Agentforce for Small Business: What It Costs and What It Can Do

As at June 2026, a small business can pilot Agentforce for free — Salesforce Foundations includes 200,000 Flex Credits on Enterprise Edition and above — and run it in production from roughly US$0.10 per agent action or US$2 per customer conversation. What it does for that money: answers customer enquiries, qualifies leads, and clears admin work, inside your existing Salesforce.

That's the summary. Below is the full pricing model with worked numbers, the three use cases we see paying off for lean teams, and — because most articles skip this part — when we'd tell you not to bother yet.

How Agentforce Pricing Works in 2026

All figures from Salesforce's official Agentforce pricing page as at June 2026 (Salesforce notes pricing is subject to change, and there is no NZD list pricing — NZ buyers see AUD or USD).

| Model | Price | Best for | |---|---|---| | Flex Credits | US$500 (AU$700) per 100,000 credits | Any use case; pay per action | | Conversations | US$2 (AU$2.80) per conversation | Customer-facing agents only | | Per-user add-ons | US$125/user/month | Unmetered employee agents (Sales/Service) | | Agentforce User License | US$5/user/month + credits | Company-wide light access | | Agentforce 1 Editions | From US$550/user/month | AI-heavy orgs; includes 2.5M credits/year |

The mechanics that matter:

• An "action" is the billing unit of the Flex model. Salesforce defines it as "a specific function that an AI agent executes on the platform, such as updating a record, summarizing a complex case, answering a product inquiry, or executing a custom prompt or flow." A standard action costs 20 credits — US$0.10 at list price. Voice actions cost 30.
• Conversations are simpler but narrower: US$2 per customer conversation, customer-facing agents only — and you cannot mix the two models in one org, so the choice is structural.
• Unused Flex Credits don't roll over between subscription terms, so don't over-buy upfront.
• There's no overage penalty — exceed your entitlement and you pay your contracted rate, billed monthly in arrears.

Worked example — employee-facing. Salesforce's own illustration: an HR-style agent answering 5 questions a month for 20 employees = 100 actions × 20 credits = US$10/month.

Worked example — customer-facing (our assumptions, list prices). Say your business receives 300 customer enquiries a month and an agent fully handles 200 of them. On conversation pricing that is 200 × US$2 = US$400/month. If the same workload averaged four actions per enquiry on Flex pricing, it would be 800 actions × 20 credits = 16,000 credits ≈ US$80/month. The spread is why model choice is worth an hour of maths before you sign — and why we model this with clients before recommending either.

The free entry point: on Enterprise Edition or above, the Salesforce Foundations add-on costs $0 and includes Agent Builder, Prompt Builder, 200,000 Flex Credits and 250,000 Data 360 credits. At 20 credits per action, that is roughly 10,000 agent actions of genuine pilot room before you spend a dollar on usage.

Three Use Cases a Small Team Can Run Today

1. A customer-question agent on your knowledge base

The workhorse. An agent grounded in your knowledge articles answers the repetitive majority of enquiries — hours, pricing, how-do-I, where-is-my — at any hour, and escalates the rest to your team with the conversation attached. Prerequisite: written answers. If your top 20 questions aren't documented anywhere, that's the project before the project. Effort: for orgs with decent knowledge content, this is typically a few weeks to a supervised launch, in our experience.

2. Lead qualification and follow-up

An agent acknowledges every new enquiry immediately, asks your standard qualifying questions, and writes a structured, qualified lead into your pipeline. Small teams lose more revenue to slow follow-up than to lost deals; this removes the gap entirely. Prerequisite: an agreed definition of "qualified" — which, usefully, forces a conversation many sales teams have never had.

3. Admin relief for your service team

Case summaries, interaction recaps, drafting first-response emails for human review. Individually trivial, collectively the difference between your service person spending the afternoon on customers or on typing. This is the lowest-risk starting point because a human reviews everything before it goes anywhere.

The adoption numbers suggest these aren't early-adopter experiments any more: 54% of sellers say they've used AI agents and nearly 9 in 10 plan to by 2027 (Salesforce State of Sales, 7th edition, February 2026 — NZ respondents included), and among SMBs using AI, 91% say it boosts revenue (Salesforce SMB Trends Report, 6th edition).

What It Actually Takes to Go Live

The pricing is the easy half. Here is the practitioner's checklist — the things that determine whether your agent is useful or embarrassing:

1. Clean CRM data. The agent answers from your records. Duplicates, stale fields and half-entered accounts become confidently wrong answers.
2. Knowledge worth grounding in. Written, current, specific. An agent cannot answer from institutional memory.
3. Guardrails and scope. Decide what the agent may do, must never do, and when it hands to a human — before it meets a customer. This design work is the actual skill in agent projects.
4. A human escalation path. Someone owns what the agent hands over, with a service-level promise attached.
5. Privacy done properly. The NZ Privacy Commissioner's AI guidance expects a privacy impact assessment before deployment, transparency with the people whose data is involved, human review of consequential outputs, and assurance that personal information is not retained or used to train the provider's models. Build this in from day one, not after the complaint.
6. A measurement plan. Deflection rate, escalation quality, customer satisfaction on agent-handled enquiries. If you can't measure it, you can't expand it with a straight face.

When Agentforce Isn't Worth It (Yet)

We'd rather tell you this now than after an invoice:

• Your data is a mess. AI on bad data is an amplifier of bad data. Spend the budget on a cleanup first — it improves everything, agent or no agent.
• Nothing is written down. If the answers live in your founder's head, document them first. (You'll need to do this for an agent anyway — but do it as the cheap project, not inside the expensive one.)
• Volume is genuinely low. If your team comfortably answers every enquiry within the hour, an agent solves a problem you don't have. Revisit when growth makes it pinch.
• You're below Enterprise Edition and don't otherwise need to upgrade. Salesforce's own line is that "AI can be added to Enterprise and above" — upgrading purely for an AI experiment rarely pencils out. See our pricing guide for what the upgrade actually costs.

Where to Start

Start smaller than feels ambitious: one agent, one job, four weeks, measured. Use Foundations' included credits so the pilot risks design effort rather than licence spend. If the pilot earns trust, expand it; if it doesn't, you've bought certainty cheaply.

If you want a second pair of eyes first, our free Salesforce health check tells you exactly where your org stands on the readiness checklist above — data, knowledge, and what your edition already includes. And if Agentforce is new to you entirely, begin with our plain-English explainer and the Einstein vs Agentforce comparison.

Frequently Asked Questions

How much does Agentforce cost? As at June 2026, three models: Flex Credits at US$500 (AU$700) per 100,000 credits, where a standard agent action uses 20 credits (about US$0.10); per-conversation pricing at US$2 (AU$2.80) for customer-facing agents; or per-user add-ons at US$125/user/month for unmetered employee use. Salesforce notes prices are subject to change.

Is there a free way to try Agentforce? Yes. If you are on Enterprise Edition or above, the Salesforce Foundations add-on costs $0 and includes Agentforce — Agent Builder, Prompt Builder, 200,000 Flex Credits and 250,000 Data 360 credits — enough to run a genuine pilot before spending anything.

How long does it take to launch a first agent? For a well-scoped pilot on clean data — one agent, one job, existing knowledge content — we typically see a few weeks from kickoff to a supervised launch. Orgs with messy data or no written knowledge base need that fixed first, which is usually the longer half of the project.

When is Agentforce NOT worth it? When your CRM data is unreliable, your processes live in people's heads rather than documentation, or your enquiry volume is so low a human handles it comfortably. Fix the data and write the knowledge down first — both make everything else better even if you never launch an agent.

Do AI agents comply with NZ privacy law? The NZ Privacy Act 2020 applies to AI tools like any other system handling personal information. The Privacy Commissioner's guidance expects a privacy impact assessment before deployment, transparency with customers, human review arrangements, and confirming personal information is not retained or used to train the provider's models.

---

SaaSKool designs and ships Agentforce implementations for NZ SMBs — including the unglamorous data and knowledge work that makes them succeed. If you want the readiness verdict on your org, the health check is free.