Agentforce Flex Credits: The Per-Action Meter and What Seat-to-Consumption Does to NRR
The question
What exactly are Salesforce Agentforce "Flex Credits," how do per-conversation vs per-action vs token pricing tiers compare, and what does the seat-to-consumption transition do to Salesforce's net revenue retention math? Context: the Agentforce strategy brief (2026-06-14) flagged this as its #1 open follow-up — we had ~$540M ARR at 330% YoY and expansion-led bookings but zero depth on the actual pricing mechanics.
What we already know (from the vault)
- The prior brief established the load-bearing re-frame: the pricing model is the tell. Per-seat pricing was the financial expression of the UI moat; consumption pricing is the financial expression of the workload surviving without the UI — and named Flex Credits / tokens / LLM calls as the meters to watch ([[2026-06-14-salesforce-agentforce-strategy]]).
- Agentforce ARR ≈ $540M at 330% YoY; >50% of new Q3 bookings came from existing customers buying more credits; 11.14T tokens served by Jan 2026; top 50 customers drove 200M+ LLM calls in a quarter — i.e. the revenue is already consumption-shaped and expansion-led ([[2026-06-14-salesforce-agentforce-strategy]]).
- The consumption-ARR accounting problem is already mapped in the vault: usage-based "ARR" is a run-rate annualization that bounces with ramp, unlike contracted seat ARR — directly relevant to how Salesforce reports and how NRR behaves ([[2026-06-04-mostly-metrics-consumption-based-arr]]).
- Usage-based AI is scaling faster than seat-based SaaS ever did because the meter tracks value delivered — top AI startups hit ~$30M ARR in ~18 months, ~3x the 2018 SaaS benchmark ([[2026-04-29-every-compute-is-new-cash]]).
- Nadella frames the end-state as a hybrid: a subscription is just "an entitlement to some consumption rate" — seats and consumption co-exist as meters rather than one cleanly replacing the other ([[2026-06-04-stratechery-satya-nadella-microsoft-core-competencies]]).
What the web says
- A Flex Credit is a $0.005 unit of consumption. The standard package is $500 for 100,000 credits. A standard agent action costs 20 Flex Credits = $0.10; a voice action costs 30 credits = $0.15; a sandbox action is discounted to ~16 credits (jitendrazaa.com, concret.io).
- An "action" is one discrete task — update a record, generate a case summary, search a knowledge base, run a flow, schedule an appointment. Actions are counted only once completed (concret.io).
- The token tier sits inside the action. Each action includes up to 10,000 tokens of processing; exceeding that bills as multiple actions (15K tokens = 2 actions = $0.20; 25K tokens = 3 actions = $0.30). So token pricing isn't a separate tier — it's a multiplier on the per-action meter (jitendrazaa.com).
- The old model was $2.00 per conversation, flat regardless of work done — a "conversation" ran from the AI's first response to issue-close or 24h inactivity, with re-engagement triggering a fresh $2 (concret.io). Salesforce moved off it on May 15, 2025 (salesforce.com press release).
- Break-even is 20 actions per interaction. A typical conversation runs 5–15 actions, so Flex Credits cost $0.50–$1.50 vs the old flat $2; a one-action FAQ is $0.10 — a ~95% cut. Above 20 actions, the old $2 flat rate is cheaper. You cannot run both models in the same org — one meter per org (concret.io, agencyq.com).
- Seats didn't disappear — they became credit-bundled tiers. Agentforce add-on ≈ $125/user/mo (unlimited internal use); Agentforce 1 Edition ≈ $550+/user/mo including 1M pooled Flex Credits/year; standalone Flex Credits need no user license (for external/customer-facing agents); Salesforce Foundations bundles 200K free credits/yr (jitendrazaa.com). This is the "3+ pricing models" SaaStr flags as deliberate (saastr.com).
- On NRR, analysts split. The bear: Agentforce automates work, so customers may need fewer human seats — consumption must more than offset seat shrinkage to keep NRR above 100%, and usage revenue introduces forecast volatility / smaller-than-expected contracts. The bull: the Agentforce pivot is the single biggest reason CRM deserves a different framework than still-seat-based monday.com/Asana — it insulates Salesforce from the "AI shrinks seat counts" threat by attaching a second, value-tracking meter (solvimon.com, investing.com, simplywall.st).
Convergences and contradictions
- Convergence (web + vault): Both confirm the revenue is consumption-shaped and expansion-led, and both frame consumption-as-ARR as a run-rate annualization that behaves differently from contracted seat ARR ([[2026-06-04-mostly-metrics-consumption-based-arr]]). The per-action meter ($0.10 / 20 credits / 10K tokens) is consistent across all independent sources.
- Refinement: "Token pricing tier" in the question is slightly mis-framed by the market's actual design — tokens are not a third standalone tier. The real architecture is two billing models (per-conversation OR per-action) with tokens as an over-threshold multiplier inside the per-action model. Seats persist as a third, credit-bundled layer.
- Contradiction (the live debate): Whether seat-to-consumption is NRR-accretive or NRR-dilutive is genuinely unresolved. The same fact — agents do work humans used to — reads as "fewer seats sold" (bear) or "new value-tracking meter that compounds with usage" (bull). No source has clean cohort data resolving it yet; Salesforce itself guided no "meaningful" revenue until FY27.
Synthesis for RDCO
The mechanics resolve cleanly into a hierarchy, which matters for how sharply Ben can render the thesis. There are two mutually-exclusive consumption meters per org — the legacy $2/conversation flat rate and the current $0.10/action rate — and tokens are a multiplier inside the action meter (one action = 10K tokens; spill over and you're billed in additional action-units). Seats did not die; they were repackaged as credit-bundled licenses ($125–$550+/user/mo, the higher tier pre-loading 1M credits). So the honest picture is not "seats → consumption" as a clean swap. It is a seat floor with a consumption meter bolted on top — exactly Nadella's "subscription = entitlement to a consumption rate" ([[2026-06-04-stratechery-satya-nadella-microsoft-core-competencies]]). The move from $2-flat to $0.10-per-action is itself the tell-within-the-tell: Salesforce cut the headline price ~95% on simple interactions to drive adoption volume, betting that metered expansion on complex multi-action workloads more than recovers the per-unit cut. That is a classic consumption-economics wager — trade margin-per-event for event-volume — and it only pencils if usage compounds.
For NRR math, this is the crux of the "SaaS premium dies, workload lives" thesis rendered financially, and it cuts both ways in a way the founder should hold precisely. Seat-based NRR is mechanically sticky and predictable: contracted seats renew, expansion is a sales motion, and >100% NRR is the engine of the SaaS premium multiple. Consumption NRR is elastic and reflexive: it rises automatically when customers use the product more, but it can also fall without a churn event — a customer who simply runs fewer agent actions next quarter shrinks net revenue with no cancellation, no logo loss, nothing for a CRO to "save." That is the volatility analysts fear, and it is structurally why a usage business trades at a lower multiple than a seat business even at the same growth rate: the revenue is real but less forecastable, less contracted, more spot-priced. The seat-to-consumption transition doesn't just change the price — it re-rates the durability of the revenue, and the multiple compresses to match. The $540M-at-330% number looks like a seat-business growth rate, but it is run-rate-annualized consumption that can de-annualize as fast as it annualized ([[2026-06-04-mostly-metrics-consumption-based-arr]]).
That is precisely "the premium dies, the workload lives" on the income statement, and the Flex Credit gives Ben the unit to make it concrete and non-derivative. The premium multiple was the capitalized value of seat-revenue stickiness — you paid a premium because humans logging in was a contract you could underwrite. Re-pricing the same customer relationship as $0.10-per-action consumption keeps the workload (the CRM is still the system of record, the agent still runs on Salesforce data and permissions) while explicitly trading the premium (sticky contracted seats) for an elastic meter (spot-priced actions). The workload survives and even grows; the multiple that rewarded the old revenue shape does not. The Sanity Check angle sharpens to: watch the Flex Credit, not the ARR headline — the $0.10 action is the price of the workload, and the gap between that price and the old per-seat premium is the SaaS multiple being demolished one action at a time. RDCO's own positioning lands in the durable layer ([[2026-06-14-salesforce-agentforce-strategy]]): someone still has to wire the data, exceptions, and regulated last-20% so those actions fire correctly — that is FDE/data-platform work, priced like a workload, not app-config work priced like a seat.
Open follow-ups
- Does Salesforce report Agentforce NRR (or any cohort retention) separately, and is consumption ARR being annualized off a trailing-30-day run rate (the bouncy method) or a contracted-minimum floor? The accounting choice changes how to read the 330%.
- Voice actions cost 30 credits vs 20 for standard — as agents move to voice/multimodal, does the blended per-action price drift up, partially re-inflating the meter? Track the action-mix.
- What share of Agentforce revenue is the credit-bundled seat tier (Agentforce 1 at $550+/user with 1M credits) vs pure standalone Flex Credits? The bundle blurs the seat-vs-consumption line and could be propping reported "consumption" growth.
- Break-even sensitivity: if typical enterprise interactions trend toward >20 actions (more autonomous, multi-step agents), the old $2-flat becomes cheaper — does Salesforce re-introduce conversation pricing or cap action-counts? Watch for a pricing re-cut.
- Cross-incumbent comparison: does SAP Joule / Microsoft Copilot meter the same way (per-action + token multiplier), making this an industry NRR re-rating rather than a Salesforce idiosyncrasy?
Related
- [[2026-06-14-salesforce-agentforce-strategy]] — the parent brief; this fills its #1 open follow-up on pricing mechanics and NRR
- [[2026-06-04-mostly-metrics-consumption-based-arr]] — how consumption ARR is annualized and why it behaves differently from seat ARR (the NRR-volatility mechanism)
- [[2026-04-29-every-compute-is-new-cash]] — usage-based AI scaling faster than seat SaaS; the bull case for consumption growth offsetting seat shrinkage
- [[2026-06-04-stratechery-satya-nadella-microsoft-core-competencies]] — "subscription = entitlement to a consumption rate"; the seat-floor-plus-meter end-state Agentforce instantiates
- [[2026-04-25-saas-death-thesis-vault-synthesis]] — the umbrella thesis this pricing analysis renders financially
Sources
- Vault: ~/rdco-vault/06-reference/research/2026-06-14-salesforce-agentforce-strategy.md
- Vault: ~/rdco-vault/06-reference/2026-06-04-mostly-metrics-consumption-based-arr.md
- Vault: ~/rdco-vault/06-reference/2026-04-29-every-compute-is-new-cash.md
- Vault: ~/rdco-vault/06-reference/2026-06-04-stratechery-satya-nadella-microsoft-core-competencies.md
- Vault: ~/rdco-vault/06-reference/research/2026-04-25-saas-death-thesis-vault-synthesis.md
- Jitendra Zaa — "Salesforce Agentforce Credits & Cost Model: Complete Guide 2026": https://www.jitendrazaa.com/blog/salesforce/salesforce-agentforce-credits-cost-model-complete-guide-2026/
- Concret.io — "From $2 Conversations to $0.10 Actions: New Agentforce Pricing Updates": https://www.concret.io/blog/new-agentforce-pricing-model
- AgencyQ — "Agentforce Flex Credits: Real Cost Math for 2026": https://www.agencyq.com/insights/article/agentforce-flex-credits-real-cost-math
- Salesforce — "Salesforce Introduces New Flexible Agentforce Pricing" (May 15, 2025): https://www.salesforce.com/news/press-releases/2025/05/15/agentforce-flexible-pricing-news/
- SaaStr — "Salesforce Now Has 3+ Pricing Models for Agentforce": https://www.saastr.com/salesforce-now-has-3-pricing-models-for-agentforce-and-maybe-right-now-thats-the-way-to-do-it/
- Solvimon — "Why is Salesforce pivoting to usage-based pricing?": https://www.solvimon.com/blog/why-is-salesforce-pivoting-to-usage-based-pricing
- Investing.com — "Salesforce Sell-Off Leaves Valuation Near Historic Lows Despite Growth": https://www.investing.com/analysis/salesforce-selloff-leaves-valuation-near-historic-lows-despite-growth-200679067
- Simply Wall St — "Consumption model shift and Agentforce will unlock market opportunities": https://simplywall.st/community/narratives/us/software/nyse-crm/salesforce/o224ptbn-consumption-model-shift-and-agentforce-will-unlock-market-opportunities