Tools are the wrong unit for learning GTM engineering, so we teach them inside the machines they serve. This is the stack Sparq Labs runs for clients and the one you'll build on, organised by machine.
Signals feed the list, the list feeds the agent, the agent feeds the copy, the sequencer sends, replies come back, and the dashboard tells you what to cut and what to scale.
Scroll sideways on a phone to see the whole diagram.
Product images come from each company's own site. We're not sponsored by any of them; we use them because they work on client accounts.
Email that lands. Domains, mailboxes, warm-up, limits.
Inboxkitinboxkit.com ↗Mailbox infrastructure for cold email: Google Workspace and Microsoft accounts on your own secondary domains.
How we use it · Week 1 sets up mailboxes here on secondary domains, with DNS records, warm-up and a cost sheet for your volume.
Google Workspaceworkspace.google.com ↗Google mailboxes.
How we use it · Half of the Azure and Google split, so one provider's filters never decide your whole send.
Microsoft 365microsoft.com/microsoft-365 ↗Microsoft mailboxes.
How we use it · The other half of the split. Mailboxes are warmed for weeks before any cold send.
ICP written as signals, cheap steps before AI credits.
Clayclay.com ↗The table where data providers, scrapers and AI steps run in one place.
How we use it · The core of M2: ICP as signals, an enrichment waterfall that runs cheap deterministic steps before a single AI credit, and tiers per account.
Apolloapollo.io ↗A large contact and company database.
How we use it · The first pass on firmographics and contacts before the waterfall narrows the list.
Prospeoprospeo.io ↗Email finding and verification.
How we use it · A step in the waterfall to find and verify work emails, so bounces never touch your domains.
Apifyapify.com ↗Ready-made scrapers and actors.
How we use it · Pulls signals that live on websites and job boards, like open SDR roles, into the Clay table.
A first line written from evidence, for every account.
Claudeclaude.ai ↗Anthropic's AI model.
How we use it · The M3 research agent: it reads each account's evidence and writes a first line from it. We review twenty of its lines together before anything sends.
Three plain-text steps by segment, tested against replies.
EmailBisonemailbison.com ↗A sequencer built for sending at volume.
How we use it · Runs the three-step cadence across many mailboxes. Students in the community use it with Prospeo and Inboxkit.
Smartleadsmartlead.ai ↗A sequencer with unlimited mailboxes, warm-up and a master inbox.
How we use it · Inbox rotation, sending limits, and the dashboard we keep open during live builds.
Instantlyinstantly.ai ↗A sequencer with built-in warm-up.
How we use it · A simpler setup for smaller senders, and warm-up while new mailboxes age.
Replies answered in minutes; calls booked; LinkedIn alongside.
n8nn8n.io ↗Workflow automation you can self-host.
How we use it · The glue in M5: classify replies, draft a first response, route positives to a human within minutes.
HeyReachheyreach.io ↗LinkedIn outreach across many sender accounts.
How we use it · The LinkedIn channel beside email, run without manual messages.
HubSpothubspot.com ↗CRM.
How we use it · Where booked calls land, and where M6 reads attribution for the weekly scoreboard.
The scoreboard someone senior actually reads.
No extra tool. M6 reads sent, replies and opportunities straight from the sequencer, and booked calls from the CRM.
How we use it · A weekly scoreboard: what to cut, what to double, which segment earned the next send.
HubSpothubspot.com ↗CRM.
How we use it · Where booked calls land, and where M6 reads attribution for the weekly scoreboard.
Deliverability, list quality and reply speed decide more outcomes than any prompt. The AI steps are maybe a fifth of the work. Week one gives you a cost sheet for your own volume, and the waterfall order keeps the credit bill down.
Never cold email from your main domain.
Deterministic filters before a single AI credit.
The agent writes only from what it found.
Positive replies answered fast.