Capacity Allocation

The idea in one minute#

A quarter does not contain a quarter’s worth of new work. Support escalations, incidents, security findings, bugs, migrations and the work somebody promised a customer all arrive whether or not they were planned, and together they routinely consume a third to a half of a team’s time. Teams that plan as though all capacity is available do not get more done; they get the same amount done and a plan that was wrong from the first week, plus an uncomfortable conversation every month about why. Allocating capacity to categories up front is not pessimism. It is the only version of the plan that can survive contact.

A picture#

flowchart LR
  CAP[":i-users: <b>A quarter of team time</b>"] --> NEW[":i-rocket: <b>New work</b><br/><small>the only part anybody plans</small>"]
  CAP --> KEEP[":i-wrench: <b>Keeping it working</b><br/><small>bugs, incidents, support</small>"]
  CAP --> DEBT[":i-recycle: <b>Debt and migrations</b>"]
  CAP --> OBL[":i-gavel: <b>Obligation</b><br/><small>security, compliance, contractual</small>"]
  NEW --> SHIP[(":i-package: <b>What actually ships</b>")]
  CAP -.->|"planned at 100% new"| OVER[":i-triangle-alert: <b>Wrong by week two</b>"]
  class CAP neutral
  class NEW compute
  class KEEP,DEBT queue
  class OBL warn
  class SHIP memory
  class OVER warn

Four categories compete for one budget. Only the first is usually on the plan, and the other three do not wait to be scheduled.

How it really works#

The four categories#

New work is what the roadmap describes and what everybody means by progress.

Keeping it working is bugs, incidents, support escalations, and the question from another team that takes half a day. It is not optional and it is not predictable item by item, but it is very predictable in aggregate, which is what makes it plannable.

Debt and migrations is work that makes future work cheaper. It never scores well against new work in a quarterly comparison, which is exactly why it needs a protected allocation rather than a place in the queue.

Obligation is security findings, compliance deadlines and contractual commitments. Not negotiable at any priority score.

Measure before you argue#

The allocation argument is usually conducted on opinion. It does not have to be: look at the last two quarters and categorise what actually happened. Most teams discover that keeping it working was somewhere between twenty five and forty percent, which is both higher than the plan assumed and stable enough to plan against next time.

That measurement converts the conversation from “we should spend less time on bugs” into “here is what it costs, so here is what we can commit to”, which is a conversation you can win.

Protected allocations beat good intentions#

An allocation only works if it is defended when the quarter gets tight, which is precisely when debt work is first to be cut. The practical protection is to treat the allocation as a commitment of the same kind as a roadmap item, with a named owner and something visible to show at the end of the quarter. Debt work with no deliverable is debt work that gets cancelled.

The unplanned work ratio is a signal#

If keeping it working rises quarter over quarter, that is a quality problem being paid for in capacity rather than fixed. If it falls after an investment in reliability, that is the return on that investment, and it is the clearest number you will ever have for justifying the next one.

Small teams, lumpy work#

On a team of three or four, a single incident week is a large share of the quarter, and averages are misleading. The honest version at small scale is to plan less, keep the committed list very short, and treat the rest as intended rather than promised.

Run the numbers#

What a quarter actually holds, and what that does to a plan built on the headline number.

// capacity.go: what is left after everything that was not on the plan.
package main

import "fmt"

const (
	people       = 6.0
	weeksQuarter = 13.0
	holidayRate  = 0.08 // leave, public holidays, sickness
	meetingRate  = 0.15 // rituals, interviews, planning, reviews
)

type bucket struct {
	name  string
	share float64 // share of remaining engineering time, measured from last quarter
}

func main() {
	gross := people * weeksQuarter
	available := gross * (1 - holidayRate) * (1 - meetingRate)
	fmt.Printf("gross capacity      %6.1f person weeks\n", gross)
	fmt.Printf("after leave         %6.1f\n", gross*(1-holidayRate))
	fmt.Printf("after meetings      %6.1f\n\n", available)

	buckets := []bucket{
		{"keeping it working", 0.30},
		{"debt and migrations", 0.12},
		{"obligation", 0.08},
	}
	left := available
	fmt.Printf("%-24s %14s %10s\n", "category", "person weeks", "share")
	for _, b := range buckets {
		w := available * b.share
		left -= w
		fmt.Printf("%-24s %14.1f %9.0f%%\n", b.name, w, 100*b.share)
	}
	fmt.Printf("%-24s %14.1f %9.0f%%\n", "new work", left, 100*left/available)
	fmt.Printf("\nA plan assuming %.0f person weeks of new work has %.0f.\n", gross, left)
	fmt.Printf("That is %.0f%% of the headline number, which is where plans go wrong in week two.\n",
		100*left/gross)
}

The final percentage is the number to carry into planning. A team of six for thirteen weeks sounds like seventy eight person weeks of capacity, and the new work available is a fraction of that once leave, meetings and the three unplanned categories are taken out. Planning against the headline figure is not optimism, it is arithmetic nobody did, and it produces a plan that is already wrong before anybody has written code.

Sector lens#

Remember this#

• A quarter does not hold a quarter of new work. Four categories compete for the time.
• Measure the last two quarters before arguing about allocation. The number is stable.
• Protect the debt allocation with an owner and a deliverable, or it is cut when things get tight.
• A rising unplanned ratio is a quality problem being paid for in capacity.
• On a small team the variance dominates the average. Commit to less.

Try it#

1. Run capacity.go with your own team size and measured shares. What percentage of the headline number is actually available for new work?
2. Categorise last quarter’s completed work into the four buckets. Compare the real split to what your plan assumed.
3. Give your debt allocation a named owner and one visible deliverable this quarter, then see whether it survives.

Check yourself#

1. Why does planning at full capacity produce the same output as planning honestly, but a worse quarter?
2. Why does debt work need a protected allocation rather than a place in the priority queue?
3. What does a rising “keeping it working” share tell you, and what would make it fall?