Decode per-resolution pricing, audit your knowledge base, and run a real 200-question bake-off before you sign an AI customer support contract.
Every vendor in this category quotes a resolution percentage, and almost none of them mean the same thing by it. Deflection, containment, and resolution are three different measurements, and the gap between them is where buyers lose money — first on an oversized contract, then on the customers who were technically resolved and are quietly annoyed. This workbook replaces the vendor's benchmark with your own measured number, produced from your tickets and graded by your team. It opens with a ticket mix audit that sorts 200 of your real conversations into four buckets — answerable from documentation, answerable from your data, requiring an action with consequences, and requiring human judgment — because those proportions predict your ceiling far better than any demo. Then the outcome-pricing decoder: what counts as a resolution at each vendor, whether a handoff bills, what the implicit-resolution window is, and worked annual math at published September 2026 rates across 2,000, 10,000, and 30,000 tickets a month. It includes the fact most buyers miss until the second invoice — success costs more, because a resolution rate that climbs from 55% to 75% raises your bill by more than a third. The rest is the process. A knowledge readiness audit scored out of 30, with a two-week rewrite protocol for teams who score low, because the strongest predictor of an agent's performance is whether a correct, current, unambiguous answer exists for it to find. A weighted scorecard that puts action capability and grounding above answer fluency. A full bake-off protocol: build a frozen 200-question gold set including 20 hard ones and 10 deliberately unanswerable ones, run two vendors on identical sources with identical setup time, grade blind on correctness, grounding, action, and tone, and count hallucinations separately rather than folding them into an average. Then a risk-sequenced rollout, a weekly improvement loop, nine metrics including repeat-contact rate, a security and procurement checklist covering subprocessors and data-training terms, and five honest reasons not to buy at all.
Every one of these is free and email-gated, same as this one.
How to choose an AI-assisted accounts payable, spend, and close stack: category map, pricing decoder, scorecard, six-document torture test, fraud controls, and a six-week migration plan.
PDFA vendor-selection workbook for AI phone agents: pricing decoder, 40-point scorecard, demo-call script, 14 adversarial test calls, contract checklist, and a 30-day pilot plan.
PDFCapture every lead, answer every call, and keep past clients buying and referring on autopilot — without hiring a full-time ISA.
PDF