The Measures Worth Reporting

Contact centre quality produces a great deal of measurable activity. Most of it does not indicate whether anything is improving.

Where workload context is required alongside QA outcomes, employee productivity reporting can provide a separate operational data layer without being merged into the quality score.

The measures that matter

Compliance failure rate, from full coverage, by requirement and by queue. Objective, complete, and directly actionable.

Repeat contact rate within a defined window. The strongest available proxy for whether the interaction actually worked.

Customer feedback, with the response rate stated, and read as a distribution rather than an average.

Process findings raised and fixed. The count of operational problems identified and closed. This is the measure that shows the programme is producing value beyond monitoring.

Inter-rater agreement, which tells you whether any of the human scores mean anything.

Coverage: proportion of interactions analysed, and proportion of the required scorecard items actually assessable.

Dispute rate and upheld rate, which is the programme measuring itself.

Seven measures. That is enough for a monthly report.

The measures that mislead

Average QA score. Compressed, gameable, and its movement is mostly noise given the sample sizes.

Agent rankings on QA score, which given inter-rater disagreement and small samples rank the evaluator assignment substantially.

Sentiment averages, which reflect the call mix.

Number of evaluations completed, which measures activity.

First contact resolution as usually calculated, which has its own note because the definitions in use are mostly not measuring resolution.

Average handle time in isolation, which is optimised at the expense of repeat contact if targeted alone.

Any single blended quality percentage, which combines heterogeneous items under arbitrary weights.

Reading them together

Compliance failures falling while repeat contact is flat: compliance improved and the customer experience did not. Both true and worth knowing.

QA scores rising while repeat contact rises: the score has decayed. This is the pairing that detects gaming and it should be on every report.

Sentiment worsening in one category while volume rises: something specific is going wrong, and the category is where to look.

Agreement falling: the scorecard or the evaluator pool has drifted, and every score in the period is less reliable.

Dispute upheld rate rising: scoring accuracy is degrading, possibly after a model update.

The pairing rule

Never report a quality measure without an outcome measure alongside it.

The single most useful discipline in this area. A quality score with no outcome next to it invites the assumption that the score is the goal, and the assumption produces the gaming described elsewhere.

What to report to whom

Agents: their own evaluations, the specific calls, what to change. Not a ranking.

Team leaders: their team's profile, process findings affecting their team, calibration status.

Operations: compliance rate, repeat contact, process findings and fixes, coverage.

Compliance and risk: failure rate by requirement, exceptions, evidence of the control operating.

Executive: compliance position, the outcome trend, process fixes and their value, and what remains unaddressed.

Each of these is one page. A single dashboard serving all five serves none of them, and it is what most operations build.

The pairing table

A single table that enforces the discipline of never reporting a quality measure alone.

Each row: a quality measure, its paired outcome measure, and both trends.

Compliance failure rate, paired with complaints or regulatory findings.

QA behavioural score, paired with repeat contact rate.

Sentiment trend by category, paired with survey scores for that category.

Silence per call, paired with handle time and repeat contact.

Automated compliance detection, paired with dispute upheld rate.

Reading down the outcome column tells you whether the quality work is doing anything.

Reading across each row tells you whether that particular measure is still honest.

A quality measure whose paired outcome has not moved in a year is either measuring something that does not matter or has been optimised, and either finding is worth having.