Handle time is the metric contact centers have always tracked. It's visible, measurable, and unambiguous -- the call lasted X minutes. What handle time doesn't capture is whether those minutes produced a resolution, whether the customer will call back next week with the same issue, or whether the agent is going to repeat the same mistake on the next 50 calls because nobody showed them what they were doing wrong.
Under-coached agents are not primarily a handle-time problem. They're a resolution quality, customer retention, and cumulative operational cost problem -- and those costs are harder to see on a dashboard.
The Compounding Cost of Unrectified Errors
An agent who consistently resolves calls incompletely -- leaving the customer with a partial fix or an unclear next step -- generates repeat contact. A customer who calls back about the same issue within 30 days has had two interactions that should have been one. The cost isn't just the second call -- it's the first call plus the second call plus whatever customer satisfaction impact the pattern produces over multiple similar experiences.
In a contact center processing 15,000 calls per month, a 12% repeat-contact rate is 1,800 calls that represent prior interactions that didn't stick. If a third of those repeat contacts are attributable to agent behavior that coaching could correct, the coaching gap is producing roughly 600 unnecessary calls per month. At $6-8 per interaction (a conservative contact center cost estimate), that's $3,600-4,800 in avoidable operational cost before you account for customer satisfaction impact.
The agents producing those repeat contacts are typically not identified as problems. Their handle time is acceptable. Their escalation rate is average. Without transcript-level scoring, the pattern doesn't emerge from the noise.
The Cost Agents Can't See and Supervisors Can't Catch
Agents who don't receive development feedback tend to stabilize at whatever performance level they reached after onboarding. Without specific, data-grounded feedback, most agents don't know which aspects of their calls are producing poor outcomes. They may sense that something isn't working -- customer frustration, repeat callers -- but they can't self-correct without knowing what to change.
Supervisors in manual programs face the same limitation from the other side. They know agent performance varies, but without scoring data on more than 2% of calls, they can't identify which agents are consistently underperforming on which criteria. The agents who need the most development stay invisible to the program unless they generate an escalation or a complaint.
What Gets Better When Under-Coached Agents Get Coaching
The most consistent effect of targeted coaching is reduction in criterion-specific failure rates. An agent who is scoring poorly on resolution commitment -- closing calls without confirming the next step -- has a specific behavior to change. A coaching session that shows them what a strong resolution commitment looks like versus what they're currently doing, drawn from their own call transcripts, produces measurable improvement on that criterion within 2-3 weeks.
This is different from general performance management. The agent isn't receiving a global assessment of "needs improvement." They're receiving a specific, call-backed observation: your resolution commitment score is 2.1 out of 5 on your last 40 calls, the team average is 3.8, and here is a call where it failed and a call where it succeeded. That's actionable.
Why the Status Quo Persists
The reason under-coaching persists is not that operations leaders don't understand its cost. It's that they don't have the data to act on it systematically. Manual QA produces enough information to identify the worst-performing agents and the most dramatic problems. It does not produce enough information to identify the large population of mid-range agents who are slightly underperforming on specific criteria and would improve with targeted coaching.
That's the gap where most of the cost lives. The agents who generate escalations and formal complaints are a small population that manual programs catch. The agents who are consistently scoring a 2.4 on a rubric criterion that should be 3.5 and generating 15% repeat contact as a result -- those agents are invisible to a 2% sampling program. They're not invisible to transcript scoring at 100% coverage.
Handle time will stay on the dashboard. The question is whether the metrics underneath it start telling you what handle time can't.