Quick start
- Make sure your organization has Plus, or that the owner has added your own Anthropic or OpenAI key in Settings › Plan.
- Open a matrix that has finished building and choose Ask AI at the top. A panel opens on the right.
- Type a question such as “Which segments should our year-end appeal focus on?” and choose Ask.
- For leadership, choose Write a summary for leadership in the same panel, then copy it into your memo.
- On a matrix's History tab, choose What changed? to explain the difference between two snapshots.
What the AI assistant does
The AI assistant reads a matrix's totals and settings and helps you read and act on them.
| Task | Where to find it | What you get |
|---|---|---|
| Ask AI | Ask AI at the top of any matrix | Answers to questions about that matrix, as a conversation |
| Summary for leadership | Write a summary for leadership in the Ask AI panel | 120–180 words ready to paste into a memo |
| What changed? | Under Compare two snapshots on a History tab | A short explanation of what changed between two snapshots |
The assistant never changes your matrices. Its answers aren't saved; copy what you want to keep.
Who can use it
- With Plus: everyone in the organization shares 150 requests a month. See your monthly allowance.
- With your own API key: on Free or Plus, once the owner adds an Anthropic or OpenAI key, everyone can use the assistant on that key, without a monthly limit from us. See using your own API key.
- On Free without a key: Ask AI carries a Plus tag, and asking answers “The AI assistant is part of Plus.”
Ask AI
Ask AI is a conversation about one matrix. Ask a question, then follow up; the assistant remembers what you've just discussed in the panel. Answers are a few short paragraphs of plain text, usually under 200 words. “Thinking…” shows while it works.
Questions that get good answers
- “Which segments hold most of our value, and which are slipping?”
- “What should we do differently for At risk and Can't lose them?”
- “How many new donors or customers do we have, and how do we keep them?”
- “Explain what R5 · FM2 means to a board member.”
- “Our frequency scores are uneven. Why, and should we use fixed breakpoints?”
- “Plan a three-step win-back for Hibernating that doesn't cost much.”
It can't help with questions about particular people, because it never sees individual records. For that, open View records.
Summary for leadership
Write a summary for leadership writes 120 to 180 words in two or three plain paragraphs: how the file breaks down, the main findings (where value is concentrated, which groups are healthy or slipping, movement since last year when shown), the main risks, and the assumptions behind the numbers (window, quintiles or fixed breakpoints, as-of date, which transaction types count). Write it again for a fresh version; each is one request.
What changed?
On a History tab, What changed? explains the difference between the From and To snapshots in 60 to 150 words, leading with the change that matters most, and says when the groups themselves changed between the two.
Your monthly allowance
With Plus, your organization gets 150 AI requests a month, shared by the whole team. You'll see how many are used in Settings › Plan, in the Ask AI panel and beside What changed?, as a line like:
AI requests this month: 38 of 150 · resets Nov 1
The count starts again on the 1st of each month (UTC). Unused requests don't carry over.
What counts as a request
| Action | Requests |
|---|---|
| Asking a question in Ask AI | 1 |
| Writing a summary for leadership | 1 |
| What changed? on the History tab | 1 |
| Opening the panel or copying an answer | 0 |
| Any request on your organization's own API key | 0 |
| A request that fails on the AI side | 0 |
When the allowance runs out
Once 150 requests have been used, each AI feature answers “You've used this month's 150 AI requests. They start again on Nov 1.” Nothing else is affected. The owner can add your own API key to keep going.
Requests that fail are given back
If the AI side can't answer — it's busy or overloaded, can't be reached, declines, or cuts its reply off — the request is given back and you see a short reason. Try again in a minute.
Using your own API key
- Only the owner can add, replace or remove the key. Members see whether a key is in use and which model.
- The key is encrypted when it's saved and is never shown again, to anyone.
- While a key is saved, every AI request in the organization uses it, including on Plus. Requests on it don't count toward Plus's 150, and their cost is billed to your Anthropic or OpenAI account.
- Your own key unlocks the AI assistant only. Scheduled CRM sync and matrix history still need Plus.
Models you can choose
| Provider | Model | Note in the app |
|---|---|---|
| Anthropic | Claude Opus 5.5 | Most capable; what Plus uses |
| Anthropic | Claude Sonnet 5.5 | Faster and cheaper |
| Anthropic | Claude Haiku 5.5 | Fastest and cheapest |
| OpenAI | GPT-6 Astra | OpenAI's most capable |
| OpenAI | GPT-6.1 Sol | Balanced cost and quality |
| OpenAI | GPT-6 Luna | Cheapest |
Adding a key
- Create an API key in your organization's Anthropic or OpenAI account.
- As the owner, go to Settings › Plan › Your own AI key and choose Add a key.
- Choose a Model, paste the key into API key and choose Save key. We check the key with a free lookup of that model before saving it.
If the key isn't accepted, the reason shows above the form: the key was rejected, can't use that model, or the provider couldn't be reached.
Replacing a key, changing the model, or removing it
Replace key or change model opens the same form; paste the key again, since we never show or reuse the saved one. Remove key puts the assistant back on Plus, or off on Free. If your provider later rejects the key, AI requests say so; replace or remove it.
What the AI can and can't see
The assistant only ever sees aggregate numbers and settings, never IDs, names or transaction records. The description of a matrix we send has no place for an ID or a row, so one can't slip in.
What's sent
- The matrix's name and type, and its as-of date.
- Each group's count and value: every segment, every composite score for a full matrix, or every cell of a grid.
- The assumptions: window, which transaction types count, each measure's breakpoints and whether they're quintiles or fixed.
- How concentrated value is (the top 20% share) and, for named segments, how many stayed, moved or are new since a year before, with the largest moves between segments.
- For Ask AI, your question and the earlier questions and answers in the panel.
- For What changed?, the same for each snapshot plus the movement counts between them. Matching people happens inside TrueRFM; only counts are sent.
What's never sent
- IDs, individual transactions, any person's scores or segment.
- Values of your kept detail columns.
- Names, emails, phone numbers or addresses — we never store those. See Data, privacy and deleting.
Note: Names you type (the matrix's name, renamed segments) are sent as written. Don't put a person's name in them.
Where requests go
- Plus: to Anthropic, using Claude Opus 5.5 on TrueRFM's account.
- Your own key: to Anthropic or OpenAI on your organization's account, with the model the owner chose. Requests to OpenAI are sent with response storage turned off.
We don't use your data to train AI models. See our Privacy Policy.
Limits and good habits
- The assistant works only from the numbers on the page. It doesn't know about a recent conversation, a major life event or a campaign you haven't run yet.
- Check numbers it quotes against the matrix before you share them.
- Treat its suggestions as a starting point for your team, not a decision.
Questions
Can the assistant see our donors or customers?
No. It sees totals per group only, never IDs or records.
Is the conversation saved?
No. Closing the panel or reloading the page starts a new conversation.
We ran out of requests. What now?
Wait for the 1st of next month, or ask the owner to add your own API key.
Which model answers?
Claude Opus 5.5 on Plus, or the model the owner chose for your own key.