Raisely AI Assistant

Raisely AI Assistant is an in-product, AI-powered assistant inside Raisely admin that helps nonprofit teams interpret fundraising performance and decide what to do next. In practice, it works like a “data analyst teammate” you can talk to in plain language: you ask questions about your organisation or campaigns, and it responds with summaries, trends, and recommended next steps.

Raisely AI Assistant is currently in beta, which means it’s ready to use and still actively improving. You may occasionally see an answer that needs review, a result that could be clearer, or a minor issue in the experience. That’s part of the beta process, and feedback from real fundraising workflows will help us make the assistant more accurate, useful, and reliable over time.

The most important expectation to set is that Raisely AI Assistant is designed as a read-only experience. That means it can help you understand what’s happening in your data and suggest actions you might take, but it will not directly change your settings, publish edits, send emails, or modify campaign configuration on your behalf. This is intentional: trust is the product, and the assistant is built to prioritize accuracy, clarity, and permission-safe responses.

What Raisely AI Assistant can do

Raisely AI Assistant is best at taking common “dashboard questions” and turning them into clear answers you can act on immediately. Typical requests include:

• “How many new donors signed up this month?”
• “How much did we raise on [Campaign] during [Date range]?”
• “Show me the latest 5 donations received today.”
• “What is the current progress to goal for [Campaign]?”

See our AI Prompt Library for more examples!

What it cannot do is “click around” in your account like a human or perform actions automatically. If you want something changed, it will typically recommend what to change and where to do it (and may provide wording you can use), but the actual change remains in your control.

For questions specifically about the Raisely product, please continue to use the support chat bot, which is separate from the Raisely AI Assistant.

Permissions and visibility: why the assistant sometimes says “I can’t access that”

Raisely AI Assistant is permission-aware by design. It can only answer using the data it has permission to access, and it should not reveal information you don’t have access to in Raisely already. In beta, access may also be limited to admins or users with the appropriate org-level permissions.

If you ask a question that requires data you don’t have access to (for example, a campaign you can’t view, or a data set restricted to higher-permission roles), the assistant may respond that it can’t access that information. When that happens, the goal is not to be vague. A good “permission-safe” response will:

• Explain that it does not have access to the requested data, and
• Suggest a more limited version of the question that it can answer, or
• Indicate what type of role/permission is required to access the missing information.

How to turn it on (enablement and consent)

To use Raisely AI Assistant, it must be enabled for your organisation.

1. In Raisely admin, open Settings.
2. Navigate to AI tools.
3. Find Raisely AI Assistant and toggle it On.
4. Review the AI Policy presented during enablement.
5. Select Agree & enable to confirm.

If you do not see an AI tools area in Settings, one of two things is usually true: either you are not an admin (or do not have the required permissions), or your organisation has not been enabled for the beta feature set yet. In either case, the next step is to ask an org admin to confirm access and enablement.

How to open the assistant in Raisely admin

Once enabled, you can open the assistant from within Raisely admin. Typically, you’ll see an AI Assistant entry point in the interface (often a button in the bottom-right corner). Clicking it opens a chat panel where you can type your questions. If you close the panel and reopen it, the experience should return you to where you left off.

How to use it well: how to ask questions that get accurate, helpful answers

Because Raisely AI Assistant is optimized for accuracy and relevance, the best results usually come from giving it enough context to answer precisely. You do not need to write a perfect “prompt,” but you should aim to include:

A timeframe. Instead of “How are we doing?”, use “How are we doing this week?” or “Compare May 1–May 31 to April 1–April 30.” If you don’t specify a timeframe, the assistant may choose a default window, which can lead to answers that don’t match what you expected.

A scope. Many questions change meaning depending on whether you’re asking about your whole organisation, a specific campaign, or a subset of campaigns. If you have a scope selector in the assistant (for example, organisation vs. a specific campaign), set it before you ask. If you’re not sure what scope is active, explicitly include it in your question (e.g., “For our Spring Appeal campaign…”).

A clear metric definition when needed. Some terms are used differently across teams and tools. If you ask about “conversion,” you may mean donation conversion, fundraiser activation, or something else. If you get an answer that feels off, a good follow-up is: “What definition are you using for conversion rate here?” This helps align on the same metric meaning.

One primary question at a time for complex analysis. It’s okay to ask multi-part questions, but if you need a detailed analysis, it can be more reliable to break it into steps. For example: first ask for a summary, then ask “why,” then ask for recommended next actions.

🔗 View the AI Prompt Library

Troubleshooting and support

Finding the AI Assistant

If you cannot find the toggle in Settings → AI tools, start by checking permissions. In most cases, you need admin-level access (or a similar org-level role) to see AI settings. If you are an admin and still don’t see it, your organisation may not be enabled for the beta yet. An org admin can confirm whether AI tools are available for your account and, if necessary, contact Support for enablement.

Not seeing the AI Assistant

If AI is enabled but you don’t see the assistant entry point:

1. Refresh the page (hard refresh if needed).
2. Confirm AI is still toggled on in Settings → AI tools.
3. If your browser has aggressive privacy settings, ad blockers, or script blockers, try an incognito/private window to rule out extensions interfering with the UI.
4. If your org recently enabled AI, give it a short moment and refresh again—some UI changes can take a short time to fully reflect.

Data returning errors

This is usually a permissions issue or a scope mismatch. First, confirm you have access to the campaign or data in Raisely normally. Then try a narrower question (for example, one specific campaign you can view, or a simpler metric over a shorter timeframe). If the assistant consistently cannot access data you believe you have access to, capture the exact question you asked and share it with Support so the team can reproduce the issue.

If a result looks surprising, it doesn’t necessarily mean it’s incorrect. It often means the timeframe, scope, or metric definition differs from what you intended. The fastest way to reconcile is:

1. Re-ask with an explicit timeframe (e.g., “From May 1 to May 31…”).
2. Specify the campaign or organisation scope (e.g., “Only the Spring Appeal campaign…”).
3. Ask the assistant to describe the definition it used (e.g., “How are you defining conversion rate here?”).
4. Verify against the corresponding Raisely dashboard/report using the same scope and date range.

If it still appears wrong, please capture: the question, the selected scope (if applicable), the timeframe you intended, and a screenshot of the response. Because the assistant is in beta, these examples are especially helpful for improving accuracy and making future responses more reliable.

Loading errors

Occasional slowness can happen during high usage or temporary system load. If you hit latency or an error:

• Try again after a short pause.
• Ask a smaller question first (narrow scope to one campaign and a shorter date range).
• Avoid stacking multiple comparisons and requests in a single message when diagnosing performance.

If errors persist, include the error message (if shown) and approximate time it occurred when contacting Support.

Raisely AI Assistant now lets you choose which AI model answers your question. You'll find the model selector in the message box, next to the send button — it's set to Auto by default.

Most of the time, you won't need to touch it. But if you're asking something unusually complex, or you want a faster answer, or you're asking the assistant to help you write something, picking a different model can make a real difference to the answer you get back.

This article explains what each model is good at, in the context of the fundraising work you're actually doing.

Choosing an AI Model

Start with Auto. If the answer isn't what you needed, switch models and ask again.

• Auto — everyday questions. Leave it here unless you have a reason not to.
• Claude Sonnet 5 — fast, straightforward answers and long conversations.
• GPT-5.6 Terra — quick multi-step questions that pull from several places at once.
• GPT-5.5 — deep analysis across multiple campaigns or long date ranges.
• Claude Opus 4.8 — the most considered explanations and the best writing.

First, the important bit: your numbers don't change

This is worth being clear about, because it's the question we get asked most.

The assistant does not ask the AI model to do maths. Totals, donation counts, averages, tickets sold, progress to goal — these all come directly from your Raisely data through a fixed calculation, every time, no matter which model you pick.

What the model does affect is everything around the number: how the answer is structured, how well it understands a vague or multi-part question, how clearly it explains what's going on, and how good it is at drafting copy.

So switching models will not turn a wrong number into a right one. If a figure looks off, it's almost always a timeframe, scope, or metric-definition mismatch — see the troubleshooting steps in the main Raisely AI Assistant article.

Also unchanged by your model choice:

• Permissions. Every model can only see the data your Raisely account can see.
• Read-only. No model can change settings, publish edits, send emails, or alter campaign configuration.
• Your data isn't used to train public models.

How to switch models

1. Open Raisely AI Assistant from the button in the bottom-right of Raisely admin.
2. In the message box, click the selector next to the send button (it says Auto by default).
3. Choose a model from the list.
4. Type your question and send it as normal.

Your choice applies to the messages you send after it. You can switch mid-conversation — a useful trick is to ask a question on a fast model, then re-ask the same question on a stronger one if you want a second read on it.

Which model for which fundraising job

The models in more detail

Auto (default)

Auto uses the model your organisation is set up with by default. It's chosen to be a good balance of speed, cost, and answer quality for the questions fundraising teams ask most — campaign performance, revenue and donation metrics, date-range queries, progress to goal.

Use it for: almost everything, until you have a specific reason to change.

Claude Sonnet 5

Quick and dependable. Sonnet 5 is a good pick when you know roughly what you want and just need it back promptly, and it copes well with long conversations where you're building on earlier answers.

Use it for: straightforward lookups, quick checks between meetings, long back-and-forth sessions, fast content drafts you plan to edit anyway.

Less suited to: questions with lots of moving parts where you want the reasoning shown.

GPT-5.6 Terra

The efficient middle ground. Terra is good at answering a question that needs the assistant to look in a few different places and then bring it together, without a long wait.

Use it for: multi-campaign comparisons, "what changed week on week" questions, anything where you want breadth quickly.

GPT-5.5

Our strongest option for multi-step reasoning. GPT-5.5 is the one to reach for when your question really has three questions inside it, or when you want the assistant to work carefully through a chain of "and then why" rather than give you a headline.

Use it for: performance investigations, longer date ranges, layered donor questions, anything where you'd otherwise be building a spreadsheet.

Trade-off: noticeably slower than Auto or Sonnet 5. Ask, then go do something else for a moment.

Claude Opus 4.8

The most thoughtful model in the list, and the best writer. Opus 4.8 is particularly good at explaining fundraising results in plain language to an audience who wasn't in the data with you — a board, an ED, a major donor, a funder.

Use it for: board and leadership summaries, narrative explanations of a result, donor and supporter comms where the wording genuinely matters, sense-checking a conclusion you're not sure about.

Trade-off: the slowest of the options. Save it for when the quality of the words or the depth of the reasoning is the point.

Tips for getting good answers, whichever model you choose

Model choice helps at the margins. Giving the assistant good context helps far more.

1. Always give a timeframe. "How are we doing?" is much weaker than "How are we doing from 1 June to 30 June?"
2. Set your scope. Use the Discussing picker in the message box to choose the organisation or a specific campaign before you ask.
3. Ask one main thing at a time for anything complex — get the summary, then ask why, then ask what it means.
4. If a number surprises you, ask about the definition. "How are you defining a new donor here?" resolves most surprises immediately.
5. Re-ask on a different model if an answer feels thin. It's a legitimate way to get a second opinion, and it costs you nothing.

Which model should we set as our default?

You don't need to decide anything at the organisation level — Auto handles that, and we tune it as models improve. The picker is there for the moments where you want something specific: faster, deeper, or better written.

Raisely AI Assistant is in beta, and the model picker is part of how we're learning what works for fundraising teams in real use. If you find a model that's consistently better (or worse) for a particular kind of question, please tell us — use the thumbs up or thumbs down on any response, or contact support. That feedback directly shapes which models we offer and what Auto defaults to.