Subscribe
ai-sre
6 min read

The Investigation Is Not the Unit: What AI SRE Agents Actually Bill For

Investigation is not the unit. AI SRE vendors use disparate pricing models including per-seat, proprietary credits, tokens, and platform plans, with only a few vendors offering transparent consumption metrics.

Chart comparing AI SRE vendor pricing models including Datadog credits, Cleric per-task costs, and PagerDuty platform tiers
Chart comparing AI SRE vendor pricing models including Datadog credits, Cleric per-task costs, and PagerDuty platform tiers

AI SRE vendors utilize diverse billing units rather than a standard investigation fee, primarily using per-seat licenses, proprietary credit systems, raw model tokens, or custom platform quotes. Only two vendors reviewed, Datadog and Cleric, provide explicit public pricing for autonomous incident response before engaging sales teams.

Every incident-response vendor now demos the same thing: an alert fires, an AI agent reads the telemetry, correlates it with recent changes, and posts a probable root cause before a human finishes coffee. The demos have converged. The invoices have not. Of the seven vendors with public pricing reviewed for this piece, none bills per investigation, and four bill in units that exist only on their own price lists.

This is what the AI SRE pricing shelf actually looks like in late 2026, meter by meter.

Four meters, one product

Per-seat, AI bundled or bolted on. The incident-management incumbents still price humans. incident.io charges $15 per user per month (annual) on Team and $25 on Pro, with on-call adding $10 to $20 per user. Its AI investigation features ("Investigations," "Root cause analysis," "Autonomous agent") appear in the plan comparison as "available as add-on" with no published add-on price. Rootly lists Incident Response at $20 per user per month and On-Call at $20, but prices its AI SRE product only as "Let's talk," with the fine print that AI SRE requires an Incident Response subscription. FireHydrant charges $25 per responder per month on Pro, and its AI features (summaries, transcripts, triage, retros) appear only in the Enterprise column, priced custom.

PagerDuty has moved its flagship "PD Reliability Platform" off per-seat pricing entirely onto annual platform plans: Starter at $2,800 per year (5 seats, plus 5 bonus seats for a limited time), Essential at $12,000, Plus at $48,000, Ultimate by quote. Each plan bundles two usage pools: Signal Intelligence Events (5,000 at Starter up to 1.5 million at Ultimate) and AI Actions (1,000 to 160,000). The AI agents, including Paige, its SRE agent, are included from Essential up and sold as add-ons on the older $21 per user Professional plan. The page does not publish how many AI Actions one investigation consumes.

Proprietary credits. Datadog sells "AI Credits" from $500 per 500 credits per month on annual billing, or $1.30 per credit on demand. The pricing page estimates consumption per task: roughly 4 credits for a Bits Investigation autonomous investigation (about $4 annual, $5.20 on demand), 0.5 for a Bits Chat message, 5 for a Bits Code fix, 1 for an Agent Builder run. Credits reset monthly with no rollover, and overages bill automatically at the on-demand rate, a 30% premium. This is itself a recent change: Bits AI SRE was renamed Bits Investigation and the old per-investigation charge was replaced by credits shared across the Bits agents, as Dash0's comparison notes. Human seats are separate: On-Call $20 per seat per month, Incident Management $30, the Incident Response bundle $40, billed annually.

Cleric also runs a credit meter but publishes the arithmetic: $0.50 per credit on every plan, 10 credits ($5) per production issue investigation, 2 credits ($1) per change verification, 1 credit per started minute of chat after the first 30 seconds free. Plans are credit floors, not seat limits: Team $700 per month for 1,400 credits, Pro $1,500 for 3,000, Scale $3,000 for 6,000. Unused credits roll over for one month, up to twice the monthly allocation.

Tokens. Grafana bills its Investigations product in raw model tokens, with no per-seat or per-investigation fee: manual investigations draw from a 40 million token per active AI user allowance, and alert-triggered investigations draw from a 25 million token monthly system pool, then pay as you go. The public pricing page links a calculator but shows no per-token rate, so the meter is only half published.

Quote. The pure-play AI SRE startups mostly decline to publish. Traversal has no pricing page as of this writing (the URL returns 404). Resolve.ai and Komodor come up quote-only in every third-party roundup checked.

Same workload, seven bills

Take a modest case: a 10-engineer team that triggers 50 autonomous investigations per month, on annual list prices.

  • Datadog: Incident Response bundle at $400 per month, plus 50 investigations at the estimated 4 credits each, 200 credits, which the $500 credit pack covers. Total about $900 per month, of which $500 is the AI meter.
  • Cleric: 50 investigations at $5 is $250 of credits, but the smallest plan that grants access is Team at $700 per month. Usage is 36% of the bill; the floor is the rest.
  • PagerDuty: Starter at $2,800 per year, about $233 per month, includes 1,000 AI Actions. Whether 50 investigations fit in 1,000 actions is unknowable from public data, and Starter's 5-plus-5 seats leave no room for a wider response roster. Essential runs $12,000 per year.
  • incident.io: Pro plus on-call is $45 per user, $450 per month, plus an unpublished investigations add-on.
  • Rootly: Incident Response plus On-Call is $40 per user, $400 per month, plus AI SRE by quote.
  • FireHydrant: $250 per month for 10 responders, but AI features are Enterprise-only, so the real AI price is not on the menu.
  • Grafana: token meters are published in tokens, not dollars; the bill cannot be computed without contacting sales or running the calculator.

The one honest conclusion: only two vendors (Datadog and Cleric) let you price a month of agentic incident response before a sales call, and even Datadog's "~4 credits" is its own average, with consumption "varying with task complexity" per its pricing page.

Five mismatched measuring instruments, a gauge, an hourglass, a balance scale, a spring scale and a ruler, all wired to one alarm bell

What the meters are hiding

Three structural problems recur across the shelf.

Unit mismatch. A Datadog credit, a Cleric credit, a PagerDuty AI Action and a Grafana token measure different work. Comparing "$1 per credit" across vendors compares nothing. Cleric is the only vendor that fixes its credit to a named task and price ("$0.50 per credit on every plan", "$5 per Issue investigation"), which is also why its math survives a procurement meeting.

Floors and cliffs. Usage meters come with minimums that dwarf expected usage. Cleric's cheapest plan is 2.8x the credits a 50-investigation month consumes. Datadog's pack is 500 credits when the same month needs about 200, and unused credits do not roll over. Alert storms run the other way: overages auto-bill at on-demand rates. Seat pricing has the mirror problem: cheap per engineer, but every stakeholder you add to the incident channel is another $15 to $45 per month, which pushes teams toward "viewers," the license class FireHydrant reserves for Enterprise.

Timing. The per-seat vendors are still selling response software with AI attached and pricing the AI last (or never, as in the four "contact us" boxes above). The credit vendors repriced mid-flight: Datadog abandoned per-investigation billing for a shared pool within roughly a year of the product's GA. Budgeting for a meter that the vendor has already renamed once is speculation, not planning.

What to ask before you sign

  1. How many credits, actions or tokens does one median investigation consume, at p50 and p95, on your incident mix? Only Cleric answers this unprompted; the others define the meter but not the work.
  2. What happens in an alert storm? Get the overage rate (Datadog: 30% premium, automatic) and the rollover rule (Cleric: one month, 2x cap; Datadog: none).
  3. What is free when the agent is wrong? Confirm in writing whether inconclusive or low-confidence investigations bill.
  4. Who else has to be licensed for the agent to run? If the AI only works where humans sit, seat math applies whether you bought seats or not.
  5. If the AI price is "Let's talk," ask for the meter definition in the SOW, not the discount. A quote without a consumption table is an unpriced product.

The demos are converged. The invoices are not. Until the units standardize, the buyer's job is converting every vendor's proprietary meter back to dollars-per-resolved-incident, and treating any vendor who will not publish that ratio as selling an unpriced product.

Sources

Frequently asked questions

How does Datadog bill for AI Investigations?

Datadog bills using AI Credits estimated at roughly 4 credits per autonomous investigation, with overages billed automatically at a 30% premium.

What is the cost of an investigation on Cleric?

Cleric charges $5 per production issue investigation, calculated as 10 credits at a rate of $0.50 per credit.

Does PagerDuty publish a per-investigation price?

No, PagerDuty bundles AI Actions into annual platform plans and does not publish how many AI Actions one investigation consumes.

How are unused credits handled by different vendors?

Unused credits roll over for one month up to twice the monthly allocation on Cleric, whereas Datadog's credits reset monthly with no rollover.

ai-sreToolingPricing