Consumption pricing for AI agents, explained

Consumption pricing for AI agents charges for completed units of work rather than for seats or messages. On Meridian, the unit is a credit: each plan includes a monthly pool of credits, each agent action draws a fixed number of credits, and usage beyond the pool is billed at $0.12 per credit.

This article explains how credits work, what counts as an action, how to estimate a monthly bill with worked examples, and how to keep consumption inside budget.

Why agents are priced on consumption

Seat pricing assumes the software's value scales with the number of people who log in. Agents invert that assumption: the fewer humans who need to touch a process, the better the agent is working. Charging per seat would penalize the outcome the customer is buying. Consumption pricing ties cost to work done, which has two consequences worth understanding before you sign.

First, cost becomes variable, which finance teams must budget for. Second, the vendor's revenue depends on the agent completing work correctly, because a wrong action that gets rolled back is consumed effort with no customer value. That alignment is a feature, but only if the credit definitions are clear.

What a credit is

A credit is a unit of consumption debited when an agent completes a defined action. Credits are not tokens, API calls, or minutes of compute. A single HR case may involve several model calls, tool calls, and a retrieval step; the customer is charged two credits for the resolved case regardless of how many internal steps it took. That is a deliberate design choice. Token-based pricing pushes cost risk onto the customer for variability the customer cannot control.

What counts as an action

As of September 2026, Meridian's published credit rates are:

Three rules govern what is billable:

• Only completed actions consume credits. A case that escalates to a human before resolution is not billed as a resolved case. A candidate screen that fails validation is not billed.
• Actions that a human rejects at an approval step are not billed. If the Contract Review Agent produces a redline and the reviewing lawyer discards it, no credits are consumed.
• Reads are free. Employees asking the Help Desk Agent questions that do not become cases, and finance users exploring plan data through the Planning Agent without generating commentary, do not draw credits.

Plans and included credits

Starter includes 3 agents and 1 workspace. Growth includes all generally available agents, Registry and Gateway, and SSO. Enterprise adds unlimited agents, custom credit pools, Data Fabric, Studio, a dedicated environment, and a 99.95% SLA.

The effective rate matters because the overage rate of $0.12 is 1.2 times Starter's included rate and about 1.45 times Growth's. Consistent overage is a signal to move up a tier, not a penalty to absorb.

Three worked examples

The examples below use the published credit rates and plan pools above; substitute your own volumes.

Worked example 1: a 1,200-employee software company

Assumptions, stated so you can substitute your own:

• HR case volume of 0.8 cases per employee per month, a common range for companies with self-service portals: 960 cases.
• The Help Desk Agent resolves 70% of them: 672 resolved cases at 2 credits each, or 1,344 credits.
• 12 hires per month with 55 screened candidates per hire: 660 screens at 1 credit, or 660 credits.
• 6 audit evidence packages per month averaged across the year, rising during audit season: 30 credits.
• 40 variance commentaries per month for the FP&A team: 120 credits.

Monthly total: 2,154 credits. Starter's 5,000-credit pool covers this with 57% headroom, at $499 per month. The effective cost per resolved HR case, allocating the whole subscription to HR cases alone, is $0.74. Most HR service centers estimate a fully loaded cost per human-handled tier-1 case between $12 and $25.

Worked example 2: an 8,000-employee retailer with frontline staff

• 6,400 HR cases per month; 75% resolved by the agent: 4,800 cases, 9,600 credits.
• 120 hires per month at 50 screens each: 6,000 credits.
• 3,000 open shifts filled per month: 3,000 credits.
• 40 audit packages: 200 credits.
• 300 variance commentaries across store P&Ls: 900 credits.
• 60 supplier contract redlines: 480 credits.

Monthly total: 20,180 credits. Growth's 30,000-credit pool covers this with 33% headroom at $2,499 per month, or $1,999 billed annually. If the retailer opens 1,500 more shifts in December, consumption rises to about 21,700 credits, still inside the pool.

Worked example 3: what overage looks like

Suppose the same retailer acquires a chain and case volume rises 40% for a quarter. Monthly consumption reaches roughly 34,000 credits. The 4,000 credits above the Growth pool are billed at $0.12, adding $480 to the $2,499 subscription, a total of $2,979. Overage is not catastrophic, but if the volume is permanent, an Enterprise credit pool sized to 40,000 credits will cost less per credit than a standing overage.

Budgeting for consumption

Finance teams budget consumption pricing the same way they budget cloud spend: a base commitment plus a forecast band.

1. Baseline the volume drivers: employees, monthly cases, hires, shifts, invoices, contracts. These are numbers your HRIS and ERP already report.
2. Apply expected automation rates conservatively. Use 60% deflection for the Help Desk Agent in budget even if modeled outcomes suggest 75%; revise upward after 90 days of measured data.
3. Add seasonality. Audit season, open enrollment, peak hiring, and year-end close all move consumption by 20% to 50% for a month or two.
4. Set the plan so that the forecast peak month sits below 85% of the included pool.
5. Review actuals monthly against the forecast in Registry, which reports credits by agent and by outcome.

Guardrails against overage

Meridian provides four controls, configurable per workspace, so that consumption does not surprise the budget owner:

• Alerts at 70%, 85%, and 100% of the monthly pool, sent to the workspace owner and the finance contact.
• Hard caps per agent. A cap on the Scheduling Agent, for instance, pauses new shift fills once a credit threshold is reached, with a notification to the owner and an option to raise the cap.
• Approval gates for high-credit actions. Contract redlines at 8 credits can be set to require a reviewer's confirmation before the agent starts.
• Rollover terms. Unused credits on Starter and Growth expire monthly; Enterprise credit pools can be structured as quarterly or annual commitments, which smooths seasonal peaks.

Questions to ask any vendor with consumption pricing

• What exactly is the billable unit, and is a rejected or rolled-back action billed?
• Are reads and retries free?
• What is the overage rate relative to the included rate?
• Can we cap consumption per agent, and what happens at the cap?
• Can we see consumption by agent, by outcome, and by cost center, in real time?

If the answers are precise, the model is safe to budget. If the billable unit is "tokens" or "requests," ask for a worked example using your own volumes before you sign.

Outcome figures in this article are modeled outcomes from design-partner deployments and are not guarantees of results.