Most support dashboards measure everything and reveal nothing. A short guide to what each customer service metric actually tells you, and which combinations contradict each other.
Three satisfaction metrics, three different questions
CSAT, NPS and CES get treated as interchangeable and are not. Each asks a different question, on a different timescale, and using the wrong one produces a number that is accurate and useless.
- CSAT asks how satisfied someone was with a specific interaction. It is immediate, granular and good for spotting a broken queue or a struggling agent. It says nothing about loyalty.
- NPS asks how likely someone is to recommend you. It is a relationship measure, moves slowly, and is heavily influenced by product and price rather than by the support call. Reading it as a support scorecard is a category error.
- CES asks how much effort the customer had to expend. It predicts repeat purchase and churn better than satisfaction does, because customers forgive problems and remember friction.
If you can only run one on the contact itself, CES is usually the most actionable, because a high-effort interaction always has a specific, fixable cause.
The operational metrics that matter
Service level — the share of contacts answered within a target time — is the staffing metric, and it should be stated as a pair. "Eighty per cent within twenty seconds" is a target; "we answer quickly" is not. Abandonment rate is service level's honest companion, because it counts the people who gave up rather than the ones who waited.
Average handle time is a capacity planning input that becomes destructive the moment it is used as a performance target, because the fastest way to reduce it is to stop solving problems. Occupancy tells you how hard agents are working within their staffed hours and is the metric most predictive of attrition; sustained above the mid-eighties, people leave.
The contradictions to design around
Several of these metrics pull directly against each other, and a scorecard that targets all of them equally guarantees that the team optimizes for whichever is measured most visibly.
- Handle time versus first contact resolution. Faster calls resolve less.
- Occupancy versus service level. Running agents hot means nobody is free when a spike arrives.
- Cost per contact versus effort. The cheapest routing is usually the most effortful for the customer.
Decide the hierarchy explicitly and write it down. A center that knows resolution beats handle time behaves differently from one where both sit on the same dashboard at the same size.
A dashboard that fits on one screen
Service level and abandonment for whether you are staffed. First contact resolution and repeat contact rate for whether you are solving. CES or CSAT for how it felt. Occupancy for whether the team is sustainable. Contacts per customer for whether the underlying problem is shrinking. Six numbers, each answering a question somebody would actually ask.
Everything else is diagnostic detail, useful when one of those six moves and unnecessary when none of them has.
See how CX reporting is built, or review the KPIs we report on client programs.
Every metric can be gamed, and knowing how is the defence
Any number attached to a target changes behaviour, usually in ways nobody intended. Knowing the specific distortion each metric produces is more useful than choosing metrics carefully once.
Average handle time under pressure produces rushed closes, transfers used as an exit, and repeat contacts that move the cost rather than removing it. Service level produces cherry-picking where routing allows it, and short-call padding to lift the answered count. Satisfaction surveys produce agents soliciting good scores, or quietly avoiding sending surveys after difficult contacts. First contact resolution produces reclassified repeats and issues closed prematurely. Quality scores drift upward when the same team both coaches and scores. None of this is dishonesty so much as ordinary response to incentive. The defence is pairing: never target a speed metric without a quality metric beside it, watch the ratio between them rather than each alone, and audit the mechanism — survey send rates, transfer reasons, disposition accuracy — as seriously as the score.
Leading and lagging, and the gap between them
Most service dashboards are almost entirely lagging: satisfaction, resolution, churn — all reported after the period they describe, which makes them useful for accountability and useless for intervention. A dashboard that only tells you last month went badly cannot help you change this month.
Add the leading indicators that move first. Forecast accuracy and schedule adherence predict service level before it fails. Queue-wait distribution, rather than the average, shows the tail that produces abandonment. Backlog age on asynchronous channels predicts satisfaction days ahead. Agent tenure mix predicts quality and handle time. And new-hire ramp progress predicts the coming quarter's capacity. The practical rule is that leading indicators get reviewed weekly and lagging ones monthly, because reviewing a lagging metric weekly produces noise-chasing rather than management.
Different audiences need different numbers
A single dashboard serving executives, operations managers and team leaders serves none of them, and the usual failure is that everyone gets the operational view. The reporting layer should differ by what the reader can actually act on.
A team leader needs today and this week at agent level: adherence, quality samples, queue state, coaching actions outstanding. An operations manager needs the month at driver and channel level: volume against forecast, service level, resolution by driver, backlog, and the ramp pipeline. An executive needs three or four numbers with direction and cause — cost per resolved contact, satisfaction trend, the largest contact drivers and what is being done about them, and any risk to coverage. Executives given the operational dashboard reliably fixate on whichever number moved most, which is usually noise. Decide who each report is for before choosing what goes on it.
Frequently asked questions
Should we run NPS on support interactions?
Generally not, though it is extremely common. NPS measures the relationship with your company, and it is driven mostly by the product, the price and the overall experience rather than by a single support contact. Asking it after a call produces a number that swings on things the support team cannot influence, which makes it unfair as a scorecard and unreliable as a signal. Keep NPS as a periodic relationship survey and use CSAT or CES on the interaction.
What survey response rate should we expect?
Post-interaction surveys commonly land somewhere in the single digits to low teens, and lower is not automatically a problem — what matters is whether responses are representative. The known bias is that very satisfied and very dissatisfied customers answer more readily than the indifferent middle, which flattens your view of the ordinary experience. Watch the trend rather than the absolute, and be sceptical of a sudden improvement that coincides with a change to how the survey is sent.
Is average handle time useless?
Not at all — it is essential for forecasting and capacity planning, and you cannot staff a center without it. The problem is exclusively what happens when it becomes an agent target: the fastest way to reduce handle time is to stop solving problems, transfer more, and end calls early, all of which increase repeat contacts and total cost. Use it to plan, watch it for unexplained drift, and keep it off individual scorecards.
How many metrics should be on an agent's scorecard?
Few enough that an agent can hold them in their head and act on them during a shift — three or four is a workable ceiling. Long scorecards do not produce balanced behavior; they produce agents optimizing for whichever metric their supervisor mentions most often, which is effectively a random selection. Choose the small set that reflects what you actually want, and keep the remaining measures at team level as diagnostics rather than targets.
Running the operation
Keep reading
The rest of this cluster, for the question you are actually working through.
- First Call Resolution: The Metric Most Contact Centers Measure Wrong
- How to Reduce Call Abandonment Without Just Adding Agents
- Call Center Attrition: What It Really Costs and What Reduces It
- Seasonal Support Staffing: Covering a Peak Without Carrying It All Year
- Writing a Call Center SLA That Means Something
- PCI Compliance in a Call Center: Where Card Data Actually Leaks