For most small charities, Agentforce for Nonprofits is not worth building yet unless your supporter records are already clean in Salesforce and a staff member has about two hours a week to own it. The free licences remove the seat cost, but agent usage is metered: Salesforce lists a $0.10 list-price action, and nonprofit rates are not published.
The decision turns on three things: data quality, a named owner, and a first job with enough volume to repay the effort. A three-person charity answering 1,500 supporter chats a year would spend about $600 on agent credits at list price and roughly $3,500 on its own staff time, so the people cost, not the licence, decides whether it pays.
What Salesforce publishes, and what it leaves out
Salesforce's own pages are written to sell, and the details that matter for a small charity are scattered across several of them, so the list below pulls them together. This is what I could verify on its pricing pages in October 2026.
- Free licences: its nonprofit pricing page says eligible nonprofits get 10 Nonprofit Cloud Core licences at no cost, with Premier Success activating on the first paid purchase. The same page lists "Agentforce for Nonprofit" and "Agentforce Coworker" in its feature comparison, so the Agentforce branding and the Nonprofit Cloud licence names overlap. Ask which product and features the free licences actually switch on.
- Paid editions: the same page lists nonprofit prices per user per month, billed annually: Sales and Service Core $70, Advanced $125, and Sales Max and Service Max $285 each. Additional licences beyond the free ten are discounted for eligible organisations.
- Credits included: the page lists Flex Credits per organisation per year: 500,000 for Core, 1 million for Advanced and 2.75 million for the Max editions.
- Agentforce itself: the nonprofit page lists it in its feature comparison but states no price. Salesforce's general Agentforce pricing page offers Flex Credits at $500 per 100,000, with an agent action using 20 credits and a voice action 30; conversations at $2 each; an Agentforce user licence at $5 a user a month that requires credits; and add-ons at $125 or $150 a user a month. It says unused credits do not roll over, and that credits and conversations cannot be used in the same org.
- Nonprofit discounts on Agentforce: the Agentforce pricing page does not mention nonprofit pricing, and several Agentforce add-ons on the nonprofit page are marked only "Available for purchase" with no price.
What is missing: the credit allowance, if any, that comes with the free licences; the nonprofit price of extra credits; and the cost of the Premier Success tier once you buy anything. Put those three questions to your Salesforce contact in writing before you plan.
Turning list prices into a yearly cost
A worked example for an imaginary charity with three staff, 6,000 supporter records and a plan to answer routine supporter questions with an agent. Every figure is illustrative and based on the list prices above; your nonprofit rate may be lower.
| Line | Assumption | Yearly cost |
|---|---|---|
| Licences | 3 of the 10 free licences used | $0 |
| Agent actions | 1,500 supporter chats a year, 4 actions each = 6,000 actions at 20 credits = 120,000 credits | about $600 at $500 per 100,000 |
| Set-up time | 40 hours of a staff member at $25 an hour | $1,000 |
| Running time | 2 hours a week for 50 weeks at $25 | $2,500 |
| Support or partner help | None assumed | $0 |
| Total in year one | about $4,100 |
The metered cost is small. The people cost is the real bill: $3,500 of the $4,100 is staff time. If the agent handles 1,500 chats that would have taken five minutes each, that is 125 hours, worth $3,125 at the same rate. In this case the project roughly breaks even in year one, and gains in year two once set-up is done. Change the volume to 300 chats and the saving falls to about $625 against the same running cost: a loss.
Using the $2 conversation option instead, 1,500 conversations would be $3,000, five times the credit route in this example. Which model is cheaper depends on how many actions your agent takes per conversation, so ask Salesforce to price both for your use case. Also check the credit pool in your edition: 500,000 credits would cover about 25,000 actions at 20 credits each, which dwarfs the example, but only if your edition includes it.
What a first agent can do, and what it should not
A good first agent does one job, answers from documents you control, and hands over to a person when unsure. For a small charity, three jobs fit that description.
Job 1: answering routine supporter questions
Questions such as "How do I change my monthly gift?", "Is my donation tax-receipted?" or "How do I cancel?" have fixed answers. An agent grounded in your published policy pages can answer these around the clock and pass anything else to a staff member. The step-by-step version is in the first service agent tutorial. Test it with 30 real past questions before it speaks to anyone: if it answers fewer than 27 correctly, fix the source documents first.
Job 2: summarising a supporter's history for staff
Before a call, a fundraiser can ask the agent for a short summary of a donor's giving, events attended and past notes. This is lower risk because a person reads and checks the output, and it saves minutes per call. It only works if your records are tidy: a duplicate donor or a missing gift will produce a confident wrong summary.
Job 3: drafting follow-up emails
Draft thank-you or lapsed-donor emails from records and approved templates, to be reviewed before sending. Treat each draft as a first attempt, and keep a person approving every message that mentions an amount.
Jobs to avoid at first
- Anything involving safeguarding, vulnerable supporters or complaints, where a person should respond.
- Refunds or changes to a donor's payment, where an error costs trust.
- Voice agents, which cost more per action and need more testing.
- Agents that write to records without review: start read-only.
Writing the agent's instructions, with a sample
Whatever tool you use, an agent behaves according to the instructions you give it. A short, strict set beats a long, clever one. An illustrative instruction block for a charity's supporter-questions agent:
You answer questions from supporters of [charity name].
Use only the approved documents provided. If the answer is not in them, say you are not sure and offer to pass the question to the team.
Never state a donation amount, a date or a fee unless it appears in the documents.
Never give tax or legal advice; say that the supporter should check with an adviser.
Never discuss safeguarding, complaints or a supporter's health; hand over to a person at once.
Keep replies under 80 words, friendly and plain.
Always end with: "If that doesn't answer it, reply and a member of the team will help."Then test it with awkward questions, not friendly ones. Illustrative results from a test of ten:
- "How do I change my monthly gift?" correct, pointed to the right page.
- "Is my gift tax deductible?" correctly declined to advise and suggested an adviser.
- "What is your refund policy for event tickets?" wrong: it blended a policy for a different event. The fix was to split the policy documents by event type and retest.
- "I think my elderly mother was pressured into donating." correctly handed over to a person immediately.
The failure in the third case is typical: when documents overlap or contradict, the agent invents a blend. Clean the sources first and the instructions have much less to do.
A before and after for a stored answer
Before: "Our giving options are varied and flexible, and we encourage supporters to consider what suits them."
After: "You can give once or monthly, by card or direct debit. To change or stop a monthly gift, use the link in your last receipt email, or reply here and a team member will do it within two working days."
The second contains the facts and the next step; the first would produce an agent that talks without answering.
Three small charities, three different answers
The verdict changes with the organisation. Three illustrative cases:
A food bank with two part-time staff and volunteers
Supporters mostly ask where and when to donate food, and the answers change weekly. Records are small and mostly in a spreadsheet. Moving to Salesforce for an agent would be a larger project than the problem deserves. A shared assistant with a weekly updated notice and a templated reply covers it. Verdict: not worth it.
A youth education charity with a membership scheme and 5,000 records
The charity already runs its supporter records in Salesforce and receives 400 repeat questions a month on membership renewals and receipts. A staff member has already tidied the data. A narrowly scoped agent on renewal questions, with a human handover, could free around 20 hours a month. Verdict: worth a pilot, with a cap on credits and a 60-day review.
A small environmental group preparing for a fundraising campaign
Staff want AI to write donor emails and summarise past giving. This is a drafting and reading job, not an agent job. A chat assistant and a clean export of donor records give most of the benefit at a fraction of the cost, and the group can revisit an agent after the campaign. Verdict: start cheaper.
What separates the three is not the size of the charity but the volume of repeat questions and the state of the data.
A short list of questions to put to Salesforce
Send these in one email and keep the reply, because verbal assurances disappear when staff change.
- Which product and edition do the 10 free licences cover, and what is its current name?
- Are any Agentforce credits or conversations included with the free licences? If so, how many, and do they reset yearly?
- What is the nonprofit price for additional credits, and does it differ from the $500 per 100,000 list price?
- What does Premier Success cost once it is activated, and can we avoid activating it?
- What happens to our data, credits and licences if our nonprofit status is re-verified or lapses?
- Can we cap usage so a testing mistake cannot create a large bill?
- Can we export our data and agent configuration if we leave?
A vendor that answers all seven clearly is easier to trust than one that offers a demo instead. The same list is a good template for any vendor, including a partner quoting for the set-up.
The data question that decides everything
An agent reads your records. If your supporter data has duplicates, missing consent flags or three spellings of one donor, the agent will pass those errors on faster than a person would. Before spending anything, run a short audit on a sample of 100 records:
- How many are duplicates of another record?
- How many have a current consent or contact preference?
- How many gifts are linked to the right person?
- When was the record last updated?
Illustrative result: 11 duplicates, 18 with no consent flag, 4 with gifts attached to a partner's record. That is not agent-ready. Spend a month on clean-up, which helps every future tool, then return. The tutorial on cleaning messy data covers the method, and the clean-up is useful even if you never buy an agent.
Who will own it
Name the person in writing, with a deputy for holidays, before the pilot starts.
An agent is a small piece of ongoing work, not a one-off install. A named person needs to review a sample of conversations weekly, update the source documents when policies change, and decide what to do about questions the agent could not answer. Without that, answers go stale: a changed event date, a new fee, an altered address. In a charity with three staff, that is a visible slice of someone's week. If every person is already over capacity, a free tool that nobody maintains is not free.
Cheaper options to try first
Several vendors price AI for nonprofits at seat prices of a few dollars a month. Verify each on the vendor's page, because eligibility rules and prices change and some differ by country.
- A shared chat assistant: OpenAI and Anthropic both run nonprofit programmes; the recorded seat prices were about $8 per user per month, but the OpenAI price is one to confirm. See the OpenAI nonprofit eligibility check.
- Google Workspace for Nonprofits: includes the Gemini app and Gemini Notebook at no extra cost, where you qualify.
- Microsoft's nonprofit Copilot add-on: listed at $25.50 a user a month, which is more than the commercial equivalent, so check before assuming nonprofit means cheaper.
- A lighter CRM with built-in AI: the cheaper alternatives to Salesforce comparison lists seven.
A concrete low-cost test: collect the 30 most common supporter questions, write approved answers in a document, and give them to a shared assistant as a reference. Staff paste in an incoming question and review the suggested reply. After a month, count how many questions were repeated and how much time it saved. If the answer is "a lot", that is your evidence for building an agent. If the answer is "almost none", you have saved yourself a project.
A decision rule you can score in ten minutes
Give one point for each yes.
- Your supporter data passed the 100-record audit with fewer than five problems.
- You already use Salesforce for supporter records, so no migration is needed.
- One person has two hours a week for the agent.
- You get at least 100 repeat supporter questions a month.
- You have confirmed in writing what the free licences include and the price of extra credits.
- You have a plan for a human to handle anything the agent cannot.
Five or six: worth a pilot. Three or four: fix the gaps and test with a cheaper assistant first. Two or fewer: not yet. The guide to Salesforce AI tiers helps you place your edition.
Mistakes charities make with a first agent
- Planning around the free licences alone: the licences are not the main cost; staff time and credits are.
- Letting the agent loose on all topics: start with one narrow topic and widen when answers are reliable.
- No fallback path: if the agent cannot answer, the supporter must reach a person quickly, with the conversation attached.
- Skipping a consent check: check that your consent records allow the contact the agent will make.
- Buying help before scoping: if you hire a partner, use the partner, freelancer or DIY comparison and ask for a fixed scope.
If you want an independent opinion on whether Agentforce or something simpler fits your charity, the contact page explains how a 1:1 call works.
How to review the pilot after 60 days
Set the success test before you start. Illustrative targets for a three-person team: the agent answers at least 60 per cent of routine questions correctly without help, escalations reach a person within one working day, and staff time on routine supporter replies falls by at least four hours a month. At day 60, compare actual figures, read twenty conversations for tone and accuracy, and check the credit usage against your estimate. If two of three targets are met, extend with a second topic. If not, either improve the source documents and retry for a month, or stop and keep what you learned. Keep the day-60 figures with the Salesforce email replies: together they are the case for, or against, renewing.
Questions charities ask about Agentforce
Is Agentforce for Nonprofits free?
Not entirely. Salesforce's Power of Us programme gives eligible nonprofits 10 licences at no cost, but agent usage is metered separately by credits or conversations, and extra licences and the support tier are paid. Ask in writing exactly what the free licences include before you plan around them.
What does one Agentforce action cost?
Salesforce's pricing page lists Flex Credits at $500 per 100,000 credits, with an action using 20 credits, which works out at $0.10 an action at list price. Nonprofit discounts are not shown on that page, so ask your account contact for the nonprofit rate.
Can a charity with two staff run it?
Technically yes, but someone must own the data, the agent's instructions and the weekly review. If nobody has about two hours a week for that, the agent will drift out of date. A small team often does better with simpler tools first.
What is the cheapest way to try AI on supporter questions?
Start with a shared assistant seat and a saved set of answers, using nonprofit pricing from OpenAI, Anthropic or Google where you qualify. It lets you test what supporters ask without building anything in a CRM. Move to an agent when the questions are repetitive and the data is clean.
Do I need a Salesforce partner?
For a first agent on clean data, a capable staff member can often do it, with Salesforce's free training. If your records are messy, or you need integrations, a partner or consultant can save time, but ask for a fixed scope and a handover you can run yourselves.
Further reads
- How to Evaluate an AI Implementation Proposal or Quote — Judge a partner's quote before you accept it.
- AI Tool Discounts for Charities and Non-Profits: What to Claim — Discounts you can claim from other vendors.
- AI Consultant for Nonprofits: What to Hire For and What to Pay — What to hire for and what to pay.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Salesforce nonprofit pricing page (Power of Us licences, Nonprofit Cloud editions and Flex Credit allowances) and Agentforce pricing page (Flex Credits, conversations, user licence, add-ons), both fetched October 2026. Eligibility and prices vary; confirm with Salesforce.