Monthly TokenOps Cost Review — Meeting Template
Duration: 60 minutes · Cadence: First Wednesday of each month
Owner: [FinOps / AI Platform Lead]
Attendees: Engineering leads, Product owners, Finance partner, AI/ML platform team
Pre-Meeting Prep (Complete 2 Days Before)
The meeting owner gathers the following data and distributes the pre-read deck 48 hours before the session.
Data to Pull
| # | Data Source | Responsible | Notes |
|---|---|---|---|
| 1 | Total token spend (current month vs. prior month vs. budget) | FinOps Lead | Break down by provider |
| 2 | Spend by team / service / feature (top-10 view) | FinOps Lead | Flag any service >15% over budget |
| 3 | Cost per request by service (P50, P90, P99) | Platform Eng | Identify outlier requests |
| 4 | Model mix breakdown (% spend by model) | Platform Eng | Track migration from expensive → cheaper models |
| 5 | Token waste metrics (cache-hit rate, retry tokens, empty responses) | Platform Eng | Target ≥30% cache-hit rate |
| 6 | Optimization initiative tracker (status, savings realized) | All teams | Update savings projections |
| 7 | Upcoming launches / feature changes that impact token spend | Product Leads | Forward-looking demand signals |
| 8 | Budget vs. actuals YTD waterfall | Finance Partner | Include committed vs. on-demand split |
Pre-Read Deliverables
- Executive summary slide (1 page) with RAG status for each budget line
- Cost trend chart (rolling 6 months)
- Top-5 optimization opportunities ranked by estimated savings
- Anomaly report: any service with >2× daily spend spike in the past 30 days
Agenda
1 · Metrics Review (15 min)
Walk through the headline numbers. The goal is shared awareness, not deep-dives.
Key Metrics Table (Example)
| Metric | Prior Month | Current Month | Δ (%) | Budget | Status |
|---|---|---|---|---|---|
| Total Token Spend | $48,200 | $52,750 | +9.4% | $55,000 | 🟡 |
| Total Tokens Consumed | 1.24 B | 1.41 B | +13.7% | — | — |
| Avg. Cost / 1K Requests | $3.82 | $3.54 | −7.3% | — | 🟢 |
| Cache-Hit Rate | 24% | 31% | +7 pp | 30% | 🟢 |
| Retry Token Waste | 4.1% | 3.8% | −0.3 pp | <5% | 🟢 |
| GPT-4o Share of Spend | 62% | 55% | −7 pp | — | 🟢 |
| Services Over Budget | 1 | 2 | — | 0 | 🔴 |
Discussion Prompts
- Are we on track to finish the month within budget? If not, what is the projected overage?
- Which services drove the largest absolute increase in spend?
- Did any model-pricing changes from providers affect our costs this month?
- Is the traffic growth organic (more users) or systemic (longer prompts, more retries)?
2 · Optimization Progress (15 min)
Review active optimization initiatives and their measured impact.
Initiative Tracker (Example)
| # | Initiative | Owner | Status | Monthly Savings (Est.) | Monthly Savings (Actual) |
|---|---|---|---|---|---|
| 1 | Migrate classification to gpt-4o-mini |
@eng-lead | ✅ Done | $4,200 | $4,850 |
| 2 | Prompt compression — support chatbot | @ml-eng | 🔄 In Progress | $2,800 | — |
| 3 | Semantic cache for FAQ answers | @platform | 🔄 In Progress | $3,500 | — |
| 4 | Batch window for data pipeline | @data-eng | 📋 Planned | $1,600 | — |
| 5 | Output token cap on email generator | @growth | ✅ Done | $900 | $1,050 |
Discussion Prompts
- Which completed initiative delivered the most value? Can we replicate the pattern elsewhere?
- Are any in-progress items blocked? What do they need to unblock?
- Should we re-prioritize the backlog based on this month's spend data?
- Are there new optimization opportunities we should add to the tracker?
3 · Budget Forecast (10 min)
Project forward based on current run-rate, known launches, and seasonal patterns.
Forecast Table (Example)
| Month | Projected Spend | Budget | Variance | Key Assumption |
|---|---|---|---|---|
| Jun 2026 | $56,400 | $55,000 | +$1,400 | New feature launch adds ~8% traffic |
| Jul 2026 | $54,100 | $55,000 | −$900 | Cache rollout offsets growth |
| Aug 2026 | $58,200 | $57,000 | +$1,200 | Seasonal traffic peak |
| Q3 Total | $168,700 | $167,000 | +$1,700 | — |
Discussion Prompts
- Do we need to request a budget adjustment for next quarter?
- Are there committed-use or reserved-capacity deals we should evaluate?
- What is the impact if the new feature launch drives 2× expected traffic?
- Should we set up an automatic scale-down for non-critical services during peak cost?
4 · Priorities for Next Month (15 min)
Agree on the top 3–5 priorities for the coming month. Each must have an owner and a measurable target.
Priority Template
| Priority | Owner | Target | Due Date |
|---|---|---|---|
| 1. Deploy semantic cache to production | @platform | ≥30% cache hit on support chatbot | Jun 15 |
| 2. Complete prompt compression for chatbot | @ml-eng | ≥20% token reduction, <1% quality loss | Jun 20 |
| 3. Negotiate batch-pricing tier with OpenAI | @finops | Secure ≥15% discount on batch calls | Jun 30 |
| 4. Build cost-per-conversation dashboard | @analytics | Dashboard live in Looker | Jun 10 |
| 5. Onboard data-eng team to tagging standard | @platform | 100% of data pipeline calls tagged | Jun 25 |
Discussion Prompts
- Do these priorities reflect our biggest cost-reduction levers?
- Are owners resourced to hit these targets, or do we need to negotiate capacity?
- What is the "do nothing" cost if we defer any of these?
5 · Action Items
Capture every commitment made during the meeting.
| # | Action Item | Owner | Due Date | Status |
|---|---|---|---|---|
| 1 | e.g., Share updated forecast model with Finance | @finops | Jun 5 | ⬜ Open |
| 2 | ⬜ Open | |||
| 3 | ⬜ Open | |||
| 4 | ⬜ Open | |||
| 5 | ⬜ Open |
Appendix: Meeting Hygiene
| Guideline | Detail |
|---|---|
| Pre-read | Distributed 48 hours before; attendees arrive having reviewed |
| Decision log | Decisions captured inline and posted to #tokenops-decisions |
| Parking lot | Off-topic items logged for async follow-up |
| Recording | Meeting recorded and transcript shared within 24 hours |
| Rotation | Note-taking rotates alphabetically each month |
| Escalation | Any budget breach >10% escalated to VP Engineering within 24 hours |
Template version 1.0 — Maintained by the TokenOps team.