Quarterly Cost Optimization Audit Checklist

Audit Period: Q__ 20__ (__________ to __________)
Auditor: _________________________
Last Audit Date: __________________

---

1. System Prompt Review

• Measured token count for all active system prompts
  Count: ____ prompts audited | Largest: ____ tokens | Average: ____ tokens

• Identified prompts exceeding 500 tokens
  Count: ____ prompts | Combined monthly cost of excess tokens: $____

• Checked for redundant instructions across prompts
  Findings: ________________________________________________________

• Tested compressed versions of top-5 costliest prompts
  Results: ____% average compression | Quality impact: ________

• Removed stale instructions (features deprecated, rules outdated)
  Removed from ____ prompts | Tokens saved: ________

• Verified system prompts are in version control
  ☐ All versioned ☐ ____ prompts not versioned

Estimated savings from prompt optimization: $____/month

---

2. Model Selection Review

• Verified each use case is on the optimal model tier
  Reviewed ____ use cases | ____ candidates for downgrade

• Checked for new cheaper models released since last audit
  New models to evaluate: ________________________________________

• Re-ran quality benchmarks with current test sets
  All use cases above threshold: ☐ Yes ☐ No — exceptions: ________

• Reviewed model pricing changes since last audit
  Price changes: ________________________________________________

• Assessed fine-tuning candidates (high-volume, consistent task)
  Candidates: __________________________________________________

• Checked for deprecated models requiring migration
  Deprecations: ________________________________________________

Estimated savings from model re-selection: $____/month

---

3. Caching Review

• Reviewed cache hit rates for all cached services

  Service	Hit Rate	Target	Status
  		≥40%	☐ OK ☐ Below
  		≥40%	☐ OK ☐ Below

• Identified new cacheable workloads
  Candidates: __________________________________________________

• Reviewed TTL settings (too short = low hit rate, too long = stale)
  Adjustments needed: __________________________________________

• Checked semantic cache threshold accuracy
  Spot-checked ____ semantic matches | Accuracy: ____%

• Validated cache invalidation on prompt version changes
  ☐ Invalidation working correctly ☐ Issues found

• Assessed cache storage costs vs. savings
  Storage cost: $/month | Savings from caching: $/month | Net: $____

Estimated savings from caching improvements: $____/month

---

4. Context Management Review

• Reviewed RAG retrieval settings (top_k, threshold)
  Current top_k: ____ | Recommended: ____ | Context utilization: ____%

• Checked conversation history management
  Max history turns: ____ | Summarization active: ☐ Yes ☐ No

• Measured average context tokens per request by service

  Service	Avg Context Tokens	Change vs Last Quarter

• Identified context bloat (tokens increasing over time)
  Services with >10% context growth: ________________________________

• Reviewed few-shot example counts
  Services using >5 examples: ______ | Tested reducing to 3: ☐ Yes

• Checked for unnecessary data in context (metadata, formatting)
  Findings: ________________________________________________________

Estimated savings from context optimization: $____/month

---

5. Batch Processing Review

• Identified real-time workloads eligible for batch migration
  Candidates: __________________________________________________

• Reviewed batch API utilization and discount realization
  Expected discount: ____% | Realized discount: ____%

• Checked batch job completion times (within SLA)
  Average: ____ hours | SLA: ____ hours | ☐ Within SLA

• Reviewed batch failure rates
  Failure rate: ____% | Target: <1% | ☐ Acceptable

Estimated savings from batch optimization: $____/month

---

6. Budget & Governance Review

• Reviewed budget utilization by team

  Team	Budget	Actual	Utilization
  	$__K	$__K	__%
  	$__K	$__K	__%

• Checked alert effectiveness (false positive rate)
  Total alerts: ____ | Actionable: ____ | False positive rate: ____%

• Reviewed chargeback accuracy
  ☐ All costs attributed ☐ Unattributed costs: $____

• Verified tagging coverage
  ☐ 100% tagged ☐ ____% tagged | Untagged services: ________

• Reviewed cost incident history
  Incidents: ____ | MTTD average: ____ minutes | MTTR average: ____ minutes

• Updated cost forecasts for next quarter
  Q__ forecast: $____K | Growth assumption: ____%

---

Audit Summary

Category	Current Quarterly Cost	Identified Savings	Priority
System Prompts	$__K	$__K/mo	☐ High ☐ Med ☐ Low
Model Selection	$__K	$__K/mo	☐ High ☐ Med ☐ Low
Caching	$__K	$__K/mo	☐ High ☐ Med ☐ Low
Context Management	$__K	$__K/mo	☐ High ☐ Med ☐ Low
Batch Processing	$__K	$__K/mo	☐ High ☐ Med ☐ Low
Total	$__K	$__K/mo

Action Items

#	Action	Owner	Deadline	Expected Impact
1				$__K/mo
2				$__K/mo
3				$__K/mo

---

Template from the TokenOps Atlas — tokenops-atlas