For the current Bedrock on-demand pricing (eu-west-3, Apr 2026):
| Component | Call | Tokens (typ.) | $/PO |
|---|---|---|---|
| Stage 0 (Client ID) | Haiku 4.5 × 1 | 1k in / 0.1k out | $0.0013 |
| Stage 1 (Doc Analyzer) | Haiku 4.5 × 1 | 2k in / 0.3k out | $0.0031 |
| Stage 2 (Extractor) | Sonnet 4.6 × 1–2 | 4k in / 2k out | $0.042 |
| Stage 3 (Rules) | local | — | $0 |
| Stage 4 (Snake) | local | — | $0 |
| Stage 4.5 (Pricing) | local | — | $0 |
| Stage 5 (Validation) | Haiku 4.5 × 1 | 3k in / 0.5k out | $0.0044 |
| Stage 6 (Router) | local | — | $0 |
| Total LLM | $0.051 |
A typical single-page ALMA PO with 6 line items costs $0.05 in LLM fees. Multi-page POs (e.g. 13 line items over 2 pages) scale roughly linearly and stay < $0.10.
Stage 4.5 is free (< 50 μs of local lookup) but it closes the pricing gap that would otherwise push the PO to a human reviewer. The leverage is not in the stage's cost, it is in its effect on routing:
| Before 4.5 | After 4.5 | |
|---|---|---|
| Avg unit_price coverage | ~35% of lines | ~85% of lines on seen families |
| Avg trust score (naked-price POs) | 40–60 | 80–95 |
| Share routed to auto-approve | < 20% | ~70% target |
| Average human time per PO | ~4 min | ~30 s (spot check) |
Assume an internal cost of €30/h for Isabel's time. Manual PO entry averages ~4 min → €2.00 per PO. The pipeline runs at $0.05 ≈ €0.046.
At 50 POs/day, that's €98/day in recovered Isabel-time, or ~€24k/year gross, before the fixed infrastructure cost.
| Resource | Monthly |
|---|---|
| EC2 t3.medium (eu-west-3) | $30 |
| EBS 20 GB gp3 | $2 |
| Route53 + CloudWatch + data out | $4 |
| Total | $36/month |
10 POs/day → $15/mo LLM + $36 infra = $51/mo (€48) 50 POs/day → $77/mo LLM + $36 infra = $113/mo (€106) 200 POs/day → $306/mo LLM + $36 infra = $342/mo (€320)
t3.medium handles 3 concurrent workers — ~20 POs/minute sustained on a 6-line single-page average. The bottleneck is Bedrock throughput, not CPU.
The model_mode parameter on /extract trades cost for accuracy:
| mode | stage 0/1/5 | stage 2 | $/PO |
|---|---|---|---|
| cheap | Haiku | Haiku | $0.01 |
| balanced (default) | Haiku | Sonnet | $0.05 |
| accurate | Sonnet | Sonnet | $0.12 |
Cheapest first. Escalate only when Stage 4.5's tariff lookup doesn't clear the bar.