> ## Content Index > Fetch the complete content index at: https://buildwire.ai/llms.txt > Use this file to discover other available public pages before exploring further. # Negative ICP for AI SDR Agents: Write the No List Before Lead Scoring - URL: https://buildwire.ai/blog/negative-icp-ai-sdr-agents/ - Published: 2026-10-07T16:10:21.000Z - Updated: 2026-10-07T16:10:21.000Z - Description: A four-step workflow for building evidence-based exclusions, checking eligibility before scoring, and routing uncertain accounts to human review. - Author: Rananjay Raj - Tags: ai agents, Prospecting, Lead scoring, claude skills An AI SDR agent will score any account you hand it. A negative ICP tells it which accounts to skip before the score is calculated. Here is the loop we use: mine your lost deals, write a negative ICP file, run it before lead scoring, and send the borderline accounts to a person. Most teams give their prospecting agent a positive ICP (ideal customer profile): headcount, industry, region and a few buying signals. The agent does exactly what it was told. It finds accounts that match, scores them and drafts the first email. Then a rep opens the queue and finds a current customer, a competitor and three companies that churned last year. The model did its job, but nobody told the agent which accounts to skip. ## What is a negative ICP? **A negative ICP** is a written, versioned list of the accounts, situations and buying conditions a team excludes, reviews or suppresses for one sales motion, in a form an agent can check before it scores. It is the inverse of the ideal customer profile, and you build it from lost, churned and stalled deals. Some teams call it an anti-ICP or a negative persona. Prospectory uses "negative persona" for the same idea. The system has four steps. Mine the no list from deals you lost. Write it as a file with three tiers. Run it as a gate before any lead scoring. Send the middle tier to a person. Three free BuildWire Claude Skills cover the research and the scoring: [ICP Research](https://buildwire.ai/claude-skills/icp-research/), [Positioning Gap Finder](https://buildwire.ai/claude-skills/positioning-gap-finder/) and [Signal Scoring](https://buildwire.ai/claude-skills/signal-scoring/). A person owns the rules. ## Why is a positive ICP not enough for an AI SDR agent? A good rep carries the no list in their head and knows which segment never renews and which stack you cannot integrate with. An AI SDR agent starts every run with only the context it was given. Lead scoring makes the gap wider. A high fit score is a strong argument to send, and nothing in a positive profile argues back. [Prospectory](https://prospectory.ai/resources/negative-persona-disqualification-framework/?ref=buildwire.ai) names this the "but their ICP score is so high!" objection. Interest, intent and a high score should not override eligibility. So the no list runs first, and when it says no, the account stops there. ## Step 1: Mine the no list from your lost deals Pull the list from evidence you already hold: closed-lost reasons, churned accounts, deals that stalled past your normal sales cycle, and support load by segment. Run the ICP Research Skill on your best customers first, so you have the positive profile and a scoring matrix. Then run it again on the losses with a different question: what do these accounts share that the winners do not? The answers can be sorted into four groups that map to data you already have: - **Structural fit.** A company shape that keeps losing, such as no team to run the product. - **Technical fit.** A stack you cannot integrate with. - **Commercial fit.** A buying process you cannot survive, such as a security review longer than your runway. - **Relationship history.** Current customers, open opportunities owned by another rep, partners, and anyone who asked not to be contacted. The last group needs no debate. Contacting someone who opted out damages trust, and in many markets it breaks the law. In the US, the FTC's [CAN-SPAM compliance guide](https://www.ftc.gov/business-guidance/resources/can-spam-act-compliance-guide-business?ref=buildwire.ai) requires commercial senders to honor an opt-out request within 10 business days, so those records belong in the file as hard exclusions. ## Step 2: Write the negative ICP as a file with three tiers A paragraph in a prompt is easy for an agent to drift past. Write the negative ICP as a file the agent loads on every run, next to your positive ICP. For Claude Code, one practical option is an anti-persona.md file in the project root, referenced from CLAUDE.md. We split the file into three tiers: | Tier | What it covers | What the agent does | Where the data comes from | | -------------------- | ----------------------------------------------------------------------------------------- | ------------------------------------------ | --------------------------------------------------- | | Hard exclusions | Customers, open opportunities, competitors, do-not-contact, blocked regions | Never scores, never sequences | Synced from the CRM on every run | | Disqualifying events | Pending acquisition, a fresh multi-year contract with a competitor, a cost-cutting round | Parks the account until an expiry date | News and account research, with a date on each line | | Review flags | One risk that needs a human look, such as an unsupported CRM or no reachable budget owner | Sends the account to a weekly review queue | Account research plus your loss data | Disqualifying events describe a moment, so each one carries an expiry date. Hard exclusions describe identity, so they stay until the CRM changes. Each line gets a reason and a source. For example: "Lost 6 of our last 8 deals in this segment, closed-lost report, Q3." Delete any line whose only reason is "feels like a bad fit." Use the Positioning Gap Finder Skill to add one more set of lines. Compare your messaging with your closest competitor's and note the buyers their product serves better. Write those down as "not us" so the agent stops chasing them. Keep every rule lawful. Each one must be a business criterion you can check. Do not target or exclude people based on protected traits, or infer them from names, photos, locations or proxies. ## Step 3: Run the negative ICP gate before lead scoring The order of operations matters: the agent checks the file first, and only accounts that pass reach the Signal Scoring Skill. Intent signals and business events then rank what is left. If you score first and filter second, the score leaks into everything downstream. A rep sees a 92 next to a competitor and starts arguing for an exception. Store the result on the account record: an eligibility flag and the rule that set it. Make outbound lists exclude flagged accounts by default, so nobody bypasses a rule while building a list in a hurry. Three terms get mixed up here. **A suppression list** is a list of specific people and accounts you must never contact. **A lead score** is a ranking of eligible accounts by fit and intent. Here is how the three layers differ: | Layer | Question it answers | When it runs | Who changes it | | ---------------- | ----------------------------------------------- | -------------------------------- | ------------------------------------- | | Suppression list | Must we never contact this person or account? | Before everything, on every send | Ops, synced from the CRM and opt-outs | | Negative ICP | Is this account eligible for this sales motion? | Before lead scoring | A person who owns the rules | | Lead score | Among eligible accounts, which comes first? | After the negative ICP gate | The scoring model, tuned by a person | ## Step 4: Send the middle tier to a person Not every match should be an automatic block. Prospectory uses a graduated rule: three or more negative signals means auto-disqualify, two means manual review, and one means proceed and flag the risk in the CRM. We run the same shape. Hard exclusions block. Disqualifying events park the account until the expiry date. Review flags go to a short queue that a person clears each week. The agent can research the account, gather the evidence and recommend an action. It does not change the rules. Dima Bilous, founder of Anfloy, puts it plainly in his [negative ICP guide](https://www.anfloy.com/blog/negative-icp?ref=buildwire.ai): "I would not let AI invent production exclusion rules." The agent proposes and a person approves. It is the same READ and PROPOSE split we use for the [Measurement Desk](https://buildwire.ai/blog/claude-marketing-write-access-measurement-desk/). ## How often should you update a negative ICP? Put a version number and a date at the top of the file. Review it every quarter, and sooner after a new product line, a new market or a pricing change. When a rule stops matching reality, delete it so it cannot quietly block good accounts. Rules are specific to a sales motion. An account you exclude from cold outbound might be fine for a partner referral. Keep one file per motion if your motions differ. ## The negative ICP loop at a glance 1. Run ICP Research on winners, then on losses. 2. Write the negative ICP file: hard exclusions, disqualifying events and review flags, each with a reason and a source. 3. Use Positioning Gap Finder to add the "not us" lines. 4. The agent checks the file before anything is scored. 5. Signal Scoring ranks only the eligible accounts. 6. A person clears the review queue and owns every rule change. This follows the same pattern as [Brand DNA](https://buildwire.ai/blog/brand-dna-extract-lock-inject/): one versioned file, approved by a person and loaded by every agent. Brand DNA tells the agent how to sound. The negative ICP tells it who to leave alone. If you want this gate built into your own prospecting agent, [book a call](https://calendly.com/rananjayrajw1/30min?ref=buildwire.ai) and tell us which accounts keep slipping through. ### How we checked this guide The linked Prospectory, Anfloy and FTC references were checked on October 7, 2026\. The three tiers, expiry dates and weekly review queue describe the supplied BuildWire workflow, not client results. Spot an error? [Contact us](https://buildwire.ai/contact/), or read [about BuildWire](https://buildwire.ai/about/). *Want the prospecting and agent workflows we run every week, written up like this one? Subscribe to [The AI Driven Marketer newsletter](https://buildwire.ai/newsletter/), and follow along as we turn slow marketing jobs into short ones.* ## Frequently asked questions What is a negative ICP? A negative ICP is a written list of the accounts, situations and buying conditions a team excludes, reviews or suppresses for one sales motion. It is the inverse of the ideal customer profile, and it is built from lost, churned and stalled deals. Is a negative ICP the same as an anti-ICP? Yes. Anti-ICP, negative ICP and negative persona usually describe the same idea: the profile of accounts you choose not to pursue. Pick one term and use it everywhere your agents and reps read it. What is the difference between a negative ICP and a suppression list? A suppression list blocks specific people or accounts that must never be contacted, such as opt-outs and current customers. A negative ICP describes types of accounts and situations that are a poor fit for one sales motion. The suppression list runs on every send. The negative ICP runs before lead scoring. Should an AI agent write the negative ICP? An agent can research accounts and suggest patterns from your loss data. A person should approve every production rule, because the agent cannot tell which pattern is a real cause and which is noise. Should the negative ICP run before or after lead scoring? Before. Run the exclusions first and score only the accounts that pass, so a high score never argues an account back in. How often should you update a negative ICP? Quarterly suits most teams, and sooner after a new product, market or pricing change. Version and date the file so agents load the current rules. Last reviewed 7 October 2026\. Linked references checked on 7 October 2026\. Workflows need human review before production use.