APPNEURAL - AI PILOT EVALUATION
Editable planning worksheet
Source: https://appneural.com/resources/templates

Define the test and go/no-go criteria before running a pilot. Keep expected
results, measured results and assumptions separate.

1. USE CASE AND OWNERS
Pilot name and version: [fill in]
Business owner: [fill in]
Evaluation owner: [fill in]
Review date: [fill in]
User and task being supported: [fill in]
Current workflow and performance baseline: [fill in]
Bounded AI task: [fill in]
Actions or decisions outside the pilot's authority: [fill in]
Human review or escalation points: [fill in]
Desired improvement and minimum acceptable result: [fill in]

2. TEST SET
Approved data sources and access owner: [fill in]
Test-set version, sample size and selection method: [fill in]
Normal cases to include: [fill in]
Edge cases, unsupported requests and failure cases: [fill in]
Representative user roles and access boundaries: [fill in]
Expected answer or outcome for each case: [fill in]
Who reviews the expected outcomes? [fill in]
Which cases are held out from development? [fill in]
Data retention and permitted processing arrangements: [fill in]

Case ID | Input/reference | Expected result | Pass/fail rule
[fill in] | [fill in] | [fill in] | [fill in]

3. EVALUATION RUBRIC
Specify thresholds before reviewing results. Use only criteria relevant to the task.
Criterion | Measurement method | Minimum threshold | Reviewer
Task completion | [fill in] | [fill in] | [fill in]
Accuracy or usefulness | [fill in] | [fill in] | [fill in]
Source support, if applicable | [fill in] | [fill in] | [fill in]
Access-boundary behavior | [fill in] | [fill in] | [fill in]
Unsupported-task handling | [fill in] | [fill in] | [fill in]
Latency | [fill in] | [fill in] | [fill in]
Human review effort | [fill in] | [fill in] | [fill in]

4. REPRODUCIBLE RUN
Application/model/provider version: [fill in]
Prompt, retrieval and configuration versions: [fill in]
Run date and environment: [fill in]
Run ID and result-log location: [fill in]
Repeat runs or variability checks planned: [fill in]
Baseline comparison method: [fill in]

5. COST ASSUMPTIONS
Expected tasks per day/month: [fill in]
Typical and maximum input/output size: [fill in]
Unit usage rates, currency and price-check date: [fill in]
Retrieval, storage and infrastructure costs: [fill in]
Retries, failures and human-review cost assumptions: [fill in]
Measured cost per completed task: [fill in]
Budget ceiling and owner of usage monitoring: [fill in]

6. RESULTS AND LIMITATIONS
Case ID | Observed result | Passed? | Reviewer | Evidence reference
[fill in] | [fill in] | [yes/no] | [fill in] | [fill in]

Criteria that met the agreed threshold: [fill in]
Failures, recurring patterns and unresolved risks: [fill in]
Conditions or populations not tested: [fill in]
Changes required before a larger rollout: [fill in]

7. GO/NO-GO REVIEW
Decision: [proceed within defined scope / revise and retest / stop]
Evidence supporting the decision: [fill in]
Residual risks accepted, by whom and for which scope: [fill in]
Operational, escalation and monitoring owners: [fill in]
Next review date and rollback/stop conditions: [fill in]

This worksheet is a planning aid. A pilot result applies to its tested task,
data and configuration; record the limits before using it for a wider rollout.