TL;DR: Two layers of targeting. ICP says "we sell to this kind of company"; Persona says "and these are the people at that company we want to reach." Records get scored against both, independently, and audiences usually require strong scores on both.
The shape of each
| ICP | Persona
**What it targets** | Companies (accounts) | People (leads)
**Lives at** | Workspace root | Inside an ICP
**Typical parameters** | Industry, employee count, revenue band, tech stack, geography, growth signals | Title patterns, seniority, function, tenure, custom questions, person-level signals
**Output field** | `icp_code` + `icp_score` | `persona_code` + `persona_score`
**Score range** | 0–100 | 0–100
A record can have an ICP score but no Persona match — that's a known state, surfaced as No persona fit in Qualification. Useful: tells you the account is interesting but you haven't defined the right buyer role yet. It's not actionable for outbound until a Persona is defined.
The hierarchical relationship
Every Persona belongs to exactly one ICP. The Qualification table renders this as a two-level tree: ICPs at the top level, Personas nested under each.
Mid-market SaaS (ICP)
├─ Founder (Persona)
├─ VP Sales (Persona)
└─ No persona fit (auto, system-managed)
SMB SaaS (ICP)
├─ Founder (Persona)
├─ VP Sales (Persona)
└─ No persona fit
Enterprise FinServ (ICP)
└─ ... (still being defined)
You cannot share a Persona definition across ICPs. If your "VP Sales" Persona looks different at SMB vs Enterprise, you create one in each. The Qualification table makes this duplication ergonomic — Copilot can clone a Persona across ICPs in one prompt.
Parallel scoring
When a signal fires for a person at a company:
- The company is scored against every Live ICP. The record gets the ICP code of whichever ICP it matched best (highest
icp_score). - The person is scored against every Live Persona inside that ICP. The record gets the Persona code of whichever Persona matched best inside that ICP.
A high ICP score with a low Persona score means "right company, wrong person." A high Persona score with a low ICP score means "right person, wrong company." Audiences typically require both above a threshold — usually 60+ on each — to act on.
What goes in an ICP
A good ICP is opinionated about companies. Examples:
- Industry — software, SaaS, healthcare, financial services
- Employee count — bands (1–10, 11–50, 51–200, 201–500, 501–2000, 2000+)
- Revenue band — ARR or revenue ranges
- Geography — country, region, sub-region
- Tech stack — installed tools that signal compatibility or competition
- Growth signals — funding stage, hiring velocity, net-new headcount
- Engagement-with-you — past pipeline, current customers, past churn
What goes in a Persona
A good Persona is opinionated about people. Examples:
- Title pattern — exact / wildcard / semantic matches
- Seniority — IC / Manager / Director / VP / C-suite
- Function — Engineering / Sales / Marketing / Security / RevOps
- Geography — when person-level geo matters (e.g. for time-zone-restricted motions)
- Years of experience — total
- Years in current role — tenure (often a buying-cycle indicator)
- Custom persona-specific question — e.g. "What are they trying to solve right now?" — answered per-record by enrichment LLM
- Person-level signals — has this person specifically posted / engaged / replied
When to use which
Three patterns:
- Account-based motion (ABM) — work at the ICP level first. Pick 200 target accounts; the Personas inside each are the buying committee.
- Volume outbound — Persona is the lead. Filter by Persona score first, then check ICP score as a tiebreaker.
- Inbound qualification — both matter equally. Inbound leads get scored against both; route based on combined score.