Salesforce Agentforce lists at $2 per autonomous conversation in 2026, with volume bands reducing to $0.50 to $1.20 at enterprise commit and Foundations bundles starting at 100,000 conversations per year for $180,000. Conversation cost is only the surface layer. Total cost per Agentforce conversation including Data Cloud credits, Einstein inference, and the Service Cloud or Sales Cloud seat licensing for human escalation lands between $3.40 and $7.80 per conversation. The ROI question is not whether Agentforce is cheaper than a human agent at $18 to $32 per contact. It is whether the deployed agents resolve enough conversations to recover the underlying Data Cloud and Einstein 1 platform cost.
Agentforce 2026 pricing structure
Agentforce uses a per-conversation pricing model rather than per-seat. A conversation is defined as a multi-turn customer interaction with the autonomous agent, scoped to a single intent or case. The pricing tier depends on commit size and whether Agentforce is purchased standalone or bundled into Einstein 1 Service or Einstein 1 Sales.
| Commercial model | List per conversation | Best fit |
|---|---|---|
| Agentforce Pay-As-You-Go | $2.00 | Pilots and seasonal workloads under 50,000 conversations |
| Agentforce Foundations (100k conversations) | $1.80 effective, $180,000 platform fee | Mid-market production deployments |
| Agentforce Enterprise (500k plus) | $0.80 to $1.20 negotiated | Enterprise contact centre or sales automation |
| Einstein 1 Service bundle | Included allowance plus overage | Full Service Cloud automation |
| Einstein 1 Sales bundle | Included allowance plus overage | SDR and lead-qualification automation |
The Foundations bundle is the most common landing point for new enterprise deployments. Conversations exceeding the 100,000 included allowance burn at the menu rate of $2.00 each by default. Negotiating an overage rate that matches the commit tier is one of the central commercial disciplines.
The full cost stack per conversation
The Agentforce conversation fee is the top line. Three additional cost layers stack underneath every conversation. Total cost per conversation is the sum of all four.
Layer 1: Agentforce conversation fee. $0.80 to $2.00 depending on tier.
Layer 2: Data Cloud credits. A typical Agentforce conversation consumes 50 to 200 Data Cloud credits to retrieve customer profile, recent interactions, calculated insights, and propensity scores. At commit-rate $0.0021 per credit, that is $0.10 to $0.42 in Data Cloud cost per conversation. See Data Cloud pricing for the full credit model.
Layer 3: Einstein inference. Each agent reasoning step consumes Einstein tokens. A 4 to 6 turn conversation typically runs 0.5 to 2 Einstein inference units at $0.50 to $1.20 each. That is $0.25 to $2.40 per conversation.
Layer 4: Human escalation. Agentforce deflection rates in production deployments range from 35 to 72 percent depending on use case maturity. The unresolved 28 to 65 percent escalate to human agents who carry the underlying Service Cloud or Sales Cloud seat license at $165 to $500 per user per month. Allocated per conversation that is $2.00 to $3.20 per escalated case.
| Cost layer | Per-conversation cost | Notes |
|---|---|---|
| Agentforce conversation fee | $0.80 to $2.00 | Direct, tier-based |
| Data Cloud credit cost | $0.10 to $0.42 | 50 to 200 credits per conversation |
| Einstein inference | $0.25 to $2.40 | Token consumption per reasoning step |
| Allocated human escalation | $2.00 to $3.20 | For deflection rate 35 to 72 percent |
| All-in per conversation | $3.40 to $7.80 | Full-stack cost benchmark |
Why pilot Agentforce ROI looks better than production ROI: Pilot deployments often exclude the Data Cloud and Einstein cost from the ROI calculation because those budgets sit elsewhere. Production deployments cannot avoid attributing those costs. The honest ROI calculation must compare the all-in $3.40 to $7.80 per conversation against the $18 to $32 per human-handled contact, not just the $2 conversation fee.
The ROI framework
The ROI calculation has four inputs. Pre-Agentforce contact volume by channel. Pre-Agentforce cost per contact by channel. Expected deflection rate by use case (which determines the resolution mix between Agentforce and human). All-in Agentforce cost per conversation.
A 1 million contact per year operation at $22 per human contact spends $22M annually. Agentforce at a 50 percent deflection rate and $5.20 all-in per conversation handles 500,000 conversations at $2.6M while the remaining 500,000 stay on human handling at $11M. Combined cost: $13.6M versus $22M baseline, a $8.4M annual reduction.
The same operation at a 30 percent deflection rate and $7.80 all-in handles 300,000 conversations at $2.34M and 700,000 humans at $15.4M, totaling $17.7M versus $22M baseline, $4.3M annual reduction. The deflection rate and the all-in cost discipline are the two ROI drivers.
The deflection rate trap
Salesforce headline deflection rate claims of 70 to 80 percent describe top-quartile use cases at maturity. Enterprise deployments in the first 12 months typically achieve 35 to 48 percent. Driving the rate higher requires investment in: knowledge content quality, intent taxonomy refinement, agent action authoring, and human-in-the-loop training data labelling.
Customers who deploy Agentforce without this content investment plateau at 35 percent deflection and conclude Agentforce does not work. Customers who invest 4 to 8 months of content engineering effort reach 55 to 68 percent within 12 months and recover the investment.
Negotiation levers for Agentforce
Five levers move Agentforce commercial terms.
Volume band placement. Commit at the volume band you expect to reach by Year 2, not Year 1. The list rate for above-commit overage applies until the next true-up otherwise.
True-up at commit rate. Identical mechanic to Data Cloud. Overage above commit converts to commit rate at the next anniversary rather than burning at menu rate during the term.
Bundle into Einstein 1 Service or Einstein 1 Sales. The bundled allowance plus the Service Cloud or Sales Cloud cross-product flexibility delivers 12 to 20 points more discount than buying Agentforce standalone.
Compete against Microsoft Copilot Studio for Service and Google Customer Engagement Suite. A documented competitive alternative moves the conversation price materially. Microsoft Copilot Studio per-message pricing approximately $0.01 to $0.05 per message at scale is the most directly competitive offer.
Phased deployment commitments. Buy the conversation commit in steps tied to deflection milestones. Reduces overcommit risk in the first 12 months.
How to optimise Agentforce spend
Pre-deployment. Model expected conversation volume by use case. Model expected deflection rate by use case. Model the all-in cost per conversation including the Data Cloud and Einstein layer. Sign the commit at the volume that matches realistic Year 2 use, not vendor-projected use.
During deployment. Track deflection rate weekly. Measure cost per conversation including all four layers. Identify low-deflection use cases that may need content investment or that should retire. Identify high-deflection use cases that should expand.
At renewal. Negotiate the commit tier at the actual Year 2 consumption volume. Add or strengthen the true-up clause. Renegotiate the unit price at the volume band actually achieved. For the broader Salesforce framework, see the Salesforce licensing guide, Salesforce pricing 2026, Sales Cloud pricing, Service Cloud pricing, Einstein pricing, and Salesforce renewal strategy. The Salesforce vendor hub aggregates the full cluster. Agentforce buyer engagement starts at AI procurement advisory.