AI Outreach That Builds Trust: Matt Symes' Warm-Collision Sales Workflow

The most useful part of Andrew Warner's conversation with Matt Symes is not "use AI to write better cold emails." That is too small. The better lesson is this: use AI to understand the market before you ask for the meeting.

Matt Symes, founder of Levership, walks through a sales workflow where AI helps find very specific prospects, build better prompts, create warm entry points, and prepare for meetings with public-source research. The interesting shift is that AI is not just drafting words. It is helping the salesperson think.

Source note

Credit for the interview goes to The Next New Thing and host Andrew Warner. The guest is Matt Symes, founder of Levership. Levership's public site describes its work as helping hands-on small business leaders install systems, cadence, and accountability, including AI-amplified sprints.

The episode claims about revenue, portfolio revenue, client examples, and workflow results are treated as interview claims, not independently audited financial facts. Product facts about Perplexity Computer are checked against Perplexity's public product page. Outreach compliance notes are grounded in FTC and ICO guidance and are not legal advice.

Matt also shared two public lead magnets: Find Your Next 25 Customers and OSINT Prompt Library. I linked the originals and organized the ideas into adapted templates below rather than pasting the PDFs in full.

Link Map

The core lesson: do not go wider, go narrower

Matt's lead-generation examples work because they are specific. In the episode, the strongest example is not "find small businesses." It is a client that sells medical supplies to people who cannot quickly get to a hospital, where analysis revealed that fishing vessels were strong buyers.

That is the key move. AI gets much more useful when the target shifts from a broad label to a real-world segment:

The operator principle is straightforward: the narrower the segment, the easier it is to define source lists, buying signals, disqualifiers, and a message that sounds like it belongs in that buyer's world.

Lead finding with Perplexity Computer

In the episode, Matt uses Perplexity Computer as a research worker. Perplexity's product page describes Computer as a general-purpose digital worker that can browse, research, connect tools, create workflows, and run background tasks. For lead generation, that matters because prospecting is not one action. It is a chain:

1. Define the narrow segment.
2. Find public directories, regulatory lists, associations, maps, or company pages.
3. Extract candidate accounts.
4. Check public signals that suggest need now.
5. Rank by fit and evidence quality.
6. Draft an outreach angle for review.
7. Preserve assumptions so the human can rerun or correct the search.

The best part of Matt's example is the assumptions tab. It makes the workflow inspectable. If the agent assumes the wrong geography, buyer title, company size, or signal, the user can fix the premise instead of manually cleaning a bad list.

Model Council: use AI to improve the prompt before the search

Matt's "Model Council" move is clever because it separates prompt design from execution. Instead of asking one model to find the leads immediately, he first asks multiple model perspectives to critique the target and improve the research plan.

The useful pattern is:

• Ask models where the segment is too broad.
• Ask what public sources are likely to contain reliable data.
• Ask what geography is too large for one clean search.
• Ask what buying signals would make the lead worth pursuing now.
• Ask what disqualifiers should prevent a company from entering the list.

In the transcript, one practical outcome was narrowing a Florida search down to county or metro batches. That is exactly the sort of planning step that makes AI lead generation less messy. A model that returns a thousand prospects is not automatically doing better work. It may simply be ignoring the operational cost of reviewing them.

Warm collisions instead of cold outreach

Matt's phrase "warm collision" is the heart of the post.

A cold approach asks the buyer to trust a stranger. A warm collision creates a credible reason for the buyer to encounter you. That might be a mutual connection, a relevant comment, a small diagnostic, a useful teardown, a shared community, a podcast angle, or a timely artifact connected to something the buyer already cares about.

The point is not to fake familiarity. The point is to stop making the first touch feel random.

This is where AI can help without becoming spam: it can search public context, identify a natural path, and suggest a respectful first move. The human still decides whether the path is appropriate.

OSINT meeting dossiers

The second half of the workflow is meeting prep. Matt shows how he uses public-source research to create a dossier before conversations, including the person's role, the organization's pressure points, likely priorities, and safe areas of overlap.

Used well, this makes sales less robotic. The salesperson walks in understanding the buyer's context, not just their own script.

Used badly, it becomes creepy. The line is simple: stay with public, professional, role-relevant information. Do not include home addresses, private emails, family details, medical details, doxxed data, or personal information that does not belong in a business conversation.

A good dossier should help with:

• a concise organization read;
• current business pressures and news hooks;
• the person's likely decision lens;
• safe rapport anchors;
• questions worth asking;
• phrases to avoid;
• a low-risk next step.

Matt also mentions recording meetings and turning transcripts into follow-ups, proposals, or better internal workflows. That can be useful, but record only with the right consent and policy checks. Recording laws vary by location, and team/company policies may be stricter than local law.

Prompt library from Matt's lead magnets

The two PDFs Matt shared are useful because they make the workflow concrete. The original files are linked here:

• Find Your Next 25 Customers
• OSINT Prompt Library

Below are adapted working versions organized for an AI outreach system. Use the originals if you want Matt's full wording. Use these when you want to turn the idea into a review-first pipeline.

1. Find 25 customer accounts

Role:
Act as a senior business development analyst.

Goal:
Find 25 candidate customer accounts I can review this month.

My business:
- Product or service:
- Outcome I deliver:
- Proof points:
- Geography priority:
- Company size:
- Target industries:
- Best-fit buyer titles:
- Hard exclusions:

Research rules:
- Use public sources only.
- Prefer company websites, public directories, association lists, public filings, credible news, and official pages.
- Do not invent companies, contacts, URLs, revenue, or buying signals.
- If a fact is estimated, label it as estimated.
- If fewer than 25 leads pass the filter, return fewer and explain why.

For each account, return:
- Account name
- Website
- Location
- Industry or sub-segment
- Estimated size
- Public buying signal and source link
- Why it fits the ICP
- Likely decision-maker titles, not private contact data
- Warm-collision angle
- Outreach angle in two short sentences
- Confidence: High, Medium, or Low

Output:
1. Lead table
2. Top 10 ranked accounts
3. Patterns across the market
4. Three messaging themes to test
5. Five accounts to review first
6. Assumptions to confirm before outreach

2. Model Council prompt builder

Role:
Act as a council of three AI sales strategists.

Task:
Before finding leads, critique my target market and improve the research prompt.

Inputs:
- What I sell:
- Best existing customer:
- Segment I think I want:
- Geography:
- Offer:
- Constraints:

Council output:
- Where the target is too broad
- What segment should be searched first
- What geography or batch size is realistic
- What public data sources are likely to work
- What buying signals matter
- What exclusions prevent bad-fit leads
- A stronger final research prompt
- A run plan: batch order, review checkpoints, and when to stop

3. Warm-collision planner

Role:
Act as a public-source relationship strategist.

Goal:
Find ethical, professional ways for me to create a warm collision with this account or person.

Inputs:
- Target account:
- Target person or role:
- My offer:
- Why this account might care:
- Existing links, communities, or content:

Rules:
- Use public professional information only.
- Do not use private contact data, personal family details, doxxed data, or sensitive assumptions.
- Do not imply a relationship that does not exist.

Return:
- Best public context
- Likely business priority
- Shared connection or community path, if any
- Useful comment or content angle
- Small artifact I could create
- First-touch draft for human review
- Risk notes: what would feel forced or creepy

4. OSINT meeting dossier

Role:
Act as an OSINT analyst and executive meeting briefer.

Goal:
Prepare a decision-grade meeting brief using lawful public sources.

Inputs:
- Person:
- Organization:
- Meeting purpose:
- My desired outcome:
- My offer and constraints:

Boundaries:
- Public sources only.
- Disambiguate identity first.
- Do not include home addresses, private emails, personal phone numbers, family details, medical details, or breached data.
- Separate verified facts from hypotheses.

Output:
- 10-line executive brief
- Organization dossier
- Person role and career-pattern read
- Current news hook
- Overlap table: their priority, my proof, best phrasing, risk, suggested artifact
- Meeting playbook: opener, 7 questions, 3 credibility bridges, next-step ask
- Gaps and open questions
- One-page meeting card

5. Morning meeting-prep automation

Every morning at 5:00:
1. Review my calendar for the next 48 hours.
2. Identify external meetings.
3. For each meeting, create an OSINT meeting brief using public sources and my saved context.
4. Mark claims as verified, probable, or hypothesis.
5. Include a short review checklist before I use it.
6. Do not contact anyone, update CRM records, or send follow-ups without approval.

A review-first outreach workflow

The safest way to turn this into a real system is not "agent sends email." It is a review queue.

This is the version I would build for a client. The agent can research and draft, but the business keeps control of sending, scheduling, CRM updates, and contact data.

Compliance and consent notes

This is not legal advice, but the workflow needs guardrails.

In the US, the FTC's CAN-SPAM guidance says commercial email must use accurate header information, non-deceptive subject lines, sender identity and location, opt-out instructions, prompt opt-out handling, and oversight of vendors. The guide also notes that CAN-SPAM applies to B2B commercial email, not just bulk consumer spam.

In the UK, the ICO's B2B marketing guidance distinguishes corporate subscribers from individual subscribers, but still warns that UK GDPR applies when personal data is processed. The practical version is simple: keep targeting role-relevant, identify yourself, provide a valid opt-out path, maintain suppression lists, and document why the message is relevant.

For Portugal and EU outreach, be extra careful with GDPR and ePrivacy rules. Do not assume US-style outreach rules apply. If the workflow touches named employees, enrichment tools, recording, or exported contact data, get legal or compliance review before running it at scale.

JQ AI SYSTEMS outreach checklist

If I were turning Matt's workflow into a real AI outreach system, this would be the build checklist:

1. Define one narrow buyer segment, not a broad audience.
2. List the public sources that are allowed for that segment.
3. Define fit signals, need signals, timing signals, and hard exclusions.
4. Generate a small lead list first: 25 to 50 accounts.
5. Require source links for every buying signal.
6. Score accounts with transparent reasons.
7. Generate warm-collision paths before generating email copy.
8. Create a meeting dossier only from public, professional context.
9. Put all drafts into a human review queue.
10. Keep opt-out, suppression, duplicate, and rejected states visible.
11. Record results: replies, meetings, objections, false positives, and closed-loop learnings.
12. Improve the ICP and prompt before increasing volume.

That last step is the one most teams miss. The goal is not more messages. The goal is sharper understanding. Once the segment, sources, signals, and offer are working, scale carefully.

Sources

• YouTube: Matt Symes shares AI workflows with Andrew Warner
• The Next New Thing
• Andrew Warner on LinkedIn
• Levership
• Perplexity Computer
• Perplexity Comet
• PLAUD AI
• Matt Symes: Find Your Next 25 Customers PDF
• Matt Symes: OSINT Prompt Library PDF
• FTC: CAN-SPAM Act compliance guide for business
• ICO: Business-to-business marketing
• ICO: Plan direct marketing