Your ICP and messaging, written for an agent

Step 1Write the ICP as rules

An ICP written for people leans on judgement: mid-sized, finance-heavy, senior. For an agent, turn each line into a value it can check in a field, name the disqualifiers, and say what to do when the field is empty: skip, or ask a person. Here is Example Software, an invented company that sells to finance teams.

In Artisan, switch on AI qualification (it starts off) and add each rule as a criterion; Ava contacts only prospects who pass all of them, at 1 credit per criterion per prospect. In 11x, enter the numbers as firmographic filters and the titles as role filters; who is off limits whatever the fit goes in its exclusions (part 8). In HubSpot’s prospecting agent, add only segments inside your numbers and personas for your titles: they decide which companies it watches and which contacts it suggests. In Apollo’s AI context center, list the disqualifiers as bad-fit customers under Additional context.

Step 2Give every claim its source

11x defines the ICP as “who you can win now, with the proof points you actually have”. So list them, each with its source: a case study the customer approved, a dated count, a signed quote. A point without one stays off: in Artisan, for one, “everything uploaded is treated as true. An unverified fact competes on equal footing with a verified one.” Then list the claims not to make. As 11x puts it, “Absence of a claim is not a prohibition on making it.”

Write each point once, in one document (support part 4’s one answer per question), and copy from it into each tool. In HubSpot, if the play’s Context tab was drafted from your website, check it against the sheet, and put the claims not to make in the advanced instructions, where HubSpot’s own example, “Emphasize our proven ability to adapt to market changes”, is a claim with no source. In Apollo, the value proposition takes only sourced outcomes, and Additional context the claims not to make. In Artisan, add only sourced points under Proof & results, with each case study’s note on when to mention it. In 11x, list the claims in the knowledge base’s boundaries.

Step 3Personalise from fields only

One rule covers personalisation: each sentence about the prospect points to a value in their record (part 4), or it isn’t written. Your own history with them counts: the last meeting and its notes are fields too. An empty field drops its sentence, and a guess has no field at all.

Read drafts against the record, starting where 11x’s review list does: “Is every factual claim true?” and “Check the research citations.” In Artisan, set Personalization to Lite, where “Ava picks one relevant detail for each lead”: one field to trace. In HubSpot, use the play’s Test button on a contact with empty fields, and read what the agent writes without them.

Step 4Keep the brief in versions

An edit doesn’t reach what is already under way. In HubSpot, “when you edit a play, the changes you make will only apply to newly-enrolled contacts”; those already enrolled continue on the version they enrolled under. Apollo “doesn’t retroactively update AI-generated content that already exists”.

So number and date each version, and log the day it went into each tool: a contact’s enrolment date then tells you which version they got. When a version withdraws a claim, unenrol the HubSpot contacts still on the old one, and in Apollo re-run AI research for anyone you’ll write to again.

Next in this series

Part 6: connecting HubSpot, Clay and your sending tool. See all 13 parts.

Why we write these guides