ICP and targeting
Define your ICP for Alice with firmographic filters, role filters, and exclusions, then enable the Qualification Agent to verify fit before any outreach is spent.
Your ICP is the definition of who Alice may contact, expressed as three things Alice can act on: firmographic filters, role filters, and exclusions. The Qualification Agent then verifies fit in real time before any outreach is spent. Start narrow — one well-defined segment produces better signal than a broad list, and it is far easier to widen a segment than to repair a sending reputation.
Exclusions are the highest-stakes part of this page. A digital worker contacting an account a rep already owns is the single fastest way to lose internal trust in a rollout, and it is entirely preventable.
Firmographic filters
| Filter | Notes |
|---|---|
| Industry | Start with the one or two where you have the strongest proof points |
| Employee count | Often a better fit proxy than revenue for B2B software |
| Revenue band | Useful where deal size correlates with it |
| Geography | Also determines language and calling-consent requirements |
| Technology stack | Strong signal where your product integrates with or displaces something |
Resist the urge to define the ICP as "everyone we could theoretically sell to." That's a market, not an ICP. The ICP is who you can win now, with the proof points you actually have.
Role filters
Define job titles, seniority, and function. For ABM motions, define each buying-committee role you want engaged, not just the primary contact.
This matters more than it appears: personas drive message framing. A CFO and a VP of Sales at the same account get different messages because they're typed differently. If every contact is typed the same way, they'll all receive the same message — which is exactly what makes multi-threading backfire.
| Committee role | Typically cares about |
|---|---|
| Economic buyer | Business outcome, cost, risk |
| Champion | Their own team's day-to-day problem |
| Technical evaluator | Integration, security, effort to adopt |
| Blocker | What breaks, and who has to maintain it |
Exclusions
Exclusions are where your CRM connection earns its value — Alice reads ownership and opportunity state directly.
At minimum, exclude:
- Accounts with an open opportunity
- Existing customers
- Accounts with an assigned owner, where policy reserves them for that rep
- Competitors and partners
- Your do-not-contact list
- Contacts who have opted out
- Accounts already active in another play (see playbooks)
Exclusion rules are only as good as your CRM hygiene. If account ownership is stale in your CRM, it will be stale in Alice. Audit ownership before launch, not after the first complaint from a rep.
The Qualification Agent
The Qualification Agent filters prospects in real time against your ICP before outreach begins, so touches aren't spent on out-of-profile accounts.
This is a structural difference from the usual pattern of qualifying after a reply. Filtering before contact protects three things at once:
Sending reputation
Out-of-profile recipients are the main source of spam complaints, and complaints are the most damaging deliverability signal there is.
Rep time
Replies from people who were never going to buy still cost a rep the time to read them.
Volume allowance
Alice is priced per prospect, so an out-of-profile prospect consumes allowance that an in-profile one could have used.
Signal quality
A clean segment tells you whether the messaging is working. A dirty one tells you nothing.
Tuning
Start narrow
One segment, clearly defined. Prove the motion before widening.
Read who actually replied
Are replies coming from the roles and company types you targeted? If not, the filters don't match the intent.
Read who got disqualified
This is where the expensive mistakes hide. A wrongly disqualified good account disappears silently — nobody reports it.
Widen one dimension at a time
Add an industry, or a seniority band, or a geography. Not all three.
Tighten before scaling volume
A threshold that's slightly wrong is a rounding error at 100 prospects and a real problem at 10,000.