GTM Engine
Revenue operations and automated outbound · Turgon AI, 2026 to now
I built a campaign factory. One problem statement goes in. A live campaign comes out, running across email, calls and LinkedIn, with every send and reply logged.
I run the BDR team at Turgon AI and built the system it runs on in Claude Code. In one quarter, the share of held meetings that turned into deals went from 18% to 31%.
- 18% → 31%held meetings that became deals, quarter over quarter
- 20% → 54%of bookings from cold outbound, in one quarter
- 40 inboxeson 10 sending domains, three channels, one scorecard
The factory
Claude Code builds every run. People approve spend and launches, read ten sample emails, and talk to prospects.
- Problem
- Signal-picked accounts
- Research check
- Tier A / B / C
- People by persona
- Verified emails
- Email gate
- CRM first
- Calls
- Reply
- Slack alert
- Daily scorecard
- Friday review
How a campaign gets built
Six steps. Each one ends on a check, so a bad list stops before it sends.
- 1Intake
One problem, one hypothesis, and the rule for keeping or killing the campaign, written before launch.
- 2Account book
Accounts picked by a dated buying signal, then tiered: A has two or more ways in, B has one, C is signal only. The tier decides the channels.
- 3People
Personas in pick order, found on the tools already paid for first. Emails pass a seven-part gate, or the contact goes to calls only.
- 4Messaging
One sequence per campaign, with a sourced opening line per person. A person reads ten rendered emails before anything loads.
- 5Send
Contacts land in the CRM first, then load interleaved by company. The first 50 sends are a probe.
- 6Measure
A scorecard refreshes every weekday morning, one row per run. Friday decides: keep, change, spread or kill.
Rules, and what broke to earn them
Every rule in the system exists because something went wrong once.
- Automate the variables, never the claimsA personalised field with no source falls back to a safer line before upload. Every claim comes from an approved list.
- The loader refuses bad rows in code“Verified only” was written in the playbook and unverified addresses loaded anyway. Those bounced at 36 to 39%, against 1% for verified ones. A rule in prose wasn’t enough.
- At most two people per company per daySix people at one company got the first email inside nine minutes, and their mail gateway blocked the whole sending domain.
- The person’s name must be in the addressA “verified” address for a CIO turned out to be the service desk. It opened a ticket.
- Preflight, probe, then a daily bounce watchOne early run was paused at 11.8% bounces. Now a preflight projects the bounce rate before launch, and a daily watch pauses any run at 4%. Back-tested, the preflight would have failed that run at 10.8%.
- Never report opens or clicksBots make about 80% of opens and 98% of clicks. A campaign is judged on meetings per 100 contacts and the pipeline behind them.
The stack: one job per tool
- Claude CodeBuilds every run and holds the scripts and scheduled tasks: 29 skills, about 9,800 lines, plus six internal apps.
- DeeplineOne connection to every other tool: loads sequences, writes the CRM, runs the email and phone waterfall and the research plays.
- Instantly · Apollo · HeyReachCold email, call sequences and LinkedIn. One email lane per contact, and cold never sends from the main domain.
- ParallelThe light check on every account before any money is spent on contacts.
- HubSpot · Notion · SlackHubSpot is the source of truth for every send, reply, call and deal. Notion holds one row per run. Slack gets every reply within the minute.
What moved the number
- IncentivesRedesigned the comp plan to pay for cold meetings and for deals that progress, not just meetings booked. Cold went from a fifth of bookings to more than half.
- TargetingAnalysed more than 6,000 logged dials by persona. The best persona booked a meeting every 35 dials and the worst every 202, so I rebuilt the prospecting lists around the ones that book.
- Context layerOne living brief per account: signals, tech stack, people and the business case, source-tagged. Reps and AEs read the same page.
What it does and doesn’t do
The factory solves supply: the list, the research, the verified contacts and the first draft. Reps still win the meeting on calls and LinkedIn, so the system’s job is to feed those conversations, not replace them.
Three reps doing the work of ten.
Same engine, other markets
The inputs are a target list and a reason to reach out. Swap enterprise data leaders for beverage brands and the same workflow builds a sponsor pipeline, which is how I ran brand partnerships for Secret Disco Society.