Drafting OKRs from Jira, GitHub and your HR system with AI
How OKR.ai uses MCP to turn real projects and teams into a first draft worth editing.
How to break one number into parts that genuinely sum, including ratios and the dependent ones.
A company key result for the year says “grow ARR from $4M to $10M”. Six teams write key results that support it. At the end of the year, every team hits its target, and the company still misses. This happens constantly, and it’s almost always arithmetic, not effort. Here’s how to split a number so the parts genuinely add up to the whole.
Key takeaways
When each team writes its OKRs separately, three things go wrong:
The fix is to split the number deliberately, from the top, before teams write their own OKRs around their slice.
| Type | Examples | How children combine |
|---|---|---|
| Additive | Revenue, bookings, deals, sign-ups, hires, cost savings | Children sum to the parent. |
| Ratio / rate | Conversion, win rate, retention %, uptime, NPS | A weighted average, by each child’s volume. |
| Threshold | p95 latency, time to hire, security findings | The parent is as good as the worst child, or a percentile across them. |
| Dependent | “Ship SSO” enabling enterprise bookings | Contributes to the parent but doesn’t sum. Link it and name the dependency. |
Write the parent as a formula of things specific teams control. For ARR growth:
Net new ARR = new business + expansion + self-serve − churn
To go from $4M to $10M, the company needs $6M of net new ARR. Suppose last quarter’s data suggests churn of about $1M is likely without intervention. Then gross additions need to reach $7M.
An even split (“three teams, $2M each”) ignores how different the channels are. Size each slice from its own baseline and capacity:
| Owner | Key result | Baseline | Target | Contributes |
|---|---|---|---|---|
| VP Sales | New mid-market bookings (ARR) | $0.8M | $4.8M | +$4.0M |
| Head of CS | Expansion ARR | $0.3M | $2.1M | +$1.8M |
| Head of Growth | Self-serve ARR | $0.4M | $1.6M | +$1.2M |
| Head of CS | Gross churn (ARR), held at or below | $1.3M | $1.0M | −$1.0M |
| Company: net new ARR | +$6.0M | |||
Ratios are where most roll-ups quietly break. Say the company key result is “trial-to-paid conversion from 10% to 14%”, and two segments run trials:
Two rules follow. Set ratio targets per segment together with the volume they apply to. And remember that shifting mix moves the company number even when no segment improves: more mid-market trials alone would lift company conversion here.
Some of the most important contributions don’t add numerically. Product shipping SSO doesn’t produce ARR by itself, but enterprise bookings can’t happen without it. Treat these as contributing key results:
In OKR.ai a key result can roll up into a super key result, usually your manager’s, so each child’s movement contributes to the parent. When you push OKRs to your reports, AI splits them so they cover the parent with no gaps and little overlap. It sizes additive targets to each owner’s realistic share, treats ratios as contributing rather than summed, and writes dependencies into the objective description. You review and edit every number before anything is saved.
See how it works in the OKR workflow docs, or start with the basics in how to write good OKRs.
Aim a little higher, often 105–115%, so the parent still lands if one slice misses. Make the buffer explicit so nobody mistakes it for padding.
Uptime and latency don’t sum. Either give the company metric to one owner with contributing key results underneath, or define the parent as a weighted average or a worst case across services.
Link its key result to the one it moves most, and mention the others in the description. Counting it under several parents is how double counting starts.
Both. The manager proposes the split from the parent target; each owner checks it against their own baseline and capacity and pushes back with data.