Define Agent Success
Agentforce surfaces a lot of data. Without a clear sense of what your agent is supposed to accomplish, that data creates noise instead of direction. Defining success before you start measuring is how you know which signals matter, which ones to ignore, and when your agent is actually improving.
Required Editions
| Available in: Lightning Experience |
| Available in: Enterprise, Performance, Unlimited, and Developer Editions. Required add-on licenses vary by agent type. |
Most of these KPIs are surfaced in Agentforce's built-in analytics dashboards. The articles that follow show which tools capture each signal and what you may need to configure to start tracking it. For a full map of the toolchain, see Understand Agentforce Observability KPIs and Tools.
We've identified three outcome areas that cover common ground across Agentforce deployments. These outcomes aren't rigid or exhaustive, and agents can touch more than one. But, knowing where your agent fits can help you select the right KPIs, reach for the right tools, and read your results with confidence.
Agent Impact (ROI)
Did the agent move the business forward?
Agent Impact covers downstream business outcomes, whether the agent is handling inbound requests, reducing cost, or driving a business result. Your agent likely falls into this area if its primary job falls into areas like deflecting cases, advancing pipeline, increasing conversion, or driving campaign engagement. In product views, the metrics that most directly reflect these outcomes appear under Effectiveness and Usage.
For example, a customer support team targeting 70% deflection handles 10,000 contacts a month. If deflection drops to 62%, that's 800 additional cases reaching a human agent, which is a concrete cost the VP of Support can act on.
Testing validates the agent behaviors behind these KPIs — correct routing, action execution, response quality. The KPIs themselves surface in production, where Analytics aggregates real user sessions into the deflection rates, abandonment rates, and engagement patterns your success statement depends on.
| KPI | What it tells you |
|---|---|
| Adoption / Enablement | Rate at which users actively engage with the agent. |
| Deflection | Share of contacts the agent handled without escalating to a human. |
| Resolution | Share of sessions where the user's underlying issue was fully resolved. This measures whether the agent solved the user's problem, not whether the agent completed its assigned task. |
| Abandonment Rate | Share of sessions where the user left without a resolution. |
| Engagement Rate | Share of sessions where the agent triggered at least one action. |
| Conversion | Business outcomes achieved, such as bookings, cart completions, opportunities progressed, or closed cases. |
| Pipeline Progression | Opportunities created or advanced through the funnel. |
| Campaign Engagement | Responses, click throughs, or interactions driven by the agent. |
| Return / Repeat Rate | Share of users who return after a prior session. |
| CSAT / NPS | Customer satisfaction at the session or relationship level. |
Agent Effectiveness & Interaction
Was this a good conversation?
Effectiveness and interaction is about in-session quality from the user's perspective, like whether the agent understood what was asked, responded accurately, and made the interaction feel smooth. Your agent likely falls into this area if the quality of individual interactions is the core success measure:
-
a Sales assistant where instruction adherence and response accuracy matter
-
an employee-facing agent where re-prompt rate is a signal of friction
-
any agent where user feedback (CSAT, thumbs up or down, NPS) is tracked as an outcome.
In product views, these metrics appear under Effectiveness and Quality Scores.
For example, while an agent with 92% task completion and a high re-prompt rate is technically finishing its job, users are having to ask the same thing in multiple ways to get there. So, User Experience flags a friction point that task completion rate alone doesn't surface.
Quality scores and Instruction Adherence are measurable in a testing environment — you can track them in Testing Center and Agentforce Builder before you deploy. Signals like CSAT and User Experience only emerge from live user sessions.
| KPI | What it tells you |
|---|---|
| Efficacy | Share of sessions where the agent completed its assigned task. This measures agent behavior, not whether the user's underlying problem was solved. |
| Quality | How accurate, safe, and relevant the agent's outputs were. |
| Instruction Adherence | How consistently the agent followed its configured instructions during the conversation. |
| Engagement Rate | Share of sessions where the agent triggered at least one action. |
| CSAT / NPS | Customer satisfaction at the session or relationship level. |
Agent Health & Security
Is the system healthy and operating within bounds?
Agent Health & Security covers the operational and technical state of the agent — whether it's fast, available, and staying within its configured guardrails. Your agent likely falls into this area if uptime, latency, or compliance are explicit requirements:
-
a regulated industry deployment where guardrail violations must be tracked
-
a high-volume deployment where latency affects user experience at scale
-
any agent where availability is a contractual concern
In product views, these metrics appear under Health and Trust.
For example, an escalation rate spike at 2 AM looks alarming in daily Analytics. However, watching availability on a 5-minute lookback reveals it was a 20-minute outage, not a sustained behavior change. Without this outcome area in view, the same signal gets misread and the wrong fix gets prioritized.
Latency, instruction adherence, and guardrail violations are detectable in a testing environment. Availability and escalation rate only become meaningful at production scale.
| KPI | What it tells you |
|---|---|
| Performance | How quickly the agent responds, measured by turn latency, retrieval latency, and throughput. |
| Availability | Whether the agent is up and responding without errors or timeouts. |
| Escalation Rate | Share of sessions that transferred to a human agent or another system. |
| Instruction Adherence | How consistently the agent's responses met its configured instruction requirements. |
| Constraints & Guardrails | Whether the agent operated within its defined compliance, privacy, and content boundaries. |
Most agents touch more than one area. Start with the area most directly tied to the reason the agent was built, then pick two to four KPIs from that area. If your agent clearly spans multiple areas, decide which lens is primary before you deploy.
You should be able to complete a success statement:
For example, for a service agent: "This agent succeeds when billing questions are resolved without a handoff, as measured by a deflection rate at or above 70% and CSAT at or above 4.0, for the customer service team handling billing contacts."
Or for a sales agent: "This agent succeeds when sales reps get accurate product recommendations on first ask, as measured by instruction adherence at or above 90% and re-prompt rate below 15%, for the enterprise sales team."
In the Observe phase, your success statement is the benchmark against which you compare production data. When a KPI drifts from this definition of success, that's the signal to take action, starting with root cause analysis. When multiple KPIs drift at once, this statement tells you which one to prioritize.
If you can fill in all three parts and point to which outcome area carries the most weight, you have what you need to move forward.
