How to Prove Agentforce ROI in 5 Metrics (Before Your CFO Asks)

By month two, your CFO’s question probably isn’t, “Does Agentforce work?” It’s: “What is the AI consumption bill buying?”
Answer with more than a demo, a deflection percentage, or a handful of encouraging conversations. Build a scorecard that connects usage to completed work, service quality, human effort, and customer outcomes.
You can instrument the five metrics below with Salesforce tools and data you may already have. The goal isn’t to buy another dashboard. It’s to make your existing data answer a business question.
The timing matters. In Salesforce’s State of Service: AI Agents Edition, a survey of 3,075 service professionals fielded March 9 to April 4, 2026, 70% of organizations with AI service agents reported measurable value within 60 days. Agentic AI adoption in service rose from 39% to 66% year over year. Those figures are useful context, not a forecast for your organization.
1. Cost per resolved conversation, not cost per case
What it answers: How much are you paying for a conversation that the agent actually resolves?
Start with Flex Credit consumption from Digital Wallet and divide the cost for the same period by the number of conversations judged resolved. Keep the denominator honest: a conversation that ends with an unresolved issue is not a low-cost win.
For planning, Salesforce’s published standard pricing is $500 per 100,000 Flex Credits, or $0.005 per credit. A standard Agentforce agent action carries a flat 20 Flex Credits, regardless of token count: $0.10 per action at that list price. Your actual cost depends on your contracted terms and usage. Use Digital Wallet as the source of truth for billing.
👉 Trap: Reporting cost per interaction while unresolved sessions quietly count as cheap. Measure cost against contextually resolved conversations, not activity alone.

2. Resolution rate judged on context, not signals
What it answers: Did the customer’s issue get resolved, or did the conversation merely look finished?
Use Agentforce Observability to evaluate the full session, not just a signal such as a customer typing “thank you.” Its session outcome metrics use context-based evaluation: an LLM reviews the conversation using an editable prompt. Define “resolved” for your business, then configure quality-score prompts and custom LLM-as-judge scores for criteria such as accuracy and tone.
Agentforce Observability is included at no additional Data Cloud cost for Agentforce customers, according to Salesforce. Session tracing structures logged events down to individual conversation turns, with citations to retrieved source chunks and waterfall traces for multi-agent chains. Use those details to investigate why a session passed or failed your resolution test.
👉 Trap: A polite dead end looks like a win under signal-based deflection. A customer-friendly closing line is not proof that the underlying request was handled.
A data-readiness warning
Metric 2 (and the escalation analysis in metric 3) can fail when the agent lacks reliable knowledge. Salesforce reports that 72% of service operations professionals call data readiness a major blocker to AI, compared with 59% of customer service leaders. Review session citations to see whether the agent retrieved relevant information or exposed a knowledge gap. Enable audit logging early, too: the usage data captures new events going forward and does not backfill historical activity.

3. Escalation quality
What it answers: When the agent hands off a conversation, was escalation the right decision, and how much human work did it actually remove?
Track the share of escalations that were appropriate, then measure human handling time on those escalated cases. Compare with your own pre-agent baseline. Review session traces to understand where handoffs happen, what the agent already gathered, and whether a knowledge or process gap caused the escalation.
For context, a Forrester Consulting Total Economic Impact study of Agentforce for Customer Service reported a 50% reduction in case handling time, alongside 35% of cases deflected and 396% ROI over three years. Treat those as study results, not a promise or a target you can assume. Test the handling-time change against your own workflow.
👉 Trap: Deflection rises while reopen rates rise, too. That’s a funnel, not a fix. A useful escalation should bring the right issue to a person with enough context to act.
4. Consumption per unit of work
What it answers: Is AI usage producing more completed work, or simply consuming more tokens?
Track Agentic Work Units (AWUs) and tokens together. Salesforce defines an AWU as a discrete task accomplished by an AI agent, such as a processed prompt, completed reasoning chain, or tool invocation. AWUs measure platform work; tokens measure model consumption. Their relationship is elastic, so watch the inference-to-work ratio rather than treating fewer tokens as the only goal. Salesforce notes that output tokens can cost up to 10 times more than input tokens.
For a weekly view, use the AiAgentGenerativeAiUsage data model to track metered versus unmetered consumption, feature, prompt input/completion/total tokens, user, and timestamp. It requires Data 360 and the Einstein Generative AI Audit and Feedback Data Collection toggle. Join it to the Agent Session DMO to attribute usage by business context, such as cost center, country, or brand.
For planning estimates, prompt-template usage is roughly four Flex Credits per 2,000-token block. Token-to-credit conversion is directional only; Digital Wallet is the billing source of truth. Configure consumption alerts at 75%, 90%, and 100% through the Flow template. The wallet does not stop execution when credits run out; overages are billed at contracted rates. Alerts are your early warning.
Salesforce reports 3.8 billion AWUs generated, with 1.6 billion in Q1 alone, up 111% quarter over quarter; Service Agents reached 232 million Q1 AWUs, up 66%, and Employee Agents rose 148%. Those are platform-level activity figures, not ROI benchmarks for your deployment.
👉 Trap: Waiting for the invoice to notice an overage. Review consumption and work together, before the billing cycle closes.
5. Time to first measurable value, and the CSAT trend
What it answers: Did the deployment show measurable business value on a defined timeline, without losing sight of customer experience?
Set a day-60 review before launch. Salesforce’s survey found that 70% of organizations with AI service agents observe measurable value within 60 days. A useful checkpoint, not a guarantee. Include customer satisfaction (CSAT) in the scorecard: respondents identified it as the KPI most improved after deploying agents, ahead of rep productivity, average handle time, retention, and first-response time.
Agree in advance on what “measurable value” means for this use case. Then compare the same period’s resolved conversations, escalations, handling time, consumption, and CSAT against the baseline you chose. Keep the review focused on evidence, not anecdotes.
👉 Trap: A pilot that never gets a day-60 review can drift into a permanent science project. Put the checkpoint, and the decision it supports, on the calendar.
Make the scorecard a monthly operating habit
Set aside 30 minutes each month. Bring one page, five numbers, and one decision per metric:
- Cost per resolved conversation: Is the spend tied to completed outcomes?
- Context-based resolution: What prompt, knowledge, or process needs adjustment?
- Escalation quality: Is the handoff useful, and is human effort changing?
- Consumption per unit of work: Is usage aligned with completed work, and are alerts configured?
- Time to value and CSAT: Is the trend strong enough to continue, change, or expand?

The metrics are yours. The operating discipline is where support helps. NextEd Consulting can help instrument and run this scorecard through Salesforce managed services, including ongoing optimization and support. We bring the Salesforce consulting services and support and maintenance to connect CRM strategy, consumption data, and service outcomes, so your team can bring numbers, not anecdotes, to the budget review.
Ready to make your Agentforce investment CFO-legible? Get in touch with NextEd to talk about managed services that keep the scorecard running.

