TL;DR: If you only track MRR, you’ll miss the cliff. Here’s the margin scorecard operators actually use.
Daily (stand-up dashboard)
- MTD Revenue / Costs / Gross Margin %
- Top 5 cost spikes by workflow
- Low-margin customers (under 40%)
- Dunning + unpaid invoices
- Alerts: RAG cost surge, TTS anomaly, retry storms
Weekly (ops review)
- Margin by customer segment (SMB, mid-market, enterprise)
- Margin by agent/feature (chat, RAG, outbound, voice)
- Price realization (what % of metered usage got billed)
- Unit economics (cost per email, per doc, per call, per outcome)
- Experiment reads (did last week’s prompt or model switch help?)
Monthly (board pack)
- Cohort margin trends (new vs existing)
- Top loss-makers + action plan (re-price, cap, or cut)
- Pricing experiments (baseline vs new version)
- Vendor mix + savings (model swaps, cache hit-rate)
- Forecast (next 2 quarters with sensitivity)
Operating rules of thumb
- A few customers will eat your GPU. Cap and call.
- If a feature is below 30% gross margin for 2 weeks, fix pricing or kill it.
- Always ship price changes behind a version with clean roll-forward rules.
How UnitPay helps
- One place to see revenue, costs, and margins by agent and customer.
- Anomaly detection on noisy AI workloads.
- Pricing simulation so you change with confidence.
Want our margin pack template (Notion + CSV)? Ping us—free for founders.
