AI AGENT PILOT TEMPLATE
ShowMeYourSaaS — October 26, 2026
Guide: https://showmeyoursaas.com/articles/ai-agents-for-business/

TASK
Audience and recurring job:
Current process and known failure points:
Smallest acceptable output:
Why a fixed workflow or one model call is insufficient (if it is):
Owner and pilot window:

INPUTS AND ACCESS
Authorized data sources and scope:
Synthetic or permitted evaluation records:
Tools: name / read or write / exact permitted action / owner
Actions requiring human approval:
Actions excluded from the pilot:
How retrieved content is kept from granting new authority:
Logging, redaction, and retention decisions:

ORIGINAL FICTIONAL SUPPORT-TRIAGE PILOT
Goal: prepare an accurate draft for a calendar-connection question.
Inputs: an authorized ticket and the approved help collection.
Allowed: search help content and produce a draft with sources.
Required: state missing facts; escalate when the sources do not answer.
Excluded: sending, changing accounts, disconnecting calendars, issuing refunds.
Success: reviewer confirms relevance, source support, correct escalation, and
absence of unauthorized action. This is a design example, not a measured rollout.

ENFORCED EXECUTION LIMITS
Maximum tool calls:
Maximum time:
Maximum retries:
Maximum spend:
Idempotency for any repeat-sensitive action:
Failure and escalation destination:
Stop/disable owner and mechanism:

EVALUATION CASES
Ordinary request with an answer in the approved sources:
Ambiguous request needing clarification:
Missing or contradictory source material:
Outdated instructions:
Text attempting to redirect the task or expand permissions:
Tool timeout, repeated response, or unavailable service:

Per case: expected result / actual result / source support / permission outcome /
review minutes / model and tool usage / failure reason / reviewer decision.
Correct escalation is an outcome to assess, not automatically a failure.

COST LOG
Completed acceptable jobs:
Attempts and retries:
Model usage or platform-billed activity:
Paid tool calls:
Review and correction minutes:
Setup and maintenance time:
Direct cost / acceptable completed job:
Total operating effort / acceptable completed job:

FICTIONAL BUDGET — ALL RATES INVENTED, NOT VENDOR PRICES
100 requests × 3 model calls × $0.01 = $3
100 requests × 2 lookups × $0.005 = $1
Direct usage cost = $4
100 requests × 2 review minutes / 60 × $30/hour = $100 review effort
Setup, maintenance, hosting, and other costs are additional.

DECISION
Comparison with existing process on a comparable request mix:
Unresolved quality or access concerns:
Continue, revise, or stop:
Next permission proposed and evidence required before enabling it:
Owner and next review date: