Resolution, deflection, ROI and pricing

Definitions that cannot be gamed, a cost-per-resolution formula and red flags in vendor benchmarks.

ProductDataDeveloper

Definitions

Metric	Means	Weakness
Containment	Conversation ended without a person	Counts people who gave up
Deflection	Request never reached a human queue	Can rise when the bot is useless
Resolution	Customer's problem actually solved	Needs a verifiable definition
Escalation rate	Share handed to a person	Low is not always good
Repeat contact	Same customer, same issue, again	Best check on false resolutions

Cost per resolution

cost per resolution =
  (platform + model + telephony + people + upkeep + evaluation)
  / verified resolutions

verified = outcome confirmed AND no repeat contact within N days

Compare with the same figure for human-handled cases of the same kind.

Pricing models

Model	You pay for	Watch for
Per resolution	Outcomes	Who decides what counts as resolved
Per conversation or session	Every conversation	Paying when nothing is solved
Per seat or flat	Capacity	Overage and low use

Red flags in benchmarks

The vendor measures its own product
Any reply counts as a resolution
No repeat-contact check
No split of AI-only from handed-off conversations
No method stated