DAHSHAN LABS — AI PILOT BRIEF
Template version: 2026-10-20
Guide: https://dahshanlabs.com/guides/business-ai-assistant#pilot-resources
Arabic version: https://dahshanlabs.com/resources/ai-pilot-brief.ar.txt

Copy and complete this template for your project. Blank fields are intentional.
This is a planning resource, not a customer case study, quote or delivery commitment.
Use redacted examples. Keep credentials and confidential records out of public copies
and contact forms; agree a secure route before sharing project data.

1. ONE WORKFLOW
Project / organization:
Business owner / technical owner:
Users and roles:
Task to improve:
Current steps and tools:
One knowledge source:
One system to connect:
Explicitly out of scope:

2. BASELINE AND BUSINESS HYPOTHESIS
Current task volume and observation period:
Baseline sample size and how cases were selected:
Median / p95 completion time, including review and correction:
Error definition and observed errors / total cases:
Proposed improvement and how it will be measured:
Costs included in the comparison:
Baseline evidence location and owner:
Keep the baseline and pilot task mix comparable. Record denominators, exclusions
and limitations. A hypothetical target is not a measured result.

3. DATA, IDENTITY AND ACTIONS
Approved source IDs, revisions and freshness rules:
Data owner and permitted use:
Roles allowed to read each source / record:
Identity provider and tenant boundary:
Required languages, terminology and dialects (if any):
Read-only operations:
Proposed write operation (if required):
Who may approve the exact proposed change:
Denied operations and escalation route:
Model, retrieval, connector and log hosting / data locations:
External data recipients and retention / deletion requirements:
Sandbox access and credential owner (do not put the credential here):

4. EVALUATION BEFORE DEVELOPMENT
Evaluation owner and reviewers for Arabic / English:
Test-set version, source revisions and cases per language:
Holdout set kept separate from tuning:
Supported-answer and citation criteria:
Expected handling of missing, stale and conflicting evidence:
Denied-access, prompt-injection and approval-bypass tests:
Retry, duplicate-write and changed-record tests:
Quality thresholds by language, with numerators / denominators:
Critical failures that stop rollout:
Latency / cost limits and how they are measured:
Required operational evidence:
Decision rules agreed on (date / approvers):
Worksheet: https://dahshanlabs.com/resources/ai-pilot-evaluation.json
Replace example scenarios with your approved fixtures. Its measurement fields
are empty; record results only after running and reviewing the actual system.

5. DELIVERY AND OPERATIONS
Reviewed data-flow design:
Sandbox connector and access checks:
Review interface and audit / receipt requirements:
Evaluation report, remaining failures and test artifacts:
Source-code, component-license and deployment ownership:
Training, runbook and support responsibilities:
Monitoring, incident owner, rollback and disable procedure:
Milestones and budget to agree:
Production access remains a separate decision after the evidence is reviewed.

6. DECISION RECORD — COMPLETE AFTER THE PILOT
Run dates and system / model / configuration versions:
Results by language and scenario, including failed / excluded cases:
Comparison with the documented baseline:
Reviewer and review date:
Decision: [ ] proceed  [ ] revise  [ ] stop
Reasons, unresolved risks, owner and next review:

7. OPTIONAL PUBLIC CASE STUDY — SEPARATE PERMISSION
Customer approved for public naming:
Problem and delivered scope that can be published:
Measured outcome, observation period and sample size:
Evidence owner and method:
Limitations and attribution:
Approved quote / logo (only if separately authorized):
Written publication approval and date:
Do not fill these fields with projected benefits, invented customers or demo results.

Prepared by Dahshan Labs, Riyadh, Saudi Arabia.
Discuss a redacted brief: https://dahshanlabs.com/contact