AI features
Ask your data, reply drafts, coaching notes and campaign diagnosis: what each one does, its limits, and exactly what is sent to the AI provider.
For: Admins and managers; reps for Ask and drafts
There are four AI features. All of them work from numbers the product has already computed. The AI never writes a query and is not allowed to invent a figure: every number it shows comes from the same reports you can open yourself.
The four features
| Feature | Where | Who | What it does |
|---|---|---|---|
| Ask your data | Ask | Everyone | Answers a question in plain language from the report data, with a link to the report behind each answer. |
| Reply drafts | Replies → Draft with AI | Anyone who may answer that reply | Suggests an answer in the reply box. A person edits it and presses Send. |
| Coaching notes | A rep's page, and Reps → Team coaching | Managers and admins | A short note per rep from the last four weeks against the four before: observations with their numbers, and suggested actions. |
| Campaign diagnosis | A campaign's page | Managers and admins | Names the most likely problem with a campaign and explains it from its numbers. |
Ask your data
Ask your data: type a question or start from an example. Your history is private to you. Demo data.
- Ask about reps, campaigns, tools, channels, replies, connections and enrichment spend, for a period such as this week, last month or a date range.
- Each answer carries source chips, for example "Rep scorecard, 5 to 11 Oct". A chip opens that report with the same period and filters, so you can check the answer.
- If a name could mean two reps or two campaigns, it asks which one instead of guessing.
- A rep can only ask about their own numbers. Asking about another rep is refused before anything is looked up.
- Limits: 1,000 characters per question, 12 questions in 5 minutes and 200 a day per person. Your last 20 conversations are kept and are private to you.
Coaching notes
A note covers the last four full weeks against the four before, whatever period the page is showing. It compares the rep with their own targets and with the team median (never with a named colleague), and it is written to describe activity, not character.
- A rep with too little activity gets "Not enough data yet" and no AI call is made.
- A statement whose number is not in the data is dropped before you see it.
- A stored note stays readable after the numbers move, marked "the numbers have changed since". Refresh writes a new one.
- Copy for 1:1 gives plain text for your meeting notes.
- Reps cannot see coaching notes, including their own.
Campaign diagnosis
The verdict is decided by fixed rules before the AI is involved, and shown as a chip on the campaign page even with AI off. The AI then explains the verdict; it may not contradict it.
A campaign page. The Diagnosis card shows the verdict from the fixed rules; Diagnose adds the explanation. Demo data.
| Verdict | Rule |
|---|---|
| Not enough data | Fewer than 100 sends, or activity in fewer than 2 weeks |
| One bad mailbox | A mailbox with 50 or more sends, a bounce rate of 5% or more, and at least twice the rate of the other mailboxes |
| List problem | Bounce rate of 5% or more |
| Wrong audience | Unsubscribes at 1% of sends or more |
| Message or targeting | Bounces fine, 300 or more sends, and a reply rate under half the median of comparable campaigns (or under 0.5% when there is no median) |
| Offer or targeting | 10 or more human replies, fewer than 15% positive |
| Slow follow-up | 3 or more positive replies and no meetings |
| No clear problem | None of the above |
Settings and limits
Settings → AI: the workspace switch, the reply guidelines and usage against the monthly allowance. The demo workspace is on a Custom plan, so its allowance is larger than Pro. Demo data.
- On or off. An admin can turn all AI features off for the workspace in Settings → AI.
- Monthly cap. Usage is counted in tokens per calendar month, across all four features: 500,000 on the trial and 1,000,000 on Pro. Settings → AI shows usage against the cap, by feature and by person. When the cap is reached, AI pauses until the next month; everything else keeps working.
- Reply guidelines. Up to 2,000 characters of tone, booking link and things not to promise. Managers and admins can edit them.
- Read-only mode. AI is paused. Past Ask answers and stored notes stay readable.
What is sent to the AI provider
AI requests go to Anthropic, and only when someone uses a feature. Prompts are built from a fixed list of fields, never a whole record, and keys or secrets are never included. This is the same statement shown in Settings → AI:
- Ask your data
- The question as typed, the earlier questions and answers of that conversation, and the report numbers needed to answer it: counts, rates and amounts per rep, campaign, tool or period, with the names of reps, campaigns and connections (and connection error messages, for managers who ask about them). Not sent: the text of emails or replies, contact names, email addresses, phone numbers, deal names.
- Reply drafts
- When someone presses "Draft with AI" on a reply they are allowed to answer: the conversation with that one contact (the prospect's replies and the text of your own messages where it is stored, newest 4,000 characters), how the reply is classified, the channel, the prospect's first name and company, the campaign name, the first name of the person drafting, this workspace's reply guidelines, and for a rewrite the text in the reply box and the instruction typed. Not sent: email addresses, last names, phone numbers, LinkedIn URLs, job titles, any other contact or conversation, deals. The draft is placed in the reply box and is never sent or stored by the AI feature; only whether a sent reply started from a draft is recorded.
- Coaching notes
- When a manager or admin asks for a coaching note: that rep's name and aggregated numbers for the last four weeks and the four before (activity against targets, reply and positive rates, the team median of those rates without names, meetings, pipeline, reply response times, waiting-reply counts, weekly trend) and the names of their best and weakest campaigns. Not sent: the text of emails or replies, contact names, email addresses, phone numbers, deal names, or other reps' names.
- Campaign diagnosis
- When a manager or admin asks for a diagnosis: that campaign's name, tool, channel and owner's name, its aggregated numbers for the last four weeks and the four before (sends, bounce, reply, positive and unsubscribe rates, meetings, deals, pipeline, weekly trend, per-step counts when the tool reports steps), the median of comparable campaigns, and per-mailbox counts with the mailbox labels. Not sent: the text of emails or replies, contact names, email addresses, phone numbers or deal names.
See also the Privacy policy and Sub-processors.
How far to trust it
- Numbers are safe to rely on: they come from the reports, and the source link lets you check.
- Explanations and suggested actions are a starting point for your own judgement, not a decision.
- A reply draft is a draft. The person who presses Send is responsible for what it says.
Something unclear or out of date? Tell us and we will fix this page.