A solo founder cannot run a six-person GTM process. Software can.
Outbound done properly is weeks of work for a full sales team.
Crewm8 is a pre-launch AI receptionist for local service businesses. Zero customers, zero brand, GTM team of one. ICP research, lead sourcing, contact verification, scoring, personalized copy, channel strategy, and compliance take a six-person team weeks. Doing it improperly with generic AI blasts burns the market before launch.
Turn the process the team would have followed into software.
A 32-skill go-to-market operating system built on Claude Code: modular, chainable AI skills that encode real sales methodology, with hard quality gates between every stage. No stage runs on unverified data from the stage before it. The AI cannot invent a personalization hook, send to an unverified email, or push past a compliance threshold.
30 leads in. 11 replies, 9 interested, 5 top Canadian barbershops onboarded.
The core pass, from list to scores to DM drafts, landed inside a two-hour window by CRM timestamps. The campaign then converted 5 of the 7 leads the agent had scored Tier-1 into the beta trial cohort. Price, not product, filtered out the rest.
A system, not a pile of prompts.
Anyone can ask a chatbot to write a cold email. The output is generic, unverifiable, and unsafe to send at volume. This suite works differently: it is a pipeline of specialized agents, and each one has a contract.
- icp-definition
- market-research
- positioning
- channel-strategy
- apollo
- clay
- web
- data-enrichment
- lead-scoring
- email-sequence
- linkedin-outreach
- cold-calling
- multi-channel
- reply-classification
- objection-handling
- call-prep
- handoff
- kpi-reporting
- forecasting
- icp-refinement-loop
every connector is a quality gate, not an arrow
The system produces outreach a professional SDR manager would sign off on, at AI speed, with an audit trail.
Real methodology, encoded as executable rules.
The skills do not just know about sales frameworks. They enforce them. The ICP scorecard is a fixed, weighted 100-point rubric applied identically to every lead.
Sales-ready. Gated further by BANT and CHAMP checks before any lead is called hot.
Warm. Enters outreach, watched for a trigger that moves it up.
Deprioritized. Never blasted just because the list exists.
Copywriting is equally constrained: the CCQ framework (Context, Connection, Question) with Pain-Trigger-Outcome openers, mobile-first formatting, and a hard ban on cliche openers like “Hope this finds you well.” Touch 1 must open with the prospect’s verified pain, not our product.
Consistency at scale. Lead #30 gets the same rigor as lead #1, something even good human teams drift on by Friday afternoon.
The call that shaped the campaign: skipping email entirely.
The obvious move for Barbershops Canada was a cold email sequence, and the suite has a full skill for it. I did not use it, for three reasons.
The audience is not at a desk.
Barbershop owners live on Instagram, which is their portfolio, and Facebook, which is their booking inbox. An email from a stranger is invisible. A DM is native.
The infrastructure gate said no.
Cold email requires a warmed dedicated domain, 14 days minimum. For a 30-lead local campaign that is the wrong cost-benefit, and the system's own email-infrastructure gate makes this explicit instead of letting you skip it.
Compliance is different in Canada.
CASL is stricter than CAN-SPAM. Conversational DMs with a soft opt-out and no link in the first message fit both the law and the culture.
Instagram and Facebook Messenger DMs routed to each prospect’s strongest channel, call-first for Tier-1 leads with a verified phone number. All of it in founder voice (“I’m building this, be one of the first”), leading with Crewm8’s wedge: the missed call during a haircut is lost revenue.
The system recommends, but strategy decides. Knowing when not to use a tool you built is the actual GTM skill.
What’s actually in the CRM: the receipts.
Everything the system produced lives in Attio as versioned, timestamped records, not in a chat log. Here is one Tier-1 lead, City Barber in Edmonton, moving through the whole machine, then the score rationale and the DM that came out the other side. All from the live workspace, phone number masked.
- found on
- Booksy listing + open web
- trigger
- booking-friction-review
- phone
- pending enrichment
- cost
- $0 · no paid database
Sourced from a WebSearch query, provenance cited.
- phone
- +1-780-758-•••• · [H] on citybarber.ca
- domain
- citybarber.ca (owned)
- address
- 5590 Windermere Blvd, Edmonton
- hook
- cited review: online booker left waiting
Contactable. The hook is quoted, not invented.
- score
- 80 / 100 · Tier-1
- breakdown
- pain 20.5 · trigger 18.0 · wtp 14.4 · reach 12.0 · ttv 7.5 · strategic 7.5
- gates
- all 4 SAL gates passed
Same 100-point rubric as all 30 leads.
- record
- Attio · Companies
- record_id
- 5ad52b5e-db38-4b52-9fe0-d993aa00b4b8
- created
- 2026-06-19
- state
- live · versioned · auditable
The row you can open in the CRM today.
“Scored vs ICP 100-pt scorecard (run web-ca-recept-20260619). Confidence: H. SAL gates: all 4 passed. Need = confirmed (the only lead with a real trigger: a public review about an online-booking client left waiting). Phone [H] confirmed. Strongest-evidence lead.”
Every one of the 30 leads carries a note like this: the score, the confidence, which gates passed, and the evidence behind it.
Notice what the DM is doing. It opens with a real, citable observation the enrichment stage captured with its source, connects it to the exact pain Crewm8 solves, and ends with a low-pressure diagnostic question. CCQ, executed. No link, no pitch, CASL-safe.
30-lead list pushed to Attio
All 30 leads scored, each with a rationale note
All 30 channel-routed DM drafts logged
List to scores to 30 finished drafts in two hours and two minutes, all logged before anything is sent.
- 30 barbershops in the Barbershops Canada list, created June 19, 2026
- 7 Tier-1 leads at 75 to 80 / 100 (23%, skewed high because the list was pre-filtered to ICP-fit shops), the majority Tier-2 warm, remainder Tier-3
- One channel-routed DM draft logged per prospect: IG, FB Messenger, or call-first
- v1 is never overwritten. Updates append as v2, so the copy's evolution stays auditable
This is the difference between “I used AI for outreach” and “I ran a professionally instrumented campaign.” Every claim on this page traces to a timestamped record.
Guardrails are the feature, not the fine print.
The most powerful thing the skills do is refuse.
The DM helper refuses to invent.
If a logged DM exists, it returns it verbatim. If none exists, it does not generate one. It asks for research on that prospect and waits. Human research in, AI drafting out. That single rule is why every DM in the campaign cites something real.
Campaigns pause themselves.
Reply, bounce, and complaint rates are watched against hard thresholds, per Google and Microsoft's 2024 bulk-sender rules. Open rates are treated as noise, because Apple Mail Privacy makes them meaningless, and the system knows this.
Every block ships with an override.
A shared Clarification Protocol across all 32 skills means the AI never dead-ends. When a rule blocks an action it presents the options, the risk, and a recommendation. The human decides.
Compliance is encoded, not remembered.
CASL, CAN-SPAM, GDPR, LinkedIn rate limits, TCPA calling rules, and RFC 8058 one-click unsubscribe each live inside the skill that needs them.
AI leverage without guardrails is a liability generator. This system’s default answer to “should the AI just do it?” is only if it can prove it is safe, which is exactly what makes it usable for a real company’s reputation.
The honest version.
The campaign ran. Here are the real numbers, the honest reason most interested leads did not convert, and what is still an estimate.
- A 32-skill GTM system, built and operational, spanning the full funnel
- A 30-lead campaign sourced, enriched, scored, and fully drafted in about one working day; the core pass took roughly 2 hours by CRM timestamps
- 11 of 30 leads replied, a 37% reply rate against the system's own 6% multi-channel floor
- 9 replies turned into real product interest; 5 of Canada's top barbershops onboarded into the beta trial cohort
- The scorecard called its shots: all 5 conversions were Tier-1 leads, 5 of the 7 the agent flagged before outreach began
- 100% of DMs carry a verified, source-cited personalization hook. Zero template blasts
- Time saved: 3 to 5 solo working days per 30-lead campaign.Assumes 30 to 45 minutes per lead for manual research, scoring, and a personalized draft of this quality.
- Repeatability: the marginal cost of campaign #2 is near zero.The ICP, scorecard, copy frameworks, and CRM plumbing are already encoded. Only the research per lead is new.
Price, not product. A voice agent is expensive to run: even after every cost optimization the floor was about $150 USD per shop per month, and that was our own first-month cost. We priced at $200 to keep a margin, and most interested shops walked. The ones who paid were exactly the leads the agent had scored Tier-1 before a single DM went out.
GTM strategy
ICP definition, channel selection against the grain of my own tooling, wedge-based positioning, and compliance-aware sequencing.
AI systems engineering
32 composable skills with typed contracts, hard gates, provenance tracking, and CRM integration through Attio via MCP.
Judgment
Knowing where the human belongs in the loop (research, strategy, the final send) and encoding that boundary so the system cannot drift past it.
The system proved two things: speed with rigor, and predictive scoring. Every paying customer came from the Tier-1 band the rubric flagged before outreach began. The icp-refinement-loop skill now has 30 real outcomes to recalibrate against, and the next problem to solve is pricing, not pipeline.
Want a GTM engine like this, or the person who builds them?
All campaign data referenced on this page is from Crewm8’s live Attio workspace. Phone numbers and personal contact details are masked. Estimates are labeled as such throughout.