esc
  • WorkflowsPages
  • Case studiesPages
  • Machine Learning FoundationsLearn
  • AI EngineeringLearn
  • AI Guardrails & Red TeamingLearn
  • UI/UXLearn
  • Design SystemsLearn
  • Production System DesignLearn
  • Book a callContact
  • Download ResumeContact
Let’s talk

I’ll build the agent for it.

Service and Health Metrics: Measuring Whether a Design System Is Working

Learn which numbers show a design system is reliable and well supported, how to pick a few, set simple targets, and act when they slip.

In short

Service and health metrics are the few numbers that show whether a design system is reliable and well supported, with simple targets and action when they slip.

14 min readFree

What You Will Learn

A design system is a shared set of parts and rules that many teams use. Once it is live, a hard question follows: is it doing well? Many teams answer with a feeling. This lesson shows how to replace that feeling with a small set of numbers. You will learn the two main kinds of numbers, how to choose a few that matter, how to set simple targets, and how to act when a number starts to slip.

Read these key words first. They come back often in the lesson.

Key words

Metric
A number you measure again and again, so you can see how it changes over time.
Health metric
A number that shows the quality of the system itself, such as its open bugs or its documentation.
Service metric
A number that shows how well the team supports the people who use the system.
Consuming team
A product team that takes the system in and uses it in its own work.
Target
The level you want a number to reach, such as a goal for the next quarter.
Threshold
A line that, once crossed, tells the team it must act.
Median
The middle value in a list of numbers. Half the values are higher and half are lower.
Version lag
How many releases a consuming team sits behind the newest version.
Deprecated
Marked as old and planned for removal, so people should move to a replacement.
Vanity metric
A number that looks impressive on a slide but does not change any decision.

Why does a system need numbers?

A design system can look busy and still fail. Teams may keep downloading the library, yet still wait days for answers or quietly build their own versions of the parts. Numbers reveal that gap. One writer notes that many teams build a system and then stop looking at it. Without tracking how it is used, how well it lines up with products, and its quality, the system slowly becomes a cost instead of a help.

A useful way to think about this is to treat the system as a product with customers. Its customers are the designers and engineers in other teams. Health metrics tell you whether the system is sound. Service metrics tell you whether those customers are being looked after. You need both kinds.

Health Metrics: Is the System Itself Sound?

Health metrics look inward. They ask whether the parts, rules, and code of the system are in good shape. Teams that study this area often list the same handful of measures.

  • Accessibility pass rate

    The share of components that pass accessibility checks. Accessibility means building things that people with different abilities can still use, such as text with enough contrast and controls that work with a keyboard. A rising pass rate means fewer parts block people.

  • Test coverage

    You do not need to write the checks to read the number. You only need to know that a low number means more surprises later.

  • Documentation share

    How many components have full, current written guidance that explains when and how to use them. Documentation that is old or missing forces teams to guess, and guessing leads to drift.

  • Open bug count and age

    How many known defects are still unresolved, and how long each one has waited. Age matters as much as count. An old bug that nobody touches signals a neglected part of the system.

  • Version lag

    How far behind the newest release the consuming teams are.

Some teams also give each page or component a simple status label, such as healthy, withering, or dormant. This gives a quick view without reading every number. The labels are only as good as the numbers behind them, so keep the rules for each label written down.

Notice that these measures are about the system, not about the people who ask for help. That is the job of the next group.

Open Bugs and Version Lag

These two numbers give some of the earliest warnings. An open bug count shows what is broken right now. Its age shows how quickly the team responds. Version lag shows whether the system's fixes ever reach the products that need them.

Version lag is easy to misread. A team that stays on an old version may not be lazy. When the upgrade is expensive and the old version still works, waiting feels rational. That is why version lag is a signal to investigate, not a score to blame people for.

Service Metrics: Are Teams Well Supported?

Service metrics look outward. They ask how well the team helps the people who depend on the system. A system can have clean parts and still serve people badly if their questions sit unanswered for weeks.

  • Time to first response

    How long a person waits before someone replies to their request. A quick reply proves the request was seen, even when the full answer takes longer.

  • Time to resolve

    How long it takes from the start of a request to its finish, whether that finish is an answer, a fix, or a decision. It shows the whole journey, not only the first step.

  • Contribution turnaround

    How long a change proposed by someone outside the core team waits for a decision. Long waits teach people that helping improve the system is not worth the effort.

  • Satisfaction survey score

    A short rating from users, often a single question asked soon after an interaction. It tells you how people feel about the help they got, which the other numbers cannot show.

Pair service numbers with health numbers. A slow reply may be a staffing issue. A slow reply that comes with a growing pile of old bugs may point to a deeper problem in the system itself.

Why do medians beat averages?

Averages can hide the truth. Imagine nine requests answered in one day and one request that waited ninety days. The average comes out near ten days, which describes nobody's real experience. The median, the middle value, is one day, which matches what most people actually waited. For response times, a median usually tells the story more fairly.

Tip

When you report a waiting time, state whether it is a median or an average, and keep using the same one. Changing the method between reviews makes trends look better or worse for no real reason.

Quick Check

Quick check

Mei leads a design system. Her dashboard shows 98 percent test coverage and a 95 percent accessibility pass rate. Yet product designers say they wait two weeks for any answer about how to space a card. Which missing measures should she add first?

How do you choose a few meaningful metrics?

Start with goals, not with the numbers your tools happen to count. A metric is a way to check a goal. If your goal is that product teams ship using shared parts without rework, then version lag and documentation share matter. If your goal is that teams feel supported, then response time and satisfaction matter. One source warns that the most common trap is measuring what is easy, such as component count or the tidiness of a design file, instead of what matters, such as adoption and downstream quality.

  1. 1

    Write two or three goals

    Use plain sentences that describe what good looks like for the teams that use the system. Keep them short enough to fit on a sticky note.

  2. 2

    Pick one or two metrics per goal

    Choose numbers that change when the goal is met or missed. Mix numbers that lead, such as how quickly people start using a fix, with numbers that show results, such as satisfaction. Keep the total list short, because too many numbers overwhelm a team.

  3. 3

    Set a target and a threshold

    The target is where you want the number to be. The threshold is the line that triggers action. Write both down before you look at the data, so you cannot move the goal after the fact.

  4. 4

    Define each measure exactly

    Say which requests count as a first response and when a request counts as resolved. Use the same definition every time, so trends stay honest.

  5. 5

    Ask what each number would change

    For every metric, ask: if this number moved, what would we do differently? If you have no answer, drop the metric from the list.

Some teams find it helpful to group their numbers into a short scorecard, with a few health numbers, a few service numbers, and one or two outcome numbers that show whether the system helps the wider organization.

How do you set simple targets?

Keep targets simple enough that anyone can check them. For example, a team might aim to have most of its consuming teams on the latest major version within two quarters, and treat any team more than one major version behind as a warning. Treat these as illustrations for planning, not as standards. Pick values that fit your team's size and history. The first targets are often guesses, so review them after a few cycles and adjust.

Be careful with the phrase set it and forget it. A target with no review date is only a wish.

How do you avoid vanity numbers?

A vanity number is one that looks good but drives no decision. These numbers are easy to collect, which is why they show up in so many reports.

  • Raw download counts

    A download may mean a project was set up once, so download volume is only a partial lens on usage. It does not by itself prove that anyone solves a real problem with the parts.

  • Total component count

    A bigger catalog is not a healthier one if many parts are unused, undocumented, or already deprecated.

  • Design file tidiness

    A clean design file looks professional, but it says little about whether product screens use the parts correctly.

  • Totals with no trend

    One total tells you a size, not a direction. Compare it with earlier periods before you decide anything.

Tip

Ask of every number: would a change in it make us do something different? If not, it is a vanity number, however nice the chart looks. Also remember that too many metrics can overwhelm a team, so fewer good numbers usually beat many weak ones.

Acting on the Metrics

A metric helps only when someone reads it on a schedule and acts on it. Choose a regular rhythm and keep to it.

  1. 1

    Review on a fixed day

    Look at the same short list of numbers, at the same time, with the same people. Keep the meeting brief and focused on change.

  2. 2

    Read the trend, not the single value

    Ask whether each number is rising, falling, or flat across several periods. One bad week may be noise. Three bad periods in a row is a pattern.

  3. 3

    Name the warning sign

    Decide whether a number has crossed its threshold or is only drifting toward it. A crossed threshold needs action now. A drift needs a closer look.

  4. 4

    Turn the finding into follow-up work

    Write one concrete action for each finding, with an owner and a date. A finding without an owner is just a worry.

  5. 5

    Tell the users what changed

    Share the finding and the plan with the teams it affects. People who see that someone is listening are more likely to report problems and to keep their versions current.

  6. 6

    Check again next cycle

    Watch whether the number moves. If it does not, the action was wrong or too small, so try a different one rather than repeating the same step.

Action sometimes needs a plan for the consuming teams, not just for the system team. These steps turn a warning into a safe change.

Warning Signs to Watch

  • A growing backlog

    Requests pile up faster than the team closes them. Watch both the count and the age of the oldest requests.

  • Teams stuck on old versions

    Version lag rises, or several consuming teams stay on a deprecated version for a long time. A single stubborn team is a question. Many stubborn teams are a sign of a hard upgrade path.

  • Slow first replies

    The median time to first response creeps up over several review cycles. One slow week is noise. A steady climb is a signal.

  • Falling satisfaction

    Survey scores drop, or the comments keep repeating the same complaint. Repeated comments are often the most useful part of a survey.

Tip

When a number gets worse, look for the cause before you blame people. Many delays come from the cost of upgrading, unclear guidance, or too few people on the support rotation, not from teams refusing to help.

Worked Example: Slow Replies and Old Versions

Imagine a team that runs a public design system such as Material Design or GitHub's Primer. The numbers below are invented to show the method. They are not real figures from those teams. The team's dashboard shows two warning numbers. The median time to first response on component requests has grown from 1 day to 6 days. Several product teams are two major versions behind the latest release.

  1. 1

    Notice the change

    During the regular monthly service review, the team sees the median first response climb from 1 day to 6 days. It does not wait for a crisis, because the trend is clear on the chart.

  2. 2

    Read the two numbers together

    Slow replies and old versions may share a cause. Product teams may be unsure which fixes apply to them, so they wait for an answer and then stay on the version they know. The team writes this guess down as a hypothesis to test.

  3. 3

    Add a triage rotation

    Triage means sorting new requests by urgency and making sure each one gets a first reply. The team creates a rotation, so one person each week owns this job. The fix targets the reply time directly.

  4. 4

    Write a migration guide

    A migration guide is a document that shows, step by step, how to move from the old version to the new one. The team writes one with before-and-after examples and names the replacement for each deprecated part. It also gives teams one or two release cycles to move, so no one is forced to upgrade overnight.

  5. 5

    Tell the product teams

    The team announces the new triage rota, shares the guide, and asks each product team to name a contact. Clear messages make it obvious what each team should do next.

  6. 6

    Watch both numbers recover

    In the scenario, the team reviews the median response time and version lag each cycle. The expected result is that response time falls back toward its earlier level and more teams move onto a current version. If one number does not recover, the team treats that as a new question rather than assuming the fix worked.

Tip

A good improvement story has two parts: a number that moved and an action that explains why. Record both, so the next review can tell what worked.

Key Takeaways

Health metrics tell you whether the system is sound. Service metrics tell you whether the people using it are supported. Choose a few numbers tied to clear goals, set simple targets, and skip vanity numbers that change nothing. Review the same numbers on a steady rhythm, watch for trends and warning signs, and turn every finding into a named action and a clear message to users.

Quick check

Your design system dashboard shows 5,000 weekly downloads of the component library. Support requests sit unanswered for two weeks, and a third of consuming teams are on a deprecated version. Is the system healthy, and what should you look at first?

Sources

How is this lesson?

Last updated on 17 November 2026

Learn with me

Every lesson is free. I add new ones as I learn. Have a question, or want help with your own work? Book a short call.