Relevance AI Pricing: Actions, Credits, and Outbound Cost

Relevance AI pricing is a three-part calculation: the subscription, Actions for tool runs, and Vendor Credits for models or paid tools. A Pro or Team price alone does not show what an outbound workflow will cost. Research, enrichment, approvals, CRM reads, retries, and failed tool calls can all change the bill. Buyers should price a representative workflow per accepted account before choosing a plan.

Relevance AI Pricing at a Glance

We reviewed Relevance AI's current pricing, plan, spend-control, and workforce documentation on October 25, 2026. The marketing pricing page and the detailed pricing documentation do not present the same buying path, so both belong in the decision.

Our opinion is that Relevance AI can be economical for a narrow, supervised process, but the plan card is the least useful unit for comparing it with another tool. The defensible unit is total monthly cost divided by accounts that pass review and reach the next controlled step.

Current Plans Have Two Public Surfaces

The detailed Relevance AI pricing documentation lists Pro, Team, and Enterprise. Pro is $29 month to month or $19 per month on annual billing. It includes 2,500 Actions each month and $20 in Vendor Credits. Team is $349 month to month or $234 per month on annual billing, with 7,000 Actions and $70 in Vendor Credits. Enterprise is quote-led.

The public marketing pricing page currently displays Enterprise and a sales conversation. It does not show the Pro or Team prices that appear in the documentation. That mismatch does not prove the documented plans are unavailable. It does mean a buyer should capture the actual checkout or quote before approving a budget.

The same check should record billing frequency and included use. Annual prices can lower the displayed monthly equivalent while increasing the commitment. A sales-led Enterprise quote may also bundle support or governance that cannot be compared with a self-serve card line by line.

An Action Is a Tool Run, Not a Completed Account

Relevance AI's plan documentation defines an Action as one tool run. A workflow that searches for a company, enriches a contact, reads a CRM record, writes a note, and creates a task can consume several Actions for one account.

Failed tool runs still count. Retries can therefore increase Action use without producing an accepted record. Conditional branches matter too. A qualified account may take a longer path than a rejected account, while a malformed input may trigger repeated calls before the workflow stops.

The platform's Workforces documentation explains how specialist agents can hand work to one another and follow mandatory or conditional steps. That flexibility is useful, but each tool call inside those routes can affect consumption. Counting workforce tasks alone hides the cost mechanism.

Build a run ledger with these fields:

The important ratio is Actions per accepted account. A flow that appears cheap per task can be expensive when most records fail qualification or require correction.

Vendor Credits Are a Separate Meter

Vendor Credits cover supported model and tool-vendor costs. The detailed pricing page publishes included credit values by plan, then lists an extra 10,000 Vendor Credits for $20. Relevance AI also documents optional bring-your-own-key arrangements that can move some model charges to the buyer's own provider account.

Bring your own key does not make model use free. It changes who bills it and can change which usage controls or contracts apply. The cost model should still include that provider invoice. It should also record which model handled each step because research, classification, and drafting do not need the same model or context size.

Third-party data and browser services can sit outside Vendor Credits or carry their own charges. The sales workflow may also need a CRM, enrichment provider, mailbox, sending tool, and monitoring service. Relevance AI coordinates those components; it does not turn every connected product into one bundled price.

Use a low-cost model for simple classification only after the team has measured error rates. Source-sensitive research and final message review may require a different model or a human decision. Lower token cost does not help when the output sends the wrong claim to a prospect.

Overage Controls Need Deliberate Setup

Published top-ups are $80 for 1,000 Actions and $20 for 10,000 Vendor Credits. Those figures make incremental use visible, but they do not answer when the platform buys or applies extra capacity for a specific account.

The spend-control documentation describes controls for managing consumption. Some documentation still refers to plan names and features that are not visible on the marketing pricing page. Confirm the controls available on the purchased plan rather than assuming every setting applies everywhere.

Set a pilot cap below the maximum approved budget. Alerts should fire before the cap, not after a team loses the ability to complete an urgent workflow. The owner should know whether the preferred response is to pause, simplify the route, buy a top-up, or move to another plan.

A daily view is useful during launch. Track spikes by agent version and trigger, then compare them with errors and accepted output. A jump in Actions may reflect healthy volume, an accidental loop, duplicate events, or a failing connector. Cost without execution evidence cannot distinguish those cases.

Approval Changes Labor Cost More Than Platform Cost

Relevance AI's tool settings include Auto Run, Approval Required, and Let Agent Decide. The approvals documentation lets a reviewer approve, reject, or guide a proposed action.

Approval should cover every external send and CRM mutation during an outbound pilot. That adds review time, which belongs in total cost. It also reduces the chance that a bad research result becomes a sent message or a damaged CRM record.

Do not remove approval only to improve the cost model. First measure rejection reasons and correction time. Automate a step only when its inputs are controlled, its result is reversible, and its error rate is acceptable. Keep approval on high-consequence actions even if the review line remains visible in the budget.

A useful labor calculation is reviewer minutes per accepted account. Combine that with platform and provider charges. This prevents an apparently inexpensive agent from hiding hours of manual cleanup.

An Outbound Cost Model Needs the Whole System

A representative pilot might process 100 accounts. It should include duplicates, missing fields, weak-fit records, and sources that disagree. Run the same routing, research, approval, and CRM steps planned for production.

For each accepted account, calculate:

(subscription allocation + Action top-ups + Vendor Credits + external provider charges + review labor) / accepted accounts

Do not divide by submitted accounts. Keep every failed call, rejected record, and correction in the numerator, while the denominator contains only records the team is willing to use.

Keep sending infrastructure outside the Relevance AI line. Gmail actions or an outbound agent do not establish mailbox warmup, domain rotation, suppression, unsubscribe handling, or reputation monitoring. Google's email sender guidelines apply independently to messages reaching personal Gmail accounts.

The same separation applies to data permission and platform rules. A tool that can collect or act on data does not grant permission to use the source. Legal review and source-policy controls remain part of the operating system.

When Pro, Team, or Enterprise Fits

Pro is the sensible test tier when one owner can keep the workflow narrow and the included 2,500 Actions cover a meaningful sample. It is a poor test if the planned process needs many tool calls per account and the sample is too small to expose failures.

Team fits a shared operating process that needs more included use and collaboration. Its price jump should be justified by a measured production route, not by the desire to build more agents. Compare the Team allowance with observed Actions and Vendor Credits from the pilot.

Enterprise belongs in the shortlist when retention, audit export, role controls, evaluations, support, or contractual terms decide the purchase. Relevance AI documents several governance and analytics capabilities as Enterprise features. Obtain the exact entitlement list and test the required export before signing.

Keep looking when the buyer needs one transparent fee for a turnkey outbound stack. The same applies when mailbox operations, sender reputation, and sequence reporting are the main requirement. Relevance AI pricing covers an orchestration platform, not every system it can call.

LeadHaste connects research, verification, sending infrastructure, reply handling, and CRM evidence in one client-owned operating system. Our outbound services show where an agent platform fits and which controls remain outside it.

Price a Real Workflow Before You Commit

Bring one representative workflow, its expected monthly volume, connected providers, and approval steps to a free ICP and campaign-fit discovery call. We will map the Action path, credit use, human review, and external costs before the plan decision. Book your free discovery call →