deciqAI OKRs (Objectives and Key Results)
Overview
OKRs separate ambition from measurement: an Objective is qualitative and aspirational; Key Results (3-5) are quantitative outcomes proving the objective was reached. KRs must be outcomes, not activities. Calibration rule: 70% achievement is success — routine 100% means goals were sandbagged. Developed by Andy Grove at Intel (1971); introduced to Google by John Doerr (1999).
Composes with north-star-metric, first-principles, metacognition, mece.
When to Use
- Team goals are vague, unmeasurable, or just "complete project X" lists
- Teams hitting all goals but the business isn't moving (sandbagging signal)
- Cross-team work failing due to private, conflicting goals; or a new company needs goal infrastructure
- Setting OKRs on AI features and the draft KR is "ship AI" / "increase AI usage" (vanity/Goodhart metric — replace with outcome KRs)
Not when: under ~10 people with sufficient informal alignment; inherently uncertain output (research labs); leadership will punish 70% achievement.
Coaching Novices (Adaptive Front Door)
- Engine mode: user has a concrete case → run The Process directly.
- Coach mode: user unfamiliar or no concrete case → guide step by step.
In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.
- OKRs separate the inspirational what we're trying to be (Objective) from the measurable how we'll know (3-5 Key Results) — aiming for 70% so goals stretch beyond what's safe.
- Check fit: under 10 people / uncertain output / leadership punishes 70% → not the right time.
- Elicit the actual goal the team is trying to set.
[WAIT — do not advance until user responds]
- Ask one question at a time: what's the outcome? What would prove it happened? Activity or outcome? Would 70% be a real win?
[WAIT — do not advance until user responds]
- Close: one well-formed Objective and 3 Key Results, plus the cadence to revisit them.
[WAIT — do not advance until user responds]
The Process
Step 1 — Write the Objective. Qualitative, aspirational, time-bounded, one sentence. Test: "If achieved, would this objectively matter to the business?"
Step 2 — Write 3-5 Key Results. Quantitative outcomes (not activities), time-bounded, 70%-success-calibrated. "Ship v2.0" = activity (wrong). "Reach 100k WAU on v2.0 within 30 days" = outcome (right). If all KRs happen and the Objective is not achieved, the KRs are wrong — iterate.
Step 3 — Calibrate ambition. >85% probability per KR → sandbagged, stretch it. <30% → unreasonable, compress. Target: 50-70% probability.
Step 4 — Cascade and align. Company OKRs first → Team OKRs derived from company → Individual OKRs (optional). Cross-team dependencies must appear explicitly in both teams' OKRs.
Step 5 — Run the quarterly cycle. Week 1: set. Mid-quarter: score 0.0–1.0. Final week: scoring + retrospective. Carryover: carry, drop, or rewrite.
Step 6 — Score honestly. Each KR scored 0.0–1.0, reported transparently. Scoring is NOT performance evaluation. Sandbagged 1.0s are worse than stretched 0.6s.
Output: Quarterly OKR Sheet
# OKR: Q[X] [Year] — [Team]
## Objective 1: <qualitative, ambitious>
- KR1: [start] → [target] (actual: [x], score: [0.0-1.0])
- KR2: ... KR3: ...
## Dependencies: depends on / others depend on us for
## Confidence: start / mid-quarter / end actual
→ Method in Action: Andy Grove at Intel, 1971; John Doerr at Google, 1999 · Bono's ONE Campaign
→ 2026 lens: OKRs in an AI-native org (2024–2026)
Pack: OKR Patterns
| Domain | Good KR (outcome) | Bad KR (activity) |
|---|---|---|
| Engineering | "Reduce P0 incidents from 4/qtr to 1" | "Implement new deployment pipeline" |
| Sales | "Close 15 new accounts at >$50k ACV" | "Hire 3 mid-market AEs" |
| Product | "Increase day-7 retention from 22% to 40%" | "Launch new onboarding flow" |
| Marketing | "Reach 50k organic blog visits/mo" | "Publish 10 blog posts" |
Applying It Well
Limit ruthlessly: ≤2 Objectives per team, ≤5 KRs each. Celebrate 70% explicitly — punishing it destroys stretch. Decouple scores from comp/performance reviews. Require cross-team OKRs to reference each other.
→ Primary sources: references/sources.md
Common Rationalizations
[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.
| Fake move | Reality |
|---|---|
| [D] "Ship feature X" as a KR | Activity, not outcome. What change in the world does shipping X cause? That's the KR. |
| [D] All KRs hit at 100% | Sandbagged. Recalibrate next quarter. |
| [D] 30% achievement is normal | Goals unreasonable. Target is 60-80%. Recalibrate. |
| [D] 20 KRs across 5 Objectives | Lack of focus. Cut to 1-2 Objectives, 3-5 KRs each. |
| [D] OKRs tied to performance evaluation | Couples stretch with punishment; sandbagging becomes rational. Decouple. |
| [D] Quarterly OKRs match what you'd have done anyway | Then they're a to-do list. The KR must change what the team chooses to do. |
| [D] No scoring at quarter end | Skipping retrospective destroys the learning loop. Score every KR. |
| → Add [O] entries here after each real use — paste the actual failure pattern | What went wrong and why |
Red Flags
- KRs that are activities ("ship," "launch," "hire," "complete") — 100% achievement every quarter
- More than 5 Objectives or 5 KRs per Objective — OKRs invisible across teams
- OKR scores tied to compensation — no quarterly retrospective
Verification
- Each Objective: qualitative, ambitious, one sentence
- Each KR: quantitative, outcome-focused, time-bounded, genuinely stretching (70% target)
- Cross-team dependencies explicit; OKRs visible org-wide
- Quarterly cadence established; scoring honest and decoupled from performance reviews
Part of deciqAI Knowledge Skills — 237 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. See it run → https://www.deciqai.com/s/okr-goal-setting · Built by deciqAI · github.com/deciqAI · Contributions welcome.
Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/okr-goal-setting.json