Pricing Models for Agentic SRE Platforms: What to Compare

Per-incident. Per-action. Per-host. Per-token. Each pricing model rewards different behaviour. How to model the spend you actually expect under each.

The four models

Four pricing models dominate agentic-SRE platforms. Per-incident scales with how many incidents you have (punishes growth); per-action scales with what the agent does (punishes broader scope); per-host or per-service scales with infrastructure size (largely independent of agent activity); per-token scales with model usage (closest to underlying compute cost).

• Per-incident. Cost scales with incident count; punishes growth.
• Per-action. Cost scales with agent activity; punishes broader scope.
• Per-host or per-service. Cost scales with fleet size; independent of agent activity.
• Per-token. Cost scales with model usage; closest to compute cost.

What each model rewards

Each model rewards different vendor behaviour. Per-incident rewards small incident counts (vendor interest aligns with reliability); per-action rewards passive agents (vendor interest is to avoid action, misaligned with making things better); per-service rewards small fleets (vendor independent of agent quality); per-token rewards efficient prompts (vendor interest aligns with cost discipline).

• Per-incident: small counts. Vendor aligns with reliability; interests aligned.
• Per-action: passive agents. Vendor avoids action; interests misaligned.
• Per-service: small fleets. Vendor independent of agent quality.
• Per-token: efficient prompts. Vendor aligns with cost discipline.

Normalise to $/incident

Comparing apples to apples requires normalisation. All four models can be expressed as cost-per-incident with assumed volumes: per-incident is straight, per-action is average actions per incident times per-action price, per-service is monthly cost divided by monthly incident count for that service, per-token is average tokens per incident times per-token price. Vendors are defensive about the conversion; that is itself a signal.

• Per-incident: the price. Direct; no conversion needed.
• Per-action: avg actions × price. The conversion to incident-cost.
• Per-service: monthly cost / incidents. Per-service slice; supports apples-to-apples.
• Per-token: avg tokens × price. The conversion via token estimate.

Hidden costs

Four cost categories deserve scrutiny. Setup fees (one-time integration charges, often $20-100k for enterprise platforms); premium support (if you need fast response, expect to pay); egress fees (pulling data out at contract end, usually disclosed); overage fees (what happens if you exceed committed volume, some vendors reasonable, some punitive).

• Setup fees. One-time integration; $20-100k for enterprise platforms.
• Premium support. Fast response costs extra; budget for it.
• Egress fees. Data extraction at contract end; usually disclosed in fine print.
• Overage fees. Above committed volume; some vendors reasonable, some punitive.

What to negotiate

Four levers matter at contract time. Trial period (90 days at reduced commitment validates the platform on real workload); caps (per-incident or per-action with a price ceiling protect against runaway costs); data ownership (your data stays yours, egress on demand without fees); termination (30-day notice, data handed over, no penalties).

• Trial period. 90 days reduced commitment; validates on real workload.
• Caps. Per-incident or per-action ceiling; protects against runaway.
• Data ownership. Your data stays yours; egress on demand without fees.
• Termination terms. 30-day notice; data handed over; no penalties.