shiphrahx/ai-for-engineering-leaders/capacity-planning

Capacity Planning

Map team bandwidth against committed work for an upcoming quarter or cycle using realistic availability — not idealistic "every engineer codes 40hrs/week" math — to answer "can we take this on?" and "what do we drop to add it?"

Inputs to gather

Gather these before calculating. If any are missing, ask for them in a single batched question — never invent absences, effort estimates, or velocity. Mark anything genuinely unavailable as Unknown.

• Team — name and size (with seniority mix if individual-level planning is needed)
• Period — the time window and number of weeks
• Planned absences — PTO, holidays, company shutdowns, parental leave, on-call rotations
• Committed work — already-agreed deliverables with effort estimates
• Requested additions — new work being considered

Steps

1. Calculate available capacity transparently, line by line: gross capacity (engineers × weeks), then deduct shutdowns, individual absences, on-call overhead, and a meetings/review/support overhead factor. Show the math; land on net available engineer-weeks.
2. Use a realistic overhead factor: ~20% for an established team; 30–40% for a new or volatile team (more process, learning, coordination). On-call ≈ 0.5 week per engineer per rotation; a full on-call week is roughly 50% reduced capacity. New hires ramp ~25% productive in month 1, ~50% month 2, ~75% month 3.
3. List committed work with estimates and a confidence level each. Subtract from net capacity to show remaining.
4. Test requested additions against remaining capacity, marking each as fitting or not, and state the capacity gap or surplus plainly.
5. If overcommitted, lay out trade-off options — what could be cut, deferred, or descoped — and lead with the gap number to force the conversation.
6. Write a recommendation — what to commit to, what to push back on, suggested sequencing (smaller/higher-confidence work first), and the caveats (e.g. "if X overruns by 3 weeks, Y slips"). Leave an explicit buffer (10–20%) for unplanned work rather than allocating to 100%.
7. Adapt to context: for individual-level planning, break capacity down per engineer when work needs specific expertise ("only two engineers can do the API integration" changes the math). For stakeholder presentation, keep the math visible — accounting for holidays, overhead, and on-call earns more trust than pretending at 100% utilisation.
8. Assemble the output in the format below.
9. Run the calculator with the gathered inputs:

`` python scripts/calculate.py --engineers <value> --sprint-days <value> --leave-days <value> --ceremony-hours <value> --velocity <value> `` Use the printed figures verbatim — do not recalculate manually. If FLAGS are printed, surface them explicitly to the user before proceeding.

Run the script. Fix every failure. Do not return output until the script passes.

Output format

**[Team] — [Period] Capacity Plan**
 
**Available Capacity**
| Factor | Calculation | Engineer-Weeks |
|--------|-------------|----------------|
| Gross capacity | [engineers × weeks] | [N] |
| [absence/shutdown] | [calc] | -[N] |
| On-call overhead | [calc] | -[N] |
| Meetings/review/support overhead | ~[X]% of remaining | -[N] |
| **Net available capacity** | | **[N]** |
 
**Committed Work**
| Item | Estimate | Confidence |
|------|----------|------------|
| [item] | [N weeks] | [high/medium/low] |
| **Total committed** | **[N weeks]** | |
 
**Remaining Capacity:** [net] − [committed] = **[N] engineer-weeks**
 
**New Requests**
| Request | Estimate | Fits? |
|---------|----------|-------|
| [request] | [N weeks] | [✅ Yes ([remaining] left) / ❌ No (gap of [N])] |
 
**Recommendation**
[What to commit to, what to push back on, sequencing, and caveats. Buffer remaining: [N] (~[X]% of net).]

Variants

Optional org-size tuning (see `docs/VARIANTS.md`). Default to growth if the user doesn't state a stage.

• startup (≤ ~15 engineers): use a 30–40% overhead factor (more context-switching), keep the plan to the Available Capacity and Recommendation tables, and skip per-engineer breakdown unless one person is a single point of failure.
• growth (~15–80): the defaults above — ~20% overhead, full committed/requested tables, 10–20% buffer.
• enterprise (80+): add per-team rollups, mark cross-org dependencies explicitly, and note where formal sign-off is required before committing capacity.

Boundaries

• Never invent absences, effort estimates, or historical velocity — mark them Unknown and use a stated default overhead (25%) only when flagged as an assumption.
• Never plan to 100% utilisation — always reserve explicit buffer for unplanned work.
• Never hide an overcommitment — when work exceeds capacity, state the gap first and force the trade-off conversation.
• Effort estimates come from the user; individual velocity varies — say so rather than implying the math is exact.

Chaining

• After this, offer sprint-planning-breakdown to slice the committed epics into well-scoped, estimated tickets that fit the planned capacity.