teams with baselines and targets, but they are mostly outputs, definitions vary, and guardrails are absent. Gaming appears, unrecognized.
• Level 3, Defined: A quarterly-within-annual cadence exists org-wide. Key results are outcome-based with baselines and targets; OKRs align top-down and bottom-up; committed and aspirational OKRs are labeled; grading is honest and separate from compensation. A coherent, owned KPI set has precise definitions and a source of truth; metrics are classified (leading/lagging, input/output/outcome); guardrails pair incentivized measures; charts follow honest-visualization standards.
• Level 4, Optimizing: OKRs are the portfolio's shared language (chapter 10.1); mid-cycle reviews drive real reallocation (chapter 1.5); outcome data from delivery and analytics (chapter 7.4) loops into the next cycle. A KPI tree ties frontline measures to a north-star metric; outcome metrics dominate; operational SLIs/SLOs and DORA/flow metrics integrate with business KPIs; definitions are audited and Goodhart effects monitored. The culture rewards stretch goals and treats low grades as learning.

Ideas for discussion¶

1. Of your current key results, how many describe an outcome, and how many are tasks you could complete by simply working hard?
2. What fraction of your OKRs originated bottom-up versus handed down, and what does that imply for ownership?
3. Are your OKRs ever consulted in performance or pay conversations, and what has that done to how ambitiously people set targets?
4. Which objectives are genuine stretch goals, and would your culture celebrate or punish a 0.6 on them?
5. Which of your KPIs would keep rising even if the product or service were getting worse?
6. For each incentivized KPI, what is the cheapest way to game it, and what guardrail would expose that cheating?
7. Which metrics you call OKRs are really KPIs, and which KPIs lack an owner or a single agreed definition?
8. For a public measure you report, could the target be met by redefining the measure rather than improving the outcome?

Key takeaways¶

• An Objective is a qualitative, inspirational end state; Key Results are the few measurable outcomes that prove it was reached, and they must be outcomes, not tasks.
• A KPI is a metric chosen because it reflects a goal; a good one is aligned, measurable, actionable, and owned.
• OKRs are the change you want; KPIs are the health you sustain. A KPI you move this cycle becomes a key result; the rest stay guardrails.
• Align both ways: leadership sets direction top-down, teams propose bottom-up. Alignment, not dictation.
• Run a cadence (quarterly within an annual frame) with mid-cycle review and honest grading on a 0.0 to 1.0 scale; label committed OKRs (aim for 1.0) versus aspirational/stretch OKRs (0.6 to 0.7 is success).
• Keep OKRs out of performance reviews and compensation to prevent sandbagging and gaming.
• Classify metrics as leading/lagging and input/output/outcome, weight toward outcomes, and assume Goodhart's Law: pair every incentivized KPI with a guardrail, prefer ratios and cohorts over raw counts.
• Reject vanity metrics; organize KPIs into a tree under a north-star metric; give each a baseline, target, and thresholds; visualize honestly (chapters 7.3, 7.4); adopt operational SLIs/SLOs (chapter 9.1) and DORA/flow metrics (chapter 11.2).
• In government, treat published measures with extra definitional rigor and audit the definition, not just the number.
• Both instruments feed the discovery pipeline (chapter 11.1).

References and further reading¶

• Measure What Matters, by John Doerr (the canonical modern account of OKRs, from Intel to Google, and their relationship to ongoing measures).
• High Output Management, by Andrew Grove (the origin of the "objectives and key results" approach).
• Radical Focus, by Christina Wodtke (OKRs in practice for smaller teams and cadence discipline).
• Objectives and Key Results, by Paul Niven and Ben Lamorte (implementation guidance for larger organizations).
• The Performance Management Revolution: related literature on separating goals from ratings.
• Google re:Work: "Set goals with OKRs" (Google's published internal guidance on OKR practice).
• Doran, G. T., "There's a S.M.A.R.T. way to write management's goals and objectives" (Management Review, 1981): the SMART criteria for writing measurable key results.
• Lean Analytics, by Alistair Croll and Benjamin Yoskovitz (vanity vs. actionable metrics, the One Metric That Matters).
• Key Performance Indicators: Developing, Implementing, and Using Winning KPIs, by David Parmenter.
• The Lean Startup, by Eric Ries (actionable vs. vanity metrics, cohort analysis).
• How to Measure Anything, by Douglas W. Hubbard (defining and quantifying the seemingly unmeasurable).
• The Visual Display of Quantitative Information, by Edward R. Tufte (honest, high-integrity data graphics).
• Site Reliability Engineering, by Betsy Beyer, Chris Jones, Jennifer Petoff, Niall Richard Murphy, eds. (SLIs, SLOs, error budgets).
• Accelerate, by Nicole Forsgren, Jez Humble, and Gene Kim (the DORA delivery and stability metrics).
• Goodhart, C. A. E., "Problems of Monetary Management: The UK Experience" (1975): origin of Goodhart's Law; see also Marilyn Strathern's widely quoted formulation.
• U.S. Government Accountability Office (GAO) guidance on performance measurement and the GPRA Modernization Act: public-sector performance reporting practice.