Anthrocentrix QA Yield Engine
Same review budget. More bad labels caught. Fewer escaped defects. Benchmark against your current QA policy — no integration required.
Without customer data, all baselines are benchmark baselines only, not the customer's true current policy.
Review budget and reviewer headcount are held constant across strategies. Anthrocentrix re-orders the queue; it does not shrink it.
Primary KPI — QA yield at 10.0% review budget vs Customer Current Policy (proxy)
Baseline comparison @ 10.0% review budget
All strategies route the same number of items to human review. Yield differences come entirely from prioritization.
| Strategy | Errors / 1k reviews | Errors caught / yr | Escaped errors / yr | Defects / hour | Cost / error |
|---|---|---|---|---|---|
| Customer Current Policy (proxy) | 177.1 | 425.0k | 1.97M | 15.94 | $2 |
| Random Sampling | 100.0 | 240.0k | 2.16M | 9.00 | $4 |
| Rule-Based Escalation | 164.0 | 393.6k | 2.01M | 14.76 | $2 |
| Inter-Annotator Disagreement | 141.7 | 340.0k | 2.06M | 12.75 | $3 |
| Consensus-Based Review | 185.7 | 445.7k | 1.95M | 16.71 | $2 |
| Annotator Scorecard Sampling | 211.1 | 506.5k | 1.89M | 19.00 | $2 |
| Anthrocentrix Risk Routing | 290.0 | 695.9k | 1.70M | 26.10 | $1 |
Frontier labs and large labeling platforms (Scale AI, Labelbox, Surge) are framed as strategic future partners, not ideal first customers. Initial GTM targets the segment list above.
Yield vs review budget — error recall by strategy
| Budget | Random | Rule | Disagreement | Consensus | Scorecard | Current policy | Anthrocentrix |
|---|---|---|---|---|---|---|---|
| 1% | 1.0% | 1.7% | 1.4% | 1.9% | 2.2% | 1.8% | 3.2% |
| 2% | 2.0% | 3.4% | 2.9% | 3.9% | 4.4% | 3.7% | 6.4% |
| 5% | 5.0% | 8.4% | 7.2% | 9.5% | 10.9% | 9.1% | 15.4% |
| 10% | 10.0% | 16.4% | 14.2% | 18.6% | 21.1% | 17.7% | 29.0% |
| 20% | 20.0% | 31.6% | 27.6% | 35.3% | 39.5% | 33.8% | 51.6% |
| 30% | 30.0% | 45.5% | 40.4% | 50.1% | 55.2% | 48.3% | 68.6% |
Sensitivity — yield lift at varying multiples of benchmark performance
If Anthrocentrix delivers only a fraction of the ImageNet-AB-calibrated lift, what yield uplift remains vs Customer Current Policy (proxy)?
Zero-Integration Retrospective Backtest
- Annotation events
- Annotator IDs & task IDs
- Timestamps
- Behavioral telemetry (if available)
- Final labels
- Gold labels or adjudicated QA results
- Current QA routing decisions (if available)
- Current QA baseline (from customer data)
- Anthrocentrix risk-ranked QA queue
- Incremental errors caught at fixed review budget
- Defect escape reduction
- Confidence intervals on yield lift
- ROI sensitivity bands
Pricing options (default models)
| Model | Basis | Annualized at current segment |
|---|---|---|
| Flat monthly platform fee | $6k / mo platform fee (typical mid-market entry) | $72.0k |
| Per 100k annotations scored | $18 per 100k annotations scored | $4.3k |
| Retrospective audit fee (one-time) | $35k one-time backtest engagement | $35.0k |
| Enterprise annual license | $180k / yr enterprise license, unlimited volume | $180.0k |
| Value-based (% of modeled downstream value) — optional | 15% of modeled downstream value (optional, not default) | $65.8k |
Value-based (percent-of-savings) pricing is optional, not the default. Default sales motion uses flat platform fees or per-volume pricing to avoid attribution disputes.
/mnt/documents: Buyer Deck, CFO One-Pager, Investor Summary, Sensitivity Report, Pilot Readiness Memo.