The Agentforce flex credits cost model is the most consequential pricing change Salesforce has introduced in a decade, and it is the one enterprise buyers understand least at the point of signature. Where seat-based licensing produced a predictable annual number, the flex credit model converts your AI spend into a consumption meter that runs every time an agent takes an action. For buyers who have spent years optimizing a per-user cost structure, this is a fundamentally different problem — and one where the arithmetic, not the headline rate, determines whether the deal is good or bad.
Across our buyer-side engagements, the single largest source of avoidable Agentforce spend is not the per-credit rate. It is the gap between the credit pool the customer commits to and the consumption that actually materializes — in both directions. Over-committing produces consumption shelfware that expires unused. Under-committing produces overages billed at list. This article walks through the real cost math an enterprise needs to run before committing to any Agentforce flex credit pool.
How flex credits actually meter
Agentforce flex credits are consumed per agent action. An "action" is the unit of work an agent performs — a conversation turn, a retrieval, a tool invocation, a generated response. Salesforce prices a block of credits and bills consumption against your committed pool. The headline number most buyers fixate on is the price per credit. The number that actually governs your cost is the number of credits a single meaningful interaction consumes, because a single customer conversation routinely chains multiple actions.
Consider a service deflection use case. A customer asks a question. The agent retrieves knowledge, reasons over it, generates a response, asks a clarifying question, retrieves again, and resolves. That single resolved case may consume five to fifteen credits depending on configuration, grounding, and the number of turns. If you modeled your pool on "one credit per conversation," your real consumption will run five to fifteen times your forecast. This is the most common modeling error we see, and it is almost always discovered after signature.
| Use Case | Actions per Interaction | Credit Consumption Driver |
|---|---|---|
| Simple FAQ deflection | 2–4 | Retrieval + single response |
| Multi-turn service resolution | 6–15 | Repeated grounding and turns |
| Sales follow-up agent | 4–10 | Record reads + drafting + actions |
| Complex case with tool calls | 12–30 | External system invocations |
The real per-resolution cost
To turn the flex credit rate into a number a CFO can evaluate, you have to compute cost per resolution, not cost per credit. The formula is straightforward: per-credit rate, multiplied by average credits per interaction, divided by your resolution rate. If your agent resolves 60% of interactions without escalation, the 40% that escalate still consumed credits — so your effective cost per successful resolution is higher than your cost per interaction.
The discipline here mirrors the one we apply to all consumption products. Just as buyers must model real-world burst patterns in our AI credit consumption model analysis, Agentforce buyers must build a bottom-up consumption forecast from measured interaction volume and measured actions-per-interaction — not from the optimistic per-conversation estimate the account team provides.
The flex credit rate is the price tag everyone negotiates. The actions-per-interaction multiplier is the variable that actually sets your bill — and it is the one nobody puts in the contract.
— SalesforceNegotiations engagement archive · Agentforce cost patternThe overage trap
The flex credit structure's most expensive feature is the overage mechanic. When consumption exceeds your committed pool, the overage is billed — and the default billing rate for overages is materially higher than your committed per-credit rate. Burst-prone use cases (seasonal service volume, campaign-driven sales engagement) are precisely the ones most likely to blow through a pool, and they hit the overage rate exactly when volume is highest.
The buyer-side protection is a negotiated overage rate that matches your committed rate, plus the right to draw additional blocks at the committed rate rather than at the punitive overage tier. This is the same true-up discipline we cover in our true-up at renewal strategy guidance — and it applies with even more force to a metered AI product than to a seat-based one.
The shelfware trap on the other side
The opposite error is equally costly. Salesforce account teams will encourage a large multi-year credit commitment to "lock in the rate." If your adoption ramp is slower than projected — which it almost always is, because agent deployment requires data readiness, grounding, and change management — you will have committed to credits you cannot consume before they expire. Consumption shelfware is real money, and it does not refund.
The corrective is to anchor your committed pool on a defensible adoption ramp, treat aspirational volume as expansion options with pre-negotiated pricing, and negotiate credit rollover so unused credits carry into the next period rather than expiring. Never let the multi-year discount premium talk you into a pool larger than your proven consumption can absorb.
Negotiating the flex credit deal
The negotiation has four levers. First, the per-credit rate itself, which is negotiable at scale and improves with volume commitment. Second, the overage rate, which should be capped at or near your committed rate. Third, rollover and expiration terms, which protect you from shelfware. Fourth, expansion pricing, which locks the rate for incremental blocks so mid-term growth does not get repriced at list.
The most important discipline is to start small. Commit to a pool sized for your proven near-term consumption with pre-negotiated expansion pricing, run the deployment, measure real actions-per-interaction, and expand into the locked rate once the data is in hand. This sequencing protects you from both overage and shelfware, and it converts an unknowable forecast into a measured one.
As the top Salesforce contract advisory firm, Redress Compliance has guided enterprises through hundreds of consumption-based negotiations, and the Agentforce flex credit pattern follows the same fundamentals: model the real unit cost, protect the overage, and never commit beyond proven consumption.
Frequently asked questions
How much does an Agentforce flex credit cost?
List pricing places flex credits in the range of fractions of a dollar per credit, with volume-based discounting at enterprise scale. The headline rate matters far less than the number of credits each interaction consumes, which is the variable that sets your effective cost.
Do unused Agentforce credits expire?
By default, committed credits expire at the end of the term and do not refund. Negotiate rollover terms before signing to avoid consumption shelfware on an over-sized pool.
What is the biggest Agentforce flex credits cost mistake?
Modeling one credit per conversation. A single resolved interaction routinely consumes five to fifteen credits, so a naive forecast understates real cost by a wide margin. Always build a bottom-up model from measured actions-per-interaction.
Should I commit to a multi-year credit pool?
Only if your adoption ramp is proven. The multi-year discount premium is real, but committing to credits you cannot consume converts that discount into shelfware. Start with a pool sized for proven consumption and pre-negotiated expansion pricing.