Sell an answerable question
The service should answer where one class of request traveled, which layer initiated paid work, and what evidence supports the conclusion. It should not promise to repair every cache, plugin, provider, and billing defect found along the way. Put the target hostname, incident window, AI feature, and one primary cost question in the order form.
Require access to edge logs, WordPress logs, provider usage, change history, and a technical contact before the clock starts. Missing inputs are not free investigative work; they trigger a pause, a reduced-confidence report, or a separately quoted evidence-recovery task.
Budget the ten-hour production model
Budget three hours for intake and normalization, five for tracing, and two for the evidence map and review. Apply the 20% incident reserve to labor cost, not by quietly adding unbounded hours. The reserve protects margin when logs need modest cleanup; it does not expand scope.
Test whether the $1,200 fee leaves contribution
At a proposed $1,200 fee, calculate loaded labor for each role, sales and payment cost, expected rework, and reserve consumption. For example, ten hours at $75 loaded cost is $750; a 20% labor reserve adds $150, leaving $300 before other overhead. If the actual loaded rate is $100, the same offer has almost no margin.
| Costing line | Planned units | Unit cost | Extended cost |
|---|---|---|---|
| Intake | 3 hours | enter loaded rate | hours × rate |
| Trace | 5 hours | enter loaded rate | hours × rate |
| Report/review | 2 hours | enter loaded rate | hours × rate |
| Incident reserve | 20% of labor | — | labor × 0.20 |
| Fee | 1 diagnostic | $1,200 example | fee minus all costs = contribution |
Make the deliverable auditable
The fixed deliverable is an edge-to-provider evidence map for representative requests, a cause statement with confidence, quantified exposure, and a list of proven and unproven paths. Include UTC timestamps and request IDs; an attractive slide with no lineage creates support liability rather than value.
Provide one decision meeting and one factual correction round. A new date range, second integration, or new root-cause hypothesis is additional scope. State how long evidence and working files are retained and exclude raw prompts unless their content is essential and authorized.
Stop the diagnostic when ownership changes
The diagnostic ends when evidence locates the divergence to a named layer or demonstrates that available records cannot distinguish the candidates. Remediation begins only under a new authorization because editing a CDN rule, plugin, or provider client changes production risk and test requirements.
Escalate security and privacy findings outside the fixed scope
Define escalation triggers in advance: evidence of credential compromise, personal-data exposure, active runaway spend, or a need to access an unsupported third-party system. These events pause the fixed service and move to the client's incident process rather than consuming the remaining hours silently.
Measure profitability after every case
Record planned versus actual hours by phase, missing-input delays, correction requests, conversion to remediation, and whether the conclusion was later supported by the invoice. Review the first five engagements before treating the $1,200 example as a standard price.
A healthy offer repeatedly produces the promised evidence map within the budget and leaves both client and agency clear about the next owner. The Agency page is the relevant next step for multi-site licensing; it should not be presented as if buying a plan substitutes for diagnostic labor.
Use the request-path service costing sheet from “Sell request-path tracing as a bounded agency diagnostic” on a real first installation. Download AI Cost Circuit Breaker for free, begin in Monitoring, and move to enforcement only after the expected signals and rollback are verified.