Outbound Pipeline Engine
10 scored prospects with verified emails and outreach drafts, every morning.
Drafts only. It finds, verifies, scores and drafts; it never sends an email or a message.
pinned once: B2B SaaS · 20-200 seats · US + EU · buyer: head of sales / RevOps
the same six steps, every day: find company · find person · get email · verify · score · draft
Prospecting isn’t hard, it’s the same six steps every day: find the company, find the person, get the email, verify it, score it, draft the note. This agent runs all six against your ICP and keeps a pipeline table current with a draft per prospect. Sending stays yours.
One paste into your agent.
Your first pipeline, minutes later.
Install superagnt's Outbound Pipeline Engine skill for me. It's the gtm-prospecting-desk plugin from the github.com/superagnt/plugins marketplace, plus the superagnt MCP server (on Claude it's a connector I add in my browser; elsewhere I sign in through my browser). The setup guide for each kind of agent is at https://www.superagnt.com/install?skill=gtm-prospecting-desk» installs the gtm-prospecting-desk skill (read every file first, or the raw SKILL.md) and enables data:linkedin, database
What AI outbound prospecting means
Outbound prospecting is finding the people who match your ideal customer and contacting them first, instead of waiting for them to find you. AI outbound prospecting hands the daily part to an agent: find the companies, find the buyer at each, find and verify the email, score the fit and draft the first note. You decide who gets it, and you send it.
Want the whole job run for you instead? See AI agents for lead generation.
One desk run: 4 verified prospects for 14.56 credits
The ICP: US advertising and marketing agencies with 11 to 50 people, and the founder or a C-suite buyer. Run on a superagnt test workspace on Nov 3, 2026: about 69 seconds of calls. Names and addresses are redacted below.
“Run the Outbound Pipeline Engine skill: find 10 US marketing agencies with 11 to 50 people, the founder or CEO at each, and verified emails where you can.”
- 01Companies that match the ICP851 ms wall time→ call
data_agnt_companies{ "action": "search", "filters": { "company_industry": { "include": ["Advertising Services", "Marketing Services"] }, "company_headcount": { "min": 11, "max": 50 }, "company_country": { "include": ["United States"] } }, "page": 1, "response_format": "json" }← result{ "items": [ { "name": "[redacted]", "industry": "Advertising Services", "type": "Private", "employee_count": 38, "hq_country": "United States" }, { "name": "[redacted]", "industry": "Advertising Services", "type": "Public", "employee_count": 48, "hq_country": "United States" } ] }» Trimmed: 23 more companies of 25, and every name, domain, description, phone number and link
- 02The founder or C-suite buyer at the 10 companies kept1.1 s wall time→ call
data_agnt_people{ "action": "search", "filters": { "company": { "websites": { "include": ["[the 10 kept domains]"] } }, "person_seniority": { "include": ["Founder/Owner", "C-Suite"] } }, "page": 1, "response_format": "json" }← result{ "items": [ { "full_name": "[redacted]", "job_title": "Chief Executive Officer", "seniority": "C-Suite", "company_employee_count": 38 } ] }» Trimmed: 20 more people of 21, and every name, profile link, location and company field
- 03Find the buyer's work email13.7 s wall time→ call
data_agnt_contacts{ "action": "find_email", "first_name": "[redacted]", "last_name": "[redacted]", "domain": "[redacted]", "response_format": "json" }← result{"email":"[redacted]","status":"valid","full_name":"[redacted]"}» Trimmed: nothing; the address and name are redacted
- 04Then always verify it1.4 s wall time→ call
data_agnt_contacts{ "action": "verify_email", "email": "[redacted]", "response_format": "json" }← result{"email":"[redacted]","status":"valid"}» Trimmed: nothing; the address is redacted
Between the calls, Claude (Opus 5.5) in Claude Code read the 25 companies against the ICP, a check this run added before the skill's people search. 14 failed it, so no person or email lookup was spent on them. The run stopped at verified emails: it stored no rows, scores or drafts.
| Step | Result | Credits |
|---|---|---|
| Company search | 25 companies back | 1.98 |
| People search, first try | Refused before it ran: its filter listed the first 10 companies, and one of their websites was a subdomain, which the search doesn't take | 0 |
| Fit read | 11 kept. Dropped: 5 offices of global networks or acquired agencies, 4 software or measurement platforms, 4 companies that aren't agencies, 1 with no website | 0 |
| People search over 10 of the 11 kept | A founder or C-suite buyer at 7 of 10 | 1.98 |
| Email finder | An address for 4 of the 7 | 9.80 |
| Verifier | 4 of 4 valid | 0.80 |
The run, step by step. Credits are what the workspace was charged.
| Company | Buyer | Fit note | |
|---|---|---|---|
| A private US agency, 11 to 50 people | Chief Executive Officer | Found and verified valid | Fits: an independent agency in the size band, and the CEO buys |
One row from the run, redacted. The fit note is from the run's fit read; the run stored no score.
»Charges read from the workspace's usage log for the run, Nov 3, 2026, 16:45 to 16:48 UTC.
A fit score is a filter, not a reply prediction
What predicted a reply was the post that surfaced the person. On 1,221 LinkedIn engagers we messaged between May 3 and Oct 4, 2026, the share tagged Interested ran from 12% to 24% by source post, while none of the four fit scores we tested told repliers from the rest (AUC 0.49 to 0.55, where 0.5 is a coin flip).
| Source post | Messaged | Replied | Tagged Interested |
|---|---|---|---|
| Founder-led sales, doing my own sales | 179 | 35% | 24% (18.4 to 30.8) |
| AI SDR, GTM AI agents | 68 | 43% | 24% |
| Cold email | 436 | 35% | 20% (16.5 to 24.0) |
| Cold calling | 232 | 31% | 16% |
| SDR and prospecting tips | 142 | 24% | 12% (7.6 to 18.3) |
So the desk scores to rank rows and pick who gets a draft, and you pick sources by what they bring back. Every person messaged had already passed the scorer and a region rule, which narrows what a score can show.
»Our per-source yields study, Oct 4, 2026: 1,221 messaged people matched to outreach outcomes; ranges are 95% intervals. One sender and one offer.
Why every email gets verified
We audited 370 stored lead rows with an email, from an older enrichment path that took a company address instead of the person's. 133 of them (36%) couldn't reach the person: 69 were role inboxes (privacy@, legal@, security@, helpdesk@) and about 64 more didn't contain the person's name.
The desk finds the person's own address and then always runs the verifier. A row that doesn't verify stays in the table as unverified, still workable on LinkedIn, and never counts as a prospect with an email.
»Our lead triage of Sep 23, 2026, aggregates only.
Three steps to your first pipeline.
- 1
Pin your ICP once
Describe who you sell to once: company size, industry, geography, buyer titles. It remembers, and every run after this one starts from that answer.
agnt_db_insert
who you sell to, rememberedexampleB2B SaaS11-50 peopleUS + EU“Head of Growth”said once · read on every run
- 2
It fills the pipeline
It finds companies that match, the right person at each, then finds and verifies their email. Every prospect lands in your pipeline table with its status visible.
data_agnt_companies_search · data_agnt_people_search · data_agnt_people_email_finder · data_agnt_people_email_verifier · agnt_db_upsert
a prospect, resolvingexampleNorthbeam Ops · M. Okafor✓ verified✓ company matched your ICP
✓ found the buyer, not the recruiter
✓ email found + verified
- 3
Review scored drafts
Each prospect gets a fit score with the reason, and the best get outreach drafts in your voice. You read, you edit, you send. Nothing goes out without you.
agnt_db_update · agnt_db_select
waiting for your reviewexampledraft · Northbeam Ops9.4“saw you just posted for a head of growth. most teams hit…”
you read · you edit · you send
»After the first good run, the skill offers the always-on version: a small hosted runner on a schedule, so the work keeps happening with your laptop closed. You confirm the cadence before anything is scheduled.
Read every file before you install it.
SKILL.md tells your agent to pin your ICP once, search companies and the buyers at each, find each email and always verify it, score every row with a one-line reason, and draft notes for the top rows in your voice. Its hard rule: drafts only, it never sends email or LinkedIn messages. Its first two headings are setup for the agent.
Plain markdown your agent reads while it runs. SKILL.md is the procedure. This is the plugin exactly as it installs from github.com/superagnt/plugins: 4 files, 7.3 KB.
Common questions
01What is AI outbound prospecting, and what does outbound prospecting mean?
Outbound prospecting is finding people who match your ideal customer and reaching out first. AI outbound prospecting hands the repetitive daily part to an agent: companies, buyers, verified emails, fit scores and drafts. You review and send.
02How do AI outbound tools handle compliance (GDPR, CCPA, DNC)?
This skill drafts and never sends, so consent, opt-outs and suppression stay in your sequencer and your process. Put your do-not-contact list and disqualifiers in the ICP: a disqualified company never re-enters the pipeline. For GDPR or CCPA questions about your own outreach, ask your counsel.
03Is it illegal to use AI for outbound calls?
This skill makes no calls; it finds prospects and writes email drafts. For calls, in the US the FCC ruled on Feb 8, 2024 that AI-generated voices count as artificial voices under the TCPA, so those calls need the same prior consent as prerecorded ones.
04Can AI fully replace SDRs?
Not this skill. It does the research half of the job every day, and a person still decides who to contact and sends. Our own data says why: on 1,221 messaged people, a fit score didn't predict who replied, and the post that surfaced them did.
05What is the best AI for prospecting?
Look for three things: live data in the AI you already use, every email verified before it counts, and a cost you can read per prospect. On Nov 3 this skill found 4 verified prospects for 14.56 credits, with nothing else to connect.
06How do you measure the ROI of AI in outbound?
Count cost per verified prospect and cost per Interested reply, not leads found. Ours: 3.64 credits per verified prospect on the Nov 3 run, and about 98 cents of data per Interested reply on our own LinkedIn pipeline from May to Oct 2026, before scoring costs. Plans are $49, $149 or $399 a month.
Already connected? Your agent can list every packaged skill: agnt_tools_search + agnt_guidance_search