Spam calls consume capacity and cost, but they should not inflate customer-success metrics. Classify them transparently and keep their operational impact visible outside the eligible-customer completion rate.
Spam exclusion needs an auditable rule
Spam-call classification separates irrelevant or abusive traffic from legitimate service interactions. Distinguish confirmed spam, wrong numbers, unclear calls, sales enquiries, and difficult but valid customer requests. An awkward conversation is not automatically spam.
For a proposed report, preserve total inbound volume, classification counts, eligible calls, and unresolved classification. Record the rule version and allow reviewed corrections. Apply the same definition before and during the pilot.
Do not let the assistant improve its score by marking failed customer calls ineligible. Review a sample of excluded calls and track classification disagreements separately from workflow failures.
Calculate a fictional spam-adjusted result
A pilot receives 200 calls. Review identifies 50 spam calls and ten wrong numbers, leaving 140 eligible interactions under the stated rule. If 98 complete the supported workflow correctly, eligible completion is 98 divided by 140, or 70 percent.
Total answered volume remains 200, and all attributable handling cost belongs in the operating-cost view. Excluding spam from the customer-outcome denominator does not make the resources it consumed disappear. These numbers are illustrative, not actual pilot data.
| Call category | Reporting treatment |
|---|---|
| Confirmed spam | Exclude from customer completion; retain cost and volume |
| Wrong number | Separate disposition under the agreed rule |
| Unclear audio | Review rather than automatically call it spam |
| Legitimate unsupported request | Show eligibility decision transparently |
| Failed valid intake | Keep in the eligible outcome funnel |
Test repeated spam patterns and a legitimate caller with similar wording. A broad filter that removes all short calls can hide frustrated customers who disconnected after a failure.
Balance filtering and access
Aggressive spam blocking can free capacity but reject legitimate callers. Conservative classification increases handling cost and staff noise. Use reviewed evidence and an approved recovery route for false positives where the service supports one.
Spam prevalence can vary by number, time period, or marketing exposure. Compare like-for-like cohorts before interpreting a change in completion rate. Report the excluded share so readers can see when denominator changes drive the headline.
This classification design is proposed. The voice platform case study describes related call workflows without claiming spam benchmarks. See same-definition before-and-after comparisons.
Spam metrics FAQ
Should spam count as a successful customer call?
No. Report its handling and cost separately from legitimate workflow completion.
Can the model’s spam label be accepted automatically?
Use the approved policy, review uncertain and sampled excluded cases, and monitor false positives.
What if classification is unknown?
Keep an explicit unresolved category rather than silently removing the call from every denominator.