Most contact centre scorecards measure how busy the team was. Very few measure whether customers got what they needed, which is the only thing that determines whether they stay. The gap between those two things is where most reporting goes wrong, and it is also where most outsourced programmes are judged unfairly, in both directions.
Every metric below can be gamed. That is not a criticism of agents. It is a property of measurement: any number that becomes a target will be optimised for, and the behaviour that optimises the number is rarely the behaviour that helps the customer. So this article treats each KPI in three parts: what it means, how it gets gamed, and which counterweight to pair it with so that gaming one number shows up in another.
First contact resolution
First contact resolution is the share of contacts resolved without the customer needing to contact you again about the same issue within a defined window. It is the single most predictive number available. A customer who has to make contact twice about the same problem is more likely to leave, regardless of how pleasant either conversation was.
It is gamed by narrowing the definition. If a repeat contact only counts when it arrives on the same channel, or is logged under the same category, or comes within a very short window, resolution looks better than it is. It is also gamed by agents closing a case and asking the customer to open a new one.
Pair it with contacts per customer per period and with a customer-side measure such as a follow-up survey asking whether the issue was resolved. If reported resolution rises while contacts per customer also rise, the definition has been narrowed rather than the service improved.
Average handle time
Average handle time is talk time plus hold time plus after-call work, averaged across contacts. It is useful for staffing, because it tells you how much agent time a given volume will consume. It is the most over-weighted number in the industry when it is turned into a target.
Managed hard, it teaches agents to end calls rather than resolve issues. Agents transfer instead of solving, skip the second question, and leave notes incomplete because after-call work counts against them. That raises repeat contact and lowers resolution, producing more total handle time across the operation, not less.
Track it. Do not target it. Pair it with first contact resolution and with quality score, and look at the distribution rather than the average. A handful of very long calls usually points at a process problem or a system problem, not an agent problem.
Track handle time. Do not target it. The moment it becomes a target, agents optimise for it and resolution falls.
Service level and average speed of answer
Service level is the share of contacts answered within a threshold, and average speed of answer is the mean wait before an agent picks up. An 80/20 service level, meaning most calls answered within twenty seconds, is a long-standing industry convention rather than a rule, and the right threshold depends on the contact type and what the customer expects.
It is gamed by counting from the wrong moment. Starting the clock after the menu and announcements, excluding short abandons, or measuring across the whole day so that a bad hour disappears into a good average all flatter the number. It is also gamed by answering quickly and placing the customer straight on hold.
Pair it with abandonment rate, with hold time, and with interval-level reporting so the worst half-hour is visible. A service level that looks fine on a daily average can hide a lunchtime queue that loses customers every day.
Abandonment rate
Abandonment rate is the share of contacts where the customer gave up before an agent answered. It is blunt, but it captures the failure that matters most: nobody was there. It also tells you something answer time does not, which is how patient your customers are willing to be.
It is gamed by excluding short abandons on the theory that they were misdials, and by defining the abandonment window generously. Some exclusion is legitimate, because a caller who hangs up within a few seconds probably did misdial, but the threshold needs agreeing and fixing rather than adjusting month to month.
Pair it with service level and with callback and retry data. If abandonment falls because callers are being offered a callback, that is a genuine improvement. If it falls because the definition changed, it is not.
Quality score
Quality score is the result of a written scorecard applied to sampled contacts, covering accuracy, process adherence, tone and whether the customer's issue was addressed. It is not a manager's impression. It is a defined rubric applied every week, with results fed back to each agent individually.
It is gamed by sampling the wrong contacts. Reviewers who pick short, simple calls, or who let agents nominate their own calls, produce a score that means nothing. It is also gamed by scorecards that weight easy items, such as using the customer's name, as heavily as hard ones, such as giving the correct answer. At volume, speech analytics can check script adherence across every call so that the sample is chosen from the calls that need a human ear.
Pair it with first contact resolution and with customer feedback. Random sampling across contact types and times of day, a scorecard that separates accuracy from manner, and calibration sessions where reviewers score the same call and compare results keep the number honest. Our call centre analytics service builds this into the reporting from the start.
Customer satisfaction
Customer satisfaction is a survey score collected after a contact. It is the customer's view rather than yours, which is its value. Reported as a single site-wide number, it hides everything useful. Segment it by contact type, by channel and by agent, or it cannot drive a decision.
It is gamed by choosing who gets surveyed. Agents who send the survey only after a good call, or systems that survey only resolved cases, inflate the score. It is also gamed by asking the question in a way that measures the agent's friendliness rather than whether the problem was solved.
Pair it with resolution and with survey response rate. A rising satisfaction score alongside a falling response rate suggests the sample has narrowed. Ask the resolution question separately from the courtesy question, and treat the two answers differently.
Contacts per customer per period
This is total contacts divided by active customers over a month or a quarter. Falling contact volume against a growing customer base is the clearest evidence that your product, your documentation and your self-service are improving. Rising volume means you are absorbing a problem rather than fixing it, however well the contacts themselves are handled.
It is hard to game inside the contact centre, which is exactly why it belongs on the scorecard. It can be distorted by changing how contacts are counted or by moving volume to channels that are not measured, so count every channel and keep the definition fixed.
Pair it with the top contact drivers. The number tells you volume is moving. The driver analysis tells you why, and it is the input your product and operations teams need to remove the cause.
Occupancy and adherence
Occupancy is the share of logged-in time an agent spends handling contacts or in after-call work. Adherence is how closely agents follow their schedule. Both are workforce management numbers, useful for staffing and cost, and dangerous when treated as performance targets. Occupancy above a sensible level predicts burnout and error, not productivity.
They are gamed by staying in after-call work, by logging into auxiliary states, and by supervisors adjusting schedules after the fact so that adherence looks clean. Pair them with quality and with attrition. A team that is highly occupied, highly adherent and steadily losing people is being run too hot.
Building a scorecard that cannot be gamed alone
The principle is that no metric stands alone. Every number that measures speed is paired with one that measures outcome. Every number the contact centre reports about itself is paired with one the customer reports. Every definition is written down, fixed for a period, and changed only by agreement, with the change noted on the report.
Every metric on a report should have an owner and a threshold that triggers an action. Numbers nobody is accountable for get read and forgotten. When we set up reporting for a client, the reporting rhythm is agreed before launch, and the first monthly review is where thresholds are set against real data rather than guesses.
Drop the rest from the front page. Total calls handled tells you nothing without context. Emails sent, chats closed and tickets touched are activity, and activity is easy. A single blended score that averages everything into one number removes the information a manager needs. Keep those figures in the appendix for the people who staff the queue. If you are evaluating an outsourced provider, ask which of the numbers above they report by default, how each is defined, and whether the definitions are yours or theirs. Our guide to writing an outsourcing RFP includes the reporting questions worth asking before you sign.
- First contact resolution, paired with contacts per customer
- Handle time, paired with resolution and quality
- Service level, paired with abandonment and interval reporting
- Quality score, paired with customer feedback and calibration
- Satisfaction, paired with response rate and a separate resolution question
- Occupancy and adherence, paired with quality and attrition