Do We Need Two Contracts to Buy Claudeforce, and How Do We Budget Two Meters?

Competitors already own the short answer. We are late to put a Director+ version on the Record with receipts.

Yes. As of the 26 August 2026 Claudeforce launch coverage, you do not buy "Salesforce in Claude" on one order form. You budget two meters that behave differently, then you refuse any Dreamforce demo that collapses them into a single seat story.

This page is the procurement decoder. The wait-versus-buy filter stays on the sibling page. Read that first if you still need the four AE questions, the premium-edition call commentary, and the dirty-CRM gate: Wait, Buy, or What to Ask the Salesforce AE Before Dreamforce?.

The buyer question, stated cleanly

Director of RevOps, VP Sales Ops, CRO partner, procurement lead, founder at 100+ headcount:

Do we need separate Salesforce and Anthropic contracts to run Claudeforce, and how do we put a number in the FY27 sales-tech budget when neither meter has a published Claudeforce rate card?

Useful answer: treat it as two independent cost lines with two failure modes. One can throttle you. One can invoice you without a ceiling you have already modeled.

What the primary sources actually say

The 15 September 2026 AIforce announcement says Salesforce in Claude / Claudeforce has been piloted by companies like Deloitte, GitLab, and Legora and is now available to all customers in beta. Still not GA. No published Claudeforce list price, no published Flex Credit rate, and no published edition requirement. The earlier Claudeforce product page is stale on availability. Source: salesforce.com/news/stories/aiforce-announcement.

Stokes, on the record to VentureBeat, described the commercial shape: Salesforce charges through headless consumption pricing tied to your user license edition's API-call headroom. Customers contract separately with Anthropic for Claude inference. His line that matters for procurement: you cannot buy this on one piece of paper at the moment. Source: VentureBeat.

Anthropic's published Enterprise plan is seat price plus usage at API rates. Anthropic's help center states Enterprise seats are priced per user per month, billed annually, and that usage is not included in the seat fee. Self-serve minimum is 20 seats. That is Anthropic's general Enterprise commercial model, not a Claudeforce SKU. Use it as the floor for the Claude side of the budget, not as a signed Claudeforce quote. Sources: anthropic.com/pricing and Enterprise plan help.

No invented Claudeforce list price appears on this page. If your AE hands you one verbally at Dreamforce, get it on an order form.

Meter 1: Salesforce headless consumption (can throttle)

What it is: API and MCP work against your org, billed or entitled as headless consumption against the edition you already run.

What breaks if you ignore it: sellers lean on Claude for meeting prep, deal health, and pipeline review. Call volume against Salesforce rises. Edition-tied headroom becomes a throttle. The "UI is the AI" pitch turns into a capacity ticket to Salesforce support.

What Director+ does before pilot scale:

1. Name the current edition and the published API / headless entitlement for that edition.
2. Ask the AE, in writing, whether Salesforce in Claude draws on that same entitlement or adds a new consumption SKU.
3. Ask how Flex Credits and Agentforce spend interact if you already buy them. Dual-path with Agentforce is already the Report's wait-or-buy position. Do not assume Claudeforce replaces that line.
4. Put a daily call budget on the pilot cohort. Measure MCP and API calls per completed seller task for two weeks before you expand seats.

Meter 2: Anthropic inference (can invoice without your model)

What it is: Claude tokens for the work sellers do inside Claude, on a contract with Anthropic.

What breaks if you ignore it: the Salesforce meter may throttle. The Anthropic meter bills. Multi-step skills that look like "one question" in the chat UI can be several model turns plus tool calls. Your FY27 model that treated Claude as a fixed seat line will understate the run rate.

What Director+ does before pilot scale:

1. Confirm whether sellers need Claude Team, Claude Enterprise, or a custom Anthropic agreement to use Salesforce in Claude in your region.
2. Use Anthropic's published Enterprise shape (seat + usage at API rates, self-serve floor of 20 seats) only as a planning floor. Replace it with the AE's written Anthropic terms before board commit.
3. Cap org and user spend in Anthropic admin before the pilot opens. A pilot without a spend ceiling is a blank PO.
4. Log tokens per completed seller workflow the same way you log Salesforce API calls. Same two-week window.

What this is not

This is not "Claudeforce is a bad deal." Sellers hate Lightning navigation. Claude is strong at synthesis over messy records. Salesforce is right that metadata and permissions are the moat.

This is not a wait-versus-buy verdict. That lives on the wait-or-buy decoder.

This is not an Agentforce replacement story. Salesforce positions Claudeforce as dual-path. If you already pay for Agentforce or Flex Credits, budget Claudeforce as an additional surface until packaging says otherwise in writing.

Procurement checklist you can read in a budget review

Copy this into the room:

1. Two legal entities on the PO path: Salesforce and Anthropic. Confirm both exist before beta week one.
2. Two meters with two owners: RevOps owns the Salesforce consumption dashboard. IT or the Claude admin owns the Anthropic spend caps.
3. No single Claudeforce seat price in the model until a primary source or signed order form publishes one.
4. Premium-edition commentary from the Q2 FY27 earnings call is still call commentary, not a Claudeforce SKU. Tie any edition upgrade to a written AE answer.
5. Kill the beta if either meter cannot be reported weekly by name of seller cohort.
6. Do not expand past the beta cohort until CRM truth tests clear. Dirty CRM on two meters is a prettier, more expensive lie. Use the pipeline truth tests.

Related: Agentforce Flex Credits forecast and close · Claudeforce wait-or-buy · Pipeline truth tests · AI SDR kill criteria