RAG ROI Calculator

The calculator models steady-state monthly cost for a production RAG system using 2026 list pricing. Four cost components:

• Vector database — base operational cost + per-million-vectors. Models pgvector (Postgres extension, AWS RDS), Pinecone Serverless, Weaviate Cloud, Qdrant (self-hosted GPU). Real costs vary with utilization and reserved-capacity discounts.
• LLM synthesis — input tokens (retrieved chunks + prompt, ~3,500/query) + output tokens (~300/query). Provider list pricing as of 2026.
• Embeddings — text-embedding-3-small at $0.02/1M tokens, assuming 10% of corpus refreshes monthly (typical for active document sets).
• Infrastructure overhead — observability, retries, supporting services. Modeled at +20% on the base cost.

FTE savings calculation — multiplies the FTE count you input by the annual loaded cost (salary + benefits + tooling, typically 1.3× base salary). Divided by 12 for monthly. Compared against monthly RAG cost to get net savings.

Payback period — your one-time build/ramp-up cost divided by monthly net savings. If RAG costs more than the FTE it displaces, payback is N/A (the system needs more scale to justify itself).

What it doesn't do: capture spikey workloads, multi-region replication costs, fine-tuning costs (separate from RAG), reranker model costs (we assume single-stage retrieval), or the cost of getting the RAG quality high enough to displace humans (which is the bigger gating factor than infrastructure cost). Real production bills are typically 15-25% higher than the calculator's projection.

For a real cost + quality audit on a RAG system you're already running — including the architecture-level optimizations the calculator can't see — book the call below.