AI assistant

What the AI assistant does

The AI assistant reads a pyramid's totals and settings and helps you plan with them. It works in four places:

Ask AI, Draft with AI and Write summary are on gift pyramids, including the gift pyramid of each campaign priority. Donor pyramids and True Pyramids use the assistant only through What changed? on their History tab.

The assistant never changes your work on its own. Drafts land in a separate scenario you can keep or delete, and summaries are yours to edit or remove.

Who can use it

• With Plus: everyone in the organization, owners and members alike, 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 in the organization can use the assistant on that key, without a monthly limit from us. See using your own API key.
• On Free without a key: the AI buttons show a Plus tag. On a gift pyramid, clicking one opens a short card with Upgrade to Plus — $29/month and Use your own API key; both take you to Settings › Plan. On the History tab, What changed? answers “The AI assistant is part of Plus.”

Ask AI

Ask AI is a conversation about one gift pyramid. Use it to read the chart, test an idea, or work out what to do about a level that's short.

1. Open a gift pyramid and pick the scenario you want to talk about. The assistant reads the scenario that's open.
2. Choose Ask AI. The panel's heading shows the pyramid, the scenario and your AI usage, for example “About Capital campaign · Current · AI requests this month: 12 of 150 · resets Nov 1”.
3. Click one of the Try asking suggestions, or type in Ask about this pyramid and choose Send (or press Enter).
4. “Thinking…” shows while the answer is written. Ask a follow-up and the assistant remembers what you've just discussed.

The three suggestions are Is $2.5M realistic for us? (with your own goal), Which levels are hardest to fill? and Why is the top of the pyramid short?. Answers are usually a few short paragraphs of plain text, under 200 words.

Questions that get good answers

The assistant knows gift pyramid conventions (a lead gift near 20% of the goal, the top 10 gifts bringing in at least half, prospect ratios of 2:1 to 5:1) and your pyramid's actual numbers. Ask about the numbers in front of you:

• “Level 3 is short by 6 prospects. What are our options?”
• “Is a 3:1 ratio realistic at our $100,000 level?”
• “How far is our concentration from the 90/10 target, and what would close the gap?”
• “Which check are we failing, and does it matter?”
• “What would change if we counted planned gifts toward levels?”
• “Explain the Short column to a board member in two sentences.”
• “Our largest reach is $150,000. Should the lead gift still be $500,000?”

It can't help with questions about particular people (“Who should we ask for the lead gift?”), because it never sees individual records. For that, open the level's prospect list with View records or work in the Prospect plan. If you ask something unrelated to gift pyramids or campaign planning, it says briefly what it can help with.

How the conversation works

• Each question you send, including a click on a suggestion, is one AI request.
• With each question, the assistant also gets your last six questions and its answers from the panel, so follow-ups like “And for level 4?” work.
• It always reads the pyramid as it is now. If you change a level and ask again, the answer uses the new numbers.
• The conversation isn't saved. Closing the panel keeps it while you stay on the page; reloading or leaving the page starts a new one. Copy anything you want to keep.
• The question box takes up to 1,000 characters.

Draft with AI

Draft with AI gives you a starting point: a set of gift levels, or a prospect rule, built from your pyramid's goal and numbers. It never overwrites your work. Each draft is copied into a new scenario, so you can compare it with your own before deciding.

Drafting levels

1. Make sure the pyramid has a goal. Without one you'll see “Set a goal before asking AI to draft levels.”
2. Choose Edit levels. If you've just changed levels, choose Save levels first and open the panel again.
3. Choose Draft with AI at the top of the panel. The button reads “Drafting…” while it works.
4. The page reloads on a new scenario tab called AI draft (with a small AI tag), showing the drafted levels.

The drafted levels follow these rules:

• 3 to 12 levels, largest gift first, each a different amount.
• Round, whole-dollar amounts ($250,000, not $247,300).
• A ratio (multiplier) from x1 to x10 at each level, usually 2 or 3 at the top and 3 to 5 lower down.
• The levels add up to no more than the goal; the rest is left to the “many” row of smaller gifts, typically 5% to 20% of the goal.
• If the pyramid has a gift file, the assistant takes into account what it can support. When your largest reach is well below a 20% lead gift, it either sizes the top to what the file can reach or keeps an aspirational lead gift, and says which.

The draft's template shows as Custom levels. The rest of the scenario (goal, prospect rule, concentration target, planned-gift setting) is copied from the scenario you were on. If the pyramid has a gift file, prospects are counted for the draft straight away.

Tip: Draft with AI works from the scenario as last saved, not from unsaved changes in the open panel. Save first if you want the draft to start from your edits.

Drafting a prospect rule

1. On the gift pyramid, choose Edit rule to open the Prospect rule panel.
2. Choose Draft with AI, beside Add condition.
3. A new AI draft scenario opens with the drafted rule. If the pyramid has a gift file, prospects are counted again with it.

A drafted rule has 1 to 4 conditions, joined by any (broader) or all (narrower), chosen from:

• Has a capacity rating at or above the level's gift.
• Gave at least a percentage of the level's gift over the last so many years.
• Gave to a fund, but only funds that actually appear in your gift file.
• Gave within the last so many years.
• Made at least a number of gifts.

It also decides whether Count planned gifts toward levels should be on. It starts from your current rule and its results: if levels are short, it may broaden the rule; if counts look more generous than your real relationships, it may narrow it. Conditions on your other kept detail columns are never drafted, because the assistant can't see those columns. See the prospect rule for what each condition means.

Keeping or discarding an AI draft

Above the table, an AI draft shows a notice: AI draft., the assistant's one-sentence reason for its main choice, then “Nothing in your other scenarios changed. Compare it before you use it.”

• Compare it: choose Compare scenarios to see the draft beside your own scenarios. See scenarios.
• Keep it: there's no separate apply step. The draft is a full scenario: switch to it, edit its levels or rule like any other, rename it from the scenario's ⋯ menu, and export it.
• Discard it: choose Delete draft in the notice, or Delete from the scenario's ⋯ menu.

Each new draft gets its own scenario: AI draft, then AI draft 2, AI draft 3 and so on. Duplicating an AI draft gives you an ordinary scenario without the AI tag.

Summary for leadership

Write summary turns a gift pyramid into two or three plain paragraphs (120 to 180 words) ready to paste into a board memo or campaign committee update. It covers where the plan stands (the goal, the lead gift, the top of the pyramid, concentration), the main findings, the main risks (short or unfillable levels, checks not met), and the assumptions behind the numbers (the prospect rule, the ratios, the as-of date and how planned gifts are treated). If there's no gift file yet, or prospects haven't been counted, it says the prospect picture isn't known yet rather than guessing.

1. Open the gift pyramid on the scenario you want summarized.
2. Choose Write summary at the top of the page. It reads “Writing…” while it works.
3. The Summary for leadership card appears above the goal, marked “Written by AI from this pyramid's totals. Check it before you share it.”

Editing, copying and exporting the summary

• Edit: type straight into the card. Your changes save as you type. A summary can be up to 4,000 characters.
• Copy: copies the text, ready to paste into an email or document.
• Include in Print and Excel exports: ticked, the summary goes into Print, and Export Excel adds a Summary for leadership sheet. Unticked, Export Excel leaves it out. See Excel export.
• Write it again: writes a fresh summary from the pyramid as it is now. One request.
• Remove: deletes the summary from the pyramid.

A pyramid has one summary, whichever scenario it was written from. It stays on the pyramid if your organization's Plus ends; you can still edit, copy, export or remove it.

Careful: Write it again and Write summary replace the whole summary, including any edits you made. Copy your edited version first if you want to keep it. Remove has no undo.

What changed? on the History tab

On any pyramid's History tab, pick two snapshots under Compare two snapshots and choose What changed?. The assistant explains the most important change in 60 to 150 words: tiers that grew or shrank, donors who moved up or down, lapsed and new donors and revenue for donor pyramids and True Pyramids; or the goal, levels, prospect rule, counts and checks for a gift pyramid. It says so when the tiers themselves changed between the snapshots, or when the cause can't be told from the numbers. The explanation shows under the button with Copy; it isn't saved. If nothing in the numbers changed, you'll see “Nothing in this pyramid's numbers changed between these two versions.” See What changed? in Pyramid history.

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 (with a meter), in the Ask AI panel, under a leadership summary, 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. Months are counted in UTC, so in US time zones the count resets in the afternoon or evening of the last day of the month. Unused requests don't carry over.

What counts as a request

When the allowance runs out

Once 150 requests have been used, each AI feature you try answers “You've used this month's 150 AI requests. They start again on Nov 1.” (with the 1st of next month). Nothing else in the app is affected. To keep going before then, the owner can add your own API key; requests on it don't count toward the 150.

Requests that fail are given back

If the AI side can't answer (it's busy or overloaded, can't be reached, declines the request, cuts its reply off, or sends back a draft that can't be used), the request is given back and doesn't count. You'll see a short reason: in the Ask AI panel as the answer, at the top of the pyramid page for drafts and summaries, or under What changed?. Try again in a minute. See messages you might see.

Using your own API key

Instead of, or as well as, Plus, your organization can connect the assistant to its own Anthropic or OpenAI account and pick the model. This is how a Free organization can use the assistant, and how a Plus organization can go beyond 150 requests or use a different model.

• 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 your key don't count toward Plus's 150, and their cost is billed to your Anthropic or OpenAI account.
• The AI lines in the app then read, for example, “Using your OpenAI key · GPT-6 Astra” instead of the monthly count.
• Your own key unlocks the AI assistant only. Scheduled CRM sync and pyramid history still need Plus.

Models you can choose

The model list shows each model with a short note, exactly as below. The provider follows from the model you pick, so choose a model from the same company as your key.

Pick the most capable model if quality matters most, for example for leadership summaries and drafts. Pick a cheaper one if your team asks a lot of quick questions and you're watching the provider bill. Every model gets exactly the same instructions and the same pyramid totals.

Adding a key

1. Create an API key in your organization's Anthropic or OpenAI account.
2. As the owner, go to Settings › Plan › Your own AI key and choose Add a key.
3. Choose a Model from the list.
4. Paste the key into API key and choose Save key.
5. We check the key with a free lookup of the model you chose (it costs nothing) before saving it. When it works you'll see, for example, Your Anthropic key is in use · Claude Opus 5.5 · added Oct 26, 2026.

If the key isn't saved, the reason shows above the form:

Replacing a key, changing the model, or removing it

• Replace key or change model: opens the same form. Choose the model and paste the key again (even if only the model is changing, since we never show or reuse the saved key), then Save key.
• Remove key: asks “Remove your AI key? The assistant goes back to Plus, or off on Free.” After removing it, Plus organizations go back to the 150-request allowance; Free organizations lose the assistant.

Note: If your provider later rejects the key (it was revoked, or the account ran out of credit), AI requests show “Your organization's AI key was rejected. The owner can check it in Settings › Plan.” They don't fall back to Plus's allowance while the key is saved; replace or remove the key.

What the AI can and can't see

The assistant only ever sees aggregate numbers and settings, never constituent IDs, names or gift records. This is built into how requests are put together: the description of a pyramid that we send has no place for an ID or a row, so one can't slip in.

What's sent

For Ask AI, Draft with AI and Write summary on a gift pyramid, the assistant gets this about the scenario that's open:

• The pyramid's name and the scenario's name.
• The goal.
• Whether a gift file is attached and, once prospects are counted, its as-of date (the date of the latest gift in the file) and the largest reach (the largest level gift anyone in the file could qualify for).
• The prospect rule in words, and whether planned gifts count toward levels.
• Fund names from your gift file, up to 40, when it has a kept detail column whose name includes “fund” (so a drafted rule can use them).
• For each level: number of gifts, gift amount, level total, running total and share of the goal, ratio, prospects needed, prospects identified, how many are short, how many reach it only through planned gifts, and whether it's unfillable.
• The “many” row's total, total prospects needed and identified, the concentration target and actual, and each check, met or not, with its detail.

Ask AI also sends your question and the last six questions and answers from the panel.

For What changed?, the assistant gets the same kind of description for each of the two snapshots. For a donor pyramid or True Pyramid that's the tier names and descriptions, each tier's count and revenue (and, for a True Pyramid, how many look ready for the tier above), the period or as-of date, and counts of how many moved up, stayed, moved down, were new, reactivated or lapsed. For a donor pyramid it also gets how many donors fit no tier and the movement counts the pyramid itself showed at the time. The matching of people between the two snapshots happens inside TruePyramids; only the counts are sent.

What's never sent

• Constituent IDs.
• Individual gifts or rows from your uploads or CRM, and any individual's dates, amounts or tier.
• Values of your kept detail columns, other than the fund names above.
• Names, emails, phone numbers or addresses. We never store those in the first place: personal-information columns are removed when a file is first read, and the raw file is deleted after processing. See Data, privacy and deleting.
• Prospect plan assignments, your team, or your organization's billing.

Note: The names you type (the pyramid's name, scenario names, renamed tiers) are sent as written. Avoid putting a donor's name in a pyramid or scenario name.

Where requests go

• Plus: to Anthropic, using Claude Opus 5.5 on TruePyramids' account.
• Your own key: to Anthropic or OpenAI on your organization's account, with the model the owner chose, under your agreement with that provider. Requests to OpenAI are sent with OpenAI's response storage turned off.

We don't use your donor data to train AI models. See our Privacy Policy.

Limits and good habits

• It can be wrong. Treat everything it writes as a draft. Check figures against the pyramid before you share them; the panel and the summary card both remind you.
• It only knows the numbers. It can't see relationships, timing, board dynamics or who's already been asked. You know things it doesn't.
• Drafts are starting points. A drafted rule is something to test: watch how the prospect counts change before you rely on it.
• Answers vary. Asking the same thing twice can give different wording or suggestions. If an answer seems off, ask it to explain, or check the numbers yourself.
• Plain text. The assistant is asked to write without headings, tables or bold, so answers paste cleanly into email.
• Thin data, cautious answers. With no gift file, or very few prospects counted, it says it can't draw conclusions instead of guessing.

Advanced: how requests are built and checked

What the assistant is told

Every request starts with the same instructions: how gift pyramids work (levels, the lead gift near 20% of the goal, the top 10 gifts at least half, concentration targets such as 90/10 or 80/20, gifts of $25,000 or less ideally under 30% of the goal, ratios of 2 to 5 per gift, tighter at the top), what “identified”, “short”, “planned”, “unfillable”, “as of” and “largest reach” mean, that it can only see the aggregate block and must never invent or imply particular donors, and to write plainly for advancement staff. Then comes the task and the pyramid's description. The assistant is told to treat everything in that description as data, not instructions, and names typed by people are kept on one line so they can't change the shape of the request.

On Anthropic models (with Plus or your own Anthropic key), if the chosen model declines a request, Anthropic may run it again on its recommended fallback model before we show “couldn't help with that request”.

A reply that hits its limit isn't saved; you'll see “The AI assistant's reply ran too long” and the request is given back.

How a levels draft is checked before it's saved

1. Amounts are rounded to whole dollars. Levels with no gifts, a zero amount, or an amount above the goal are dropped.
2. Ratios outside 1 to 10 are moved to the nearest end of that range.
3. Two levels with the same amount are merged: their gifts are added together and the higher ratio is kept.
4. Levels are sorted largest gift first, and only the top 12 are kept.
5. If the levels add up to more than the goal, gifts are removed starting from the smallest level, as few as needed, until the total fits. A level that loses all its gifts disappears. This keeps the top of the pyramid intact.
6. If fewer than 3 levels are left, the draft is rejected with “The AI's draft wasn't usable. Try again.” and the request is given back.

How a rule draft is checked before it's saved

• “Gave a percentage over N years”: percent kept between 1 and 100 (10 if missing), years between 1 and 50 (5 if missing).
• “Gave within N years”: years between 1 and 50 (2 if missing).
• “At least N gifts”: between 1 and 1,000 (3 if missing).
• “Gave to a fund”: kept only if the fund matches one in your file, ignoring capitals, and saved with your file's spelling.
• Duplicate conditions are dropped, and at most 4 are kept. Conditions on detail columns are never kept.
• If no usable condition is left, you'll see “The AI's draft wasn't usable. Try again.” and the request is given back.

Messages you might see

While a draft or summary is being written, the button you pressed shows “Drafting…” or “Writing…”, and further Draft with AI and Write summary clicks are ignored until the page reloads, so a double click doesn't spend two requests.

Questions

Can the AI see our donors' names or IDs?

No. It sees totals and settings only: counts and amounts by level, your goal, your rule in words, tier summaries. Constituent IDs and gift rows are never sent, and names and contact details aren't stored in TruePyramids at all.

Did Draft with AI change my levels?

No. Drafts always go into a new scenario called AI draft. The notice above the table confirms “Nothing in your other scenarios changed.” Switch back to your own scenario's tab to see it as it was.

Why didn't the draft start from the edits I just made?

Draft with AI reads the scenario as last saved. Choose Save levels (or Save rule) first, then open the panel again and choose Draft with AI.

How do I apply an AI draft?

There's nothing to apply: the draft is already a working scenario. Use it as it is, edit it, rename it, or delete it. If you'd rather copy its levels into your main scenario, compare the two with Compare scenarios and copy the numbers across in Edit levels.

Can I ask the AI about a particular donor?

No, it can't see individual records and will say so. Use View records on a level or tier to see the constituents behind it.

Does a request that fails still count?

No. Requests that fail on the AI side are given back, and the reason shows on the page. Only answers count.

Is the 150 per person or per organization?

Per organization. Everyone on the team shares the same 150 requests each month.

We've used all 150. What now?

The assistant pauses until the 1st of next month; everything else keeps working. If you need it sooner, the owner can add your own Anthropic or OpenAI key, and requests on it don't count toward the 150.

We have Plus and our own key. Which is used?

Your own key, for every request, with the model the owner chose. Those requests don't use Plus's allowance. Remove the key to go back to the allowance.

Can a member add the API key?

No, only the owner can add, replace or remove it. Members see whether a key is in use, which model, and when it was added.

Can I see the key after saving it?

No. It's encrypted and never shown again, to anyone. To change it, paste a new one with Replace key or change model.

Which model should we choose?

Claude Opus 5.5 is what Plus uses and is the most capable Anthropic model; GPT-6 Astra is OpenAI's most capable. Choose a cheaper model if you mainly ask quick questions and want a smaller bill from your provider.

Is the Ask AI conversation saved?

No. It stays in the panel while you're on the page; reloading or leaving starts a new conversation. Copy anything you want to keep.

Does Write it again keep my edits?

No. It replaces the whole summary. Copy your edited text first if you want to keep it.

How do I put the summary in our Excel export?

Tick Include in Print and Excel exports on the Summary for leadership card. Export Excel then adds a Summary for leadership sheet.

Can I use Ask AI on a donor pyramid or True Pyramid?

No. Ask AI, Draft with AI and Write summary are on gift pyramids. On donor pyramids and True Pyramids, the assistant explains changes over time with What changed? on the History tab, which needs Plus for the history itself.