Agentforce ROI: how to calculate it honestly before you build (2026)

“What’s the ROI on Agentforce?” is the first question in every executive review and the one with the least honest answer floating around. The vendor decks quote payback periods; the case studies quote deflection percentages; neither tells you what your number will be, because ROI isn’t a property of the platform — it’s a property of your volume, your cost per interaction, and how honestly you count the savings.

The good news is that the model is simple and you can run it yourself before committing a rupee or a dollar. This post walks through the actual calculation — the cost side, the benefit side, the break-even volume most use cases need to clear, and the savings you should refuse to count. If you’d rather just plug in numbers, the Agentforce ROI Calculator does the arithmetic; this is the reasoning behind it.

The cost side: what an agent actually costs to run

You can’t compute a return without the denominator, and Agentforce’s cost has moved around, so pin it down for your org before anything else.

• Flex Credits is the model Salesforce steers most deployments toward: a standard agent action runs about $0.10 (20 credits, with credits sold at $500 per 100,000). A single resolved interaction fires several actions, so cost-per-resolution is a small multiple of that, not a flat fee.
• The legacy per-conversation model charged a flat $2 per conversation — a complete interaction from start to resolution, regardless of message count. Some orgs are still on it, and it’s a useful mental anchor because it’s the easiest number to compare against a human’s cost per case.
• Per-user licensing exists if you’d rather not meter at all. And Enterprise Edition and up get 100,000 Flex Credits free through Salesforce Foundations, which is enough to run a real pilot and measure your own cost-per-resolution before you forecast anything.

The single most common ROI error is forecasting cost from a demo. An agentic interaction fires more model calls than you expect once it’s grounding, reasoning, and calling actions. Measure credits-per-resolved-interaction under real traffic — the free Foundations credits exist precisely so you can — before you extrapolate. The full mechanics are in our note on how Agentforce pricing really works.

The benefit side: case deflection is the math that holds up

Most defensible Agentforce ROI comes from case deflection — the agent fully resolving interactions a human would otherwise have handled. It’s the cleanest to quantify because both sides of the equation are things you already know.

The calculation:

1. Baseline cost. Take the query type you’re targeting (say, tier-1 “where’s my order” cases) and compute your fully-loaded cost per case — agent salary and overhead divided by cases handled. Most orgs land somewhere between a few dollars and low-double-digit dollars per tier-1 case.
2. Autonomous resolution rate. What share will the agent resolve end-to-end without escalating? Industry benchmarks for well-configured, well-grounded agents cluster around 55–70% for suitable case types — but that’s a target, not a promise, and it collapses if your data or resolution paths are messy.
3. Labor displaced. Multiply your interaction volume by the resolution rate to get the cases the agent absorbs, then by your loaded cost per case. That’s your gross annual saving.
4. Net it against agent cost. Subtract the annual Agentforce cost (resolved interactions × your measured cost-per-resolution, plus any licenses). What’s left is your real return.

The break-even reality most decks skip

Here’s the part that determines whether a use case is worth building at all: deflection only pays when volume is high enough that the per-interaction agent cost is below the human cost it replaces. As a rule of thumb, a use case needs to be deflecting on the order of 200–400 interactions per month before the economics clearly favor the agent. Below that, you’re paying to automate something a human handles more cheaply, and the “ROI” is a rounding error against the build cost.

This is why use-case selection is the ROI lever, not the agent itself. A high-volume, repetitive, well-documented query type clears the threshold easily. A low-volume, high-variability workflow never will, no matter how good the agent is. Picking the wrong first use case is the most expensive ROI mistake, and it happens before a line of configuration is written — which is exactly the failure pattern we see in what actually breaks Agentforce projects in production.

A worked example (with numbers you’d supply)

Illustrative only — plug in your own:

• Target: tier-1 order-status cases. Volume: 1,000/month.
• Loaded cost per case (human): $8.
• Autonomous resolution rate: 60% → 600 cases/month deflected.
• Gross saving: 600 × $8 × 12 = $57,600/year.
• Agent cost: 600 resolutions/month × ~$0.30 cost-per-resolution × 12 ≈ $2,160/year (measure your own).
• Net first-year return: roughly $55,000 from a single agent topic — before build cost.

Change any input and the answer moves a lot. Drop volume to 200/month and the same agent barely clears break-even. That sensitivity is the whole point: run your numbers, not the example’s. The ROI calculator lets you do exactly that; for non-agent process automation, the Automation Savings Calculator runs the same logic on reclaimed hours.

The savings you should refuse to count

Honest ROI means excluding benefits you won’t actually bank:

• “Soft” deflection savings that never leave the P&L. If deflecting 600 cases doesn’t let you redeploy or avoid hiring, you haven’t saved money — you’ve created slack. Count a saving only if it maps to a real cost you reduce or a hire you don’t make.
• Payback periods borrowed from someone else’s org. Public figures range widely — some implementations report payback inside a year, others take two. Yours depends on your volume and build cost, so treat external numbers as range-finders, not forecasts.
• Revenue and CSAT upside, in the hard-ROI column. Faster responses and better experience are real and often the bigger prize — but they’re difficult to attribute cleanly. Track them, present them honestly as directional, and don’t let a shaky revenue estimate carry a business case that deflection alone should justify.

If a use case only pencils out once you add soft savings and borrowed benchmarks, that’s your signal it isn’t the right first build. The ones worth doing clear the bar on hard deflection math alone.

Understanding the basics

How do you calculate Agentforce ROI?

Compare the cost of running the agent against the labor it displaces. Take your target query type’s fully-loaded human cost per interaction, multiply your interaction volume by a realistic autonomous resolution rate (industry benchmarks run 55–70% for well-configured agents) to get deflected volume, and multiply that by your cost per case for the gross saving. Then subtract the annual agent cost — resolved interactions times your measured cost-per-resolution (roughly a small multiple of the $0.10-per-action Flex Credit rate), plus any licenses. The net is your return. Only count savings that map to a real cost you reduce or a hire you avoid.

How much does an Agentforce conversation cost?

Under Flex Credits, a standard agent action is about $0.10 (20 credits; credits sell at $500 per 100,000), and a resolved interaction fires several actions, so cost-per-resolution is a small multiple of that. The older per-conversation model charged a flat $2 per complete interaction. Enterprise Edition and above receive 100,000 free Flex Credits through Salesforce Foundations, which is enough to measure your real cost-per-resolution before forecasting.

What case deflection rate should I expect from Agentforce?

Well-configured, well-grounded agents on suitable case types typically resolve 55–70% of interactions end-to-end without escalating. That’s a benchmark, not a guarantee — the rate depends heavily on data quality, how narrowly the use case is scoped, and how clean your resolution paths are. Broad, vaguely-scoped agents fall well short of it, which is why narrow first use cases both perform better and pay back faster.

How much volume does an Agentforce use case need to be worth it?

As a rule of thumb, a deflection use case needs to be handling on the order of 200–400 interactions per month before the per-interaction agent cost clearly beats the human cost it replaces. Below that threshold the economics rarely justify the build. This makes use-case selection — picking high-volume, repetitive, well-documented query types first — the single biggest lever on whether Agentforce pays off.

---

Want the number for your org instead of a range? Run the Agentforce ROI Calculator, or tell us the use case and we’ll pressure-test the business case with you — including when the honest answer is that it doesn’t clear the bar yet.