Book qualified meetings
from the right signals.
AI SDRs ship volume. The Outbound Automation agent ships qualified conversations — signal, fit, timing, hypothesis, draft, approval, learn.
Find 6 high-fit accounts based on today's signals and draft personalized openers.
I'll break this into agent steps and execute with your guardrails.
Queued…
Queued…
Claude Code can write the email. It cannot run the play.
Generic agents do not score signals. They do not check ICP fit. They do not remember which plays converted. They do not respect a do-not-contact list. They draft. The harness is the rest.
“Personalization without a hypothesis is trivia.”
Write a cold email for Acme. Make it sound personalized.
Subject: Quick question about Acme's growth plans
- I noticed Acme is doing great work in your space and wanted to reach out.
- We have helped similar companies scale their pipeline.
- Would you be open to a 15-min chat next week?
Static lists and a sequencer vs. one signal-to-play layer.
The list is six months old. The sequencer fires on schedule. Personalization tokens replace relevance. Replies get triaged manually. Plays never retire.
ICP B · cold · 4,212 contacts
Last refreshed 38 days ago
Step 1 · email
Step 2 · email
Step 3 · LinkedIn
acme · ?
globex · ?
initech · ?
"Write a cold email for Acme"
No signal data · no CRM
What replaces volume motion.
Four operating principles. Each one carries a method anchor and a piece of product evidence.
Signals are play inputs, not magic triggers.
A signal only matters when it is fresh, relevant, tied to ICP fit, and strong enough to support a reason to reach out. Treat signal selection as a first-class design step.
Method anchor — Clay-style signal selection
- hiring growth rolet½ 14d
- tech stack changet½ 21d
- series fundingt½ 30d
- pricing visitt½ 3d
From signal detected to play memory.
The agent watches signals, scores fit and timing, enriches the account, builds a relevance hypothesis, drafts outreach, runs QA, routes for approval, classifies replies, and writes outcomes to memory.
Each run writes outcomes to memory. The next run starts with the prior decision graph and review boundary already loaded.
Production-grade agents need more than a clever prompt. Each layer below is required for governed autonomy.
What the agent reads before every play.
Every Metaflow agent is grounded in a domain-specific skill file — a structured operating procedure that defines inputs, workflows, evaluation criteria, anti-patterns, and output contracts.
The skill file is editable, versioned, and inspectable. It is not a hidden prompt.
# Mission Create qualified conversations from real buying signals — not from list volume. Signal selection. Fit and timing. Enrichment. Relevance hypothesis. Governed drafting. Human approval. CRM coordination. Reply classification. Play memory. ## Optimizes for - qualified conversation rate per play - time-to-meeting from signal detection - play attribution accuracy ## Does not promise - guaranteed meetings - "AI SDR" replacing the operator - set-and-forget outbound at scale
Drafts are scored before they ever land in an inbox.
The agent does not send. It scores every draft and routes anything below threshold to an operator with the failing element flagged.
Where the agent stops or hands back instead of guessing.
- Account is on do-not-contact in any source.
- Signal stale beyond freshness window.
- Relevance hypothesis incomplete.
- Audience saturation flagged on the play.
- Watch signals
- Score ICP fit and timing
- Run enrichment waterfall
- Compose relevance hypothesis
- New play candidates
- New signal sources
- Audience expansion within ICP
- Sequencer cadence changes
- New play launch
- Drafts before send
- Cross-channel push
- CRM field overwrite
System evidence, not feature cards.
Five artifacts the agent produces, scores, or maintains. Hover to pause.
Signals scored before any outreach.
Audience fit, baseline conversion, freshness window, execution difficulty. Saturated or stale signals retire automatically.
| Signal | Half-life | Fit | Status |
|---|---|---|---|
| hiring_growth_role | 14d | 0.84 | Active |
| tech_stack_change | 21d | 0.81 | Active |
| series_funding | 30d | 0.78 | Active |
| pricing_page_visit | 3d | 0.72 | Active |
| general_award | 7d | 0.31 | Retired |
Clay-style signal selection
Common signals become noisy. Advantage comes from unique signal selection, strong enrichment, and fast execution.
Signal library scored on freshness, fit, baseline conversion, saturation risk before plays are designed.
UnifyGTM signal-based outbound
Replace static lists with real-time signal-led plays. Start with one strong signal before scaling.
Plays designed first — signal, ICP filter, enrichment, message angle, routing, measurement.
Common Room signal scoring
Signals scored by fit, frequency, baseline conversion, pipeline potential, execution difficulty.
Signal scoring runs before any outreach is composed. Generic activity is filtered out.
ColdIQ relevance discipline
Personalization is a relevance hypothesis: why this account, this person, now, this offer.
The four-question test ships with every draft. Trivia personalization does not pass.
Production-grade agent governance
Serious agents need explicit instructions, evals, traces, human oversight, and stopping conditions.
External actions require human approval. Confidence drives routing. Plays retire on rule, not vibe.
Reforge-style growth loops
Execution should produce learning that improves the next cycle.
Play-level attribution and objection memory persist. Future drafts pre-empt patterns the team already saw.
Not another AI SDR. A governed signal-to-play layer.
Four contenders. One operating layer that compounds. Hover the Metaflow column for product evidence.
| Dimension | Claude Code, Cursor Generic agents in chat windows. | AI SDR products Volume-led outbound automation. | Outbound agency Outsourced humans with sequencer access. | Metaflow agent Agentic operating layer. |
|---|---|---|---|---|
| Sourcing | Whatever you paste in. | Static lists, bulk enrichment. | Whatever the agency supports. | Real-time signal-led plays, scored before outreach. signals.watch Hiring + tech-change + funding stacked on Acme. |
| Personalization | Generic prompts, surface tokens. | Token-based with light AI rewriting. | Variable across SDRs. | Four why-questions answered explicitly per draft. hypothesis.compose Why Acme · why VP Marketing · why now · why offer. |
| Memory | Resets every chat. | Limited, often per-sequencer. | Lives in slack and the SDR. | Workflow memory persists plays, retired plays, objections. memory.write “VP post-funding · outcome-led offers” — persisted. |
| Attribution | None. | Reply rate, open rate. | Meeting volume. | Play-level: signals + audience + message → qualified conversations. play_attribution.json Hiring play · 14 sends · 4 qualified · 28% rate. |
| Compliance | Unbounded. | Variable. | Process-dependent. | DNC enforced. Refusal conditions explicit. Saturation flagged. review.queue Stark Inc · DNC verified · blocked. |
Repeatable plays the agent runs end-to-end.
Each play has the same shape: signal, fit, timing, enrichment, hypothesis, draft, QA, approval, push, classify, learn.
Hiring signal play
A target persona is hiring into a function the buyer would build with your category.
Outcome
Drafts grounded in speed-to-hire pressure, scoped to a 90-day mandate.
Hired 3 growth roles · 14 days · post-funding
Design your first signal-led play in one focused session.
A 30-minute working session with a Metaflow operator. We pick the strongest signal you have access to, define the play, score the audience, propose a hypothesis, and outline the approval boundary.
- 01Scored signal library of the top 5 signals available to you.
- 02A defined first play: signal, ICP filter, enrichment, message angle.
- 03Relevance hypothesis template against your top-fit account.
- 04A written 1-page memo with the next 3 plays we would design.
- hiring growth rolet½ 14d
- tech stack changet½ 21d
- series fundingt½ 30d
- pricing visitt½ 3d
Hired 3 growth roles · 14 days · post-funding
A focused diagnostic. No slides. Walk away with a designed play whether or not we work together.