The QA Workflow, End to End
From selecting an interaction to a closed action. Most programmes have the first three steps and stop, which is why the output does not change anything.
A QA programme is a workflow. Where it breaks is usually at the end, after the evaluation and before anything happens.
The steps
One: selection. Which interactions are evaluated, and why.
Two: evaluation. Scoring against the scorecard.
Three: calibration check, periodically, on a shared set.
Four: delivery, to the agent.
Five: action, by the agent or by whoever owns the underlying cause.
Six: verification, that the action happened and had an effect.
Seven: aggregation, into findings about the operation.
Eight: routing, of process findings to their owners.
Nine: closure, with the outcome recorded.
Most programmes execute one through four and treat the rest as optional. Five through nine are where the value is.
Selection
Random sampling is the default and is rarely the best use of scarce evaluator time once analytics provides coverage.
Better: a mix. Calls flagged by compliance rules. Calls with poor outcomes. Acoustic outliers. A category under investigation. Plus a small random sample to catch what the rules miss.
Record why each call was selected, because a queue selected entirely by rules has a blind spot and the random component is what reveals it.
Delivery and action
Timeliness matters more than thoroughness. Feedback a month late is about an event nobody remembers.
One or two actions, specific, as covered in the feedback note.
Recorded, so the next conversation can reference it.
Verified. The most commonly skipped step: did the behaviour change, observably, in a later interaction. Without it, coaching is a conversation rather than an intervention.
Aggregation and routing
Individual findings roll up into patterns.
A pattern across agents is not an agent finding. It goes to the process owner with the quantification described in the root cause note.
Track routed findings to closure, with the owner named and a date. Findings raised into a void stop being raised.
The measures of the workflow itself
Distinct from the quality measures, and worth watching.
Time from interaction to evaluation.
Time from evaluation to delivery.
Proportion of evaluations delivered at all, which is frequently well below 100 percent.
Proportion with a recorded action.
Proportion verified.
Process findings raised, routed and closed.
A programme scoring calls that are never discussed with anyone is producing a number and nothing else, and these measures are how that becomes visible.
Where it usually breaks
Delivery. Supervisors are busy, the coaching conversation is unpleasant, and the evaluation sits undelivered.
Verification. Nobody checks whether anything changed.
Routing. No path from a finding to a process owner.
Closure. Findings raised and never resolved, so people stop raising them.
Each of these is an ownership problem rather than a tooling problem, and buying a platform does not address any of them.
The minimum viable version
For a small operation with no formal programme:
Select five calls a week per agent, mixed random and flagged.
Score against a ten-item scorecard.
Discuss two of them, weekly, for fifteen minutes.
Record one agreed action and check it next week.
Note anything that is not an agent problem, and send it somewhere.
That is a functioning programme. It requires no software and it executes all nine steps, which is more than most large operations manage.
Measuring the workflow itself
Six numbers that reveal whether the programme is a process or a formality.
Evaluations completed against planned.
Median days from interaction to evaluation.
Median days from evaluation to delivery.
Proportion of evaluations actually delivered to the agent. This is frequently well below 100 percent and nobody reports it.
Proportion with a recorded, specific action.
Proportion where the action was verified in a later interaction.
The last two are usually near zero, which explains why the same findings recur.
Report these alongside the quality measures, because a programme scoring calls that are never discussed is producing a number at considerable cost.