AI personalization
How Alice writes per-prospect copy from four inputs, adapts tone by channel and persona, and how to keep control with review and approval workflows.
AI Personalization writes copy for an individual prospect rather than filling variables in a template. It synthesizes four inputs — external data, your knowledge base, CRM context, and signal data — then adapts length, formality, and structure to the channel, and framing to the persona. You keep control through review, approval workflows, or full autopilot.
The four inputs
| Input | What it contributes |
|---|---|
| External data | Public information about the prospect and their company, via Deep Research |
| Knowledge base | Your positioning, product detail, differentiation, and proof points |
| CRM context | Account history, lead data, and previous interactions |
| Signal data | Job changes, funding rounds, tech stack changes, and timing indicators |
In 11x's words: "Lead and account data from your CRM, job changes, funding rounds, tech stack, and account history all feed into how the message is framed."
Output quality is bounded by the weakest input. Teams usually invest in external research and neglect the knowledge base — which produces well-researched messages that make generic claims about your product.
Channel-aware formatting
An email reads differently from a social message. Alice adjusts length, formality, and structure based on where the message lands, so it reads as native to the channel rather than an email pasted into LinkedIn.
| Channel | What changes |
|---|---|
| Longer, more structured, subject line carries the hook | |
| Shorter, more conversational, no subject line | |
| SMS / WhatsApp | Very short, direct, single ask |
| Phone | Talking points rather than prose |
Persona-aware framing
A CFO and a VP of Sales at the same company get different messages, because they care about different things. Alice frames the same underlying value around what that role is accountable for.
This is what makes multi-threading an account work. Sending the same message to four people on a buying committee is transparent and counterproductive; sending four role-appropriate messages is how ABM is supposed to function. See ABM plays.
Control and guardrails
Three modes, in increasing order of autonomy:
Every message can be reviewed, edited, or overridden before it sends. Use this while you're calibrating a new segment or a new knowledge base.
Set an approval step so a named person signs off before prospects see anything. Appropriate for regulated industries, named strategic accounts, or a first rollout where internal trust matters.
Alice sends without per-message approval. Appropriate once you've validated output quality on a segment and your knowledge base boundaries are complete.
Don't start on autopilot. Review the first sample, fix what the output reveals about your inputs, then graduate. Teams that skip the review phase typically conclude the technology doesn't work when the real problem was a thin knowledge base.
Reviewing output well
Reading generated copy is a skill. What to look for, in priority order:
Is every factual claim true?
Check the research citations. Accuracy first — everything else is style.
Would you let a rep send this under their own name?
The most reliable single test. If the answer is no, the gap is in your inputs.
Does it make a specific claim, or a category claim?
"Teams like yours struggle with pipeline" is a category claim. Category claims mean thin product detail in the knowledge base.
Is the ask clear and small?
One ask, easy to say yes to.
Read five in a row
Individually good messages that are structurally identical still read as automated. Variation across prospects is the signal to check.