Your Ideal Customer Profile (ICP)
Your ICP is the answer to one question: who is worth a connection request. SDRPilot does not ask you to describe your buyer from memory. It reads the customers you already have and writes the profile from them, and you keep a set of hard rules on top that nobody gets past.
Two halves, one page
The profile is the written half. It describes the people who buy from you, the signals that say somebody is a good fit, and the things that disqualify them. The AI reads it when it scores a lead, and your agent reads it when it writes.
The filters are the machine half. They are yes or no gates that run before any AI is asked anything. A lead that fails a gate is never scored and never invited.
One profile or several
Most workspaces run a single ICP, and it opens on the page with no click. If you keep more than one, for example a profile per offer, the page lists them a line each instead: the name, whether it is built or still learning, what the build job is currently doing, and the roles and industries it targets. Click a line to open that profile in place, click it again to close it. The line also shows the version, when it last built, and how many agents are pinned to it, so you can tell two profiles apart without opening either.
How the profile gets written
- Mark the people who bought from you as customers, on the lead itself or in bulk from the leads list. You can also import them, see below.
- SDRPilot reads those customers and writes the profile from what they have in common: countries, languages, industries, the roles they actually hold, the signals that repeat.
- Every rebuild is saved as a version, so nothing is ever overwritten beyond recovery.
A brand new workspace has no customers yet, so the first draft is written from the business description you gave during onboarding. It is a starting point on purpose, and it is rebuilt from real customers as soon as you have some. Under five customers the profile is still learning, and SDRPilot leans on what you pinned by hand rather than on a pattern drawn from too few people.
Anything you edit yourself is pinned and always wins over what was derived. Unpin it and the derived value comes back.
The hard filters
Every gate is off unless you set it. A profile with no gates lets everybody through to scoring, which is the safer default: a wrong rule silently rejects the people who would have bought.
- Countries. Either Only these or All but these. Core markets, expanded markets and other countries are just groupings to make a long list readable: a country cannot be moved from one heading to another, and the heading has no effect on scoring. Only the ticked countries count. If Singapore is your market, tick it under expanded markets and untick the core ones you do not sell in.
- When we cannot tell where they are. LinkedIn does not always show a location. Choose Pass, Cap the score with a ceiling you set, or Reject.
- Languages. Tick every language you are happy to sell in, and choose Pass or Cap the score when the language is unclear.
- The lead’s own industry. Only these industries, never these industries, or both.
- Roles. Decision makers only, plus a list of job titles to skip.
- Exclusions. Skip self employed and freelancers, skip people at big corporates (with your own list of company names), and skip engagement pod members.
- Minimum engagement. For leads found through a post: A like is enough, or Must have commented.
- Network size. A minimum follower count. LinkedIn does not publish how many connections another member has, so followers is the number that can actually be used. Choose what happens when the number is unknown: pass, cap the score, or reject. Leads that were researched before 15 September 2026 have no follower count stored yet and count as unknown until they are researched again.
- Score thresholds. The minimum AI score a lead needs, with a different threshold per source if you want one. A profile view converts very differently from a keyword search, so it can earn a different bar.
Changing filters in plain English
You do not have to hunt through the form. Type what you want, for example “only Benelux and DACH, and at least 500 followers”, and press Suggest. SDRPilot turns it into a proposal, tells you in a sentence what it changed, and shows you which settings are affected. Nothing is saved until you press Save, so a suggestion you dislike costs you nothing.
See the impact before you save
The impact preview replays a set of filters over your most recent leads and tells you how many would be rejected, by which gate, how many would only be capped, and how many would still pass. Use it as a sanity check: a gate that rejects almost everything is usually a typo, not a strategy.
Filters also apply to what is already waiting to be sent. See The Connection Queue for how to re-check the queue against your new rules. Leads the old profile had already rejected are not re-judged on their own: select them on the Leads page and choose Re-score with AI.
History and restore
Every build and every edit writes a version. Open the history to see what changed between versions, which values were added or removed, and whether the gates moved. Restore puts an earlier version back as your current profile, and that restore is itself a new version, so you can undo the undo.
Importing customers
Press import and paste either a list of LinkedIn profile URLs or a spreadsheet, up to 500 rows at a time. Rows with a LinkedIn URL become leads (or mark a lead you already have, keeping its history). Rows without one are kept as written evidence about a customer, which still teaches the profile. Rows SDRPilot cannot read are listed back to you with their row number, so you can fix just those.
Importing also researches the profiles you pasted, spread out over time on purpose rather than in one burst. You can import your customer list before you have connected LinkedIn: the marks land either way and the research happens once a seat is connected.