What it is and why it matters
Not every item costs the same when it is late. A defect that stops customers from paying costs a lot every hour. A legal compliance change costs little until the deadline and a lot after it. A library upgrade costs nothing today, but can turn into a crisis in six months. Handling everything in a single queue, ordered by whoever speaks loudest, produces expensive decisions.
The Kanban Method answers this with classes of service: explicit policies that say how each type of risk is handled in the flow. The question stops being “what is most important?” and becomes “what costs the most to wait?”.
Key concepts
Cost of Delay and its profiles
Cost of Delay is the value lost per unit of time while an item is not delivered: revenue that does not come in, a penalty that draws closer, a risk that grows. The concept was spread by Don Reinertsen in product development. The way this cost evolves over time gives rise to typical profiles:
- Expedite: high and immediate cost. Every day counts, starting now.
- Fixed Date: low cost until a date and high, sometimes total, after it. An event, a regulatory requirement, a market window.
- Standard: cost that starts right away and grows fairly continuously. Most work falls here.
- Intangible: small cost today, which can rise sharply in the future. Technical debt, upgrades, prevention.
You do not need an exact monetary value: recognizing the shape of the curve already changes the decision.
Classes of service and their policies
Each profile becomes a class with its own policies. A plausible example for a product team:
- Expedite: at most one item at a time in the system; can jump ahead of everything; someone with availability pulls it immediately.
- Fixed Date: enters the system ahead of time, with a margin calculated from the lead time percentiles, not the day before; it is tracked by age.
- Standard: pulled in order of arrival within its allocation, with the service’s SLE.
- Intangible: has a reserved allocation so it is not postponed forever; can give way to Fixed Date items at risk.
The policies should be on the board, usually in a swimlane or by card color, and apply to everyone. The value of the class lies in the agreed rule, not in the label. Explore the differences in the classes of service experience.
CD3: Cost of Delay divided by duration
When several items compete for the same capacity, a useful economic rule is to do first what has the highest Cost of Delay per unit of duration. The name of this ratio is CD3 (Cost of Delay divided by duration).
Consider two items. Item A costs $10,000 per week of delay and takes 2 weeks. Item B costs $6,000 per week and takes 1 week. The CD3 of A is 5,000; that of B, 6,000. Doing A first, the total Cost of Delay is $20,000 from A plus $18,000 from B, which waits 3 weeks: $38,000. Doing B first, it is $6,000 from B plus $30,000 from A: $36,000. The “most valuable” item is not the one that should go first. Work through this reasoning in the sequencing by Cost of Delay challenge.
CD3 is a guiding idea, not a formula with false precision: cost and duration are uncertain estimates, and the gain comes from making the conversation economic and explicit.
Capacity allocation
Capacity allocation splits the system’s WIP across types or classes of work. With a total limit of 10 items, a team can reserve 6 for Standard, 2 for Fixed Date and 2 for Intangible, with one extra slot for Expedite. This ensures that improvements and technical debt move forward even under pressure. The proportion is reviewed in the cadences, with demand data.
Expedite abuse
When everything is urgent, nothing is. The signs are well known: the expedite lane is never empty, Standard items age and the service’s predictability disappears. The answer is policy, not effort. Define who can declare an item urgent, limit how many can exist at once, record each occurrence and examine the proportion in the Service Delivery Review. Many “emergencies” are actually Fixed Date items that entered too late.
Prioritizing and sequencing
Prioritizing usually means ordering a long list all at once, often weeks before the work starts. That order goes stale quickly and eats up discussion time. In the Kanban Method, the decision is split into two moments. In the Replenishment Meeting, the team and the requesters select, among the available options, which ones cross the commitment point, considering Cost of Delay, risk and capacity. After commitment, the pull order follows the class of service policies. This way the decision is made as late as possible, with the best information available.
Common mistakes
- Classes without policies. Labeling a card “Fixed Date” without an entry rule or tracking changes nothing.
- Starting the Fixed Date item the day before. The start margin should come from the lead time percentiles, as in the fixed date commitment challenge.
- No allocation for Intangible. The work that prevents crises never starts, until it becomes one.
- Treating backlog order as a commitment. The order of options is not a delivery promise.
How this shows up on the exam
Questions ask for sequencing and policy decisions. Examples of the kind of situation: two items with different costs and durations competing for one slot; a Fixed Date item due in 15 days in a system whose 85th percentile is 18 days; an expedite lane that is always occupied; a team that wants to reserve capacity for technical debt. You need to choose the action most consistent with Cost of Delay and with the policies, and recognize an item’s cost profile from its description.
Frequently asked questions
Are class of service and work item type the same thing?
No. The work item type describes the nature of the work, such as defect or feature. The class of service describes how it is treated according to the risk of delay. A defect can be Expedite or Standard, depending on its impact.
Do I need to calculate Cost of Delay in money?
Not necessarily. Recognizing the profile of the curve already improves the decision. When data is available, estimating approximate values lets you compare items with CD3; what matters is the order of magnitude and the explicit conversation.
How do I keep everything from becoming urgent?
With policy: who can declare, how many can exist at the same time and periodic review of occurrences. Reviewing urgent items in the cadences shows whether the problem is real demand or items that enter too late.
