Signal-Based Outbound with AI Agents: The Complete Guide for B2B SaaS
How to replace spray-and-pray cold outreach with signal-based outbound: the four buying signals worth monitoring, how AI agents research and draft messages, the deliverability and legal rules you can't ignore, and a step-by-step build with human approval at every send.
By Alexandre Contador, Fractional CGO & GTM Advisor

Cold email didn't stop working. Cold email that has no reason to exist stopped working. The inbox is crowded with sequences that open with a flattering line about "your recent growth" and then pitch the same product to everyone.
Signal-based outbound flips the logic: don't start with a list and invent a reason — start with a real event and reach out to the people it affects. AI agents make that practical, because the monitoring, research and first-draft writing are exactly the repetitive, evidence-gathering steps they're good at. This guide shows how to build it end to end, safely.
Why signals beat static lists
A static list tells you who might buy. A signal tells you why now. That difference changes every part of the funnel:
| Static list outreach | Signal-based outreach | |
|---|---|---|
| Reason to reach out | Generic problem statement | A specific, verifiable change |
| Message | Template with merge fields | One or two sentences referencing the event |
| Volume | High | Lower, targeted |
| Risk | Spam complaints, burnt domain | Lower complaint risk if relevance is real |
| Work required | List building | Monitoring and research (automatable) |
None of this removes the need for a clear ICP and offer. Signals tell you when; your ICP and positioning determine whether the conversation should happen at all. The free ICP scorecard is a quick way to check you're pointing the agents at the right accounts.

The four signals worth monitoring
1. Hiring signals
What it is: a company posts a role that implies the problem you solve (a first RevOps hire, a "Head of Growth", several SDRs, a data engineer).
Why it works: headcount is budget and intent made public. A new head of a function usually reviews tools and agencies in their first 90 days.
How to detect it: job boards and company careers pages, filtered by title keywords and seniority; LinkedIn job posts; data providers with job-posting feeds.
Message angle: "Saw you're hiring a [role]. Teams building that function usually hit [specific problem] in the first quarter. We [one concrete way you help]."
2. Funding signals
What it is: a seed, Series A or later round announced.
Why it works: new capital creates pressure to deploy it and board-level targets for growth.
How to detect it: funding databases, press releases and newsletters, news alerts.
Message angle: reference the round only if it's relevant to what you do. Tie it to the stage-specific problem ("post-round, the question usually shifts from finding customers to proving repeatable channels").
3. Job-change signals
What it is: a champion, past customer or user moves to a new company — or a new decision-maker joins a target account.
Why it works: people bring trusted tools with them, and a warm relationship converts far better than a cold name.
How to detect it: CRM contact monitoring, LinkedIn alerts or data providers with job-change feeds.
Message angle: congratulate, mention the shared history, and offer help in the new context. Keep it human.
4. Technology install and removal signals
What it is: a company adds, switches or drops a tool (a CRM, analytics stack, a competitor's product).
Why it works: stack changes create integration needs, migration pain and replacement windows.
How to detect it: technographic data providers and website technology detection.
Message angle: "Noticed you moved to [tool]. The usual gap after that switch is [specific]. We [fit]."
How an AI-agent system works
Anthropic's guidance on agents distinguishes workflows (predefined code paths) from agents (models directing their own tool use) and recommends starting with the simplest design. Signal-based outbound is a good fit for a workflow with agent-powered steps, because the overall flow is fixed even though the research inside it varies.
The pipeline:
- Monitor — scheduled checks for signals on a defined account list.
- Qualify — score the account against your ICP; discard misfits.
- Research — in parallel, gather the facts about the signal, the company and the right person.
- Draft — write a short message that cites the signal and one verified fact.
- Verify — check every claim against the research notes; reject anything unsupported.
- Review — a person approves, edits or kills each item.
- Send — through your sequencer, under deliverability limits.
- Learn — log replies and outcomes to improve signals and angles.
This maps to patterns such as routing (classify the signal type), parallelisation (research several accounts at once) and evaluator-optimiser (a verifier that scores each draft against a rubric).
A practical tool stack
| Layer | Purpose | Examples |
|---|---|---|
| Data and signals | Companies, contacts, hiring, funding, tech installs | Vibe Prospecting, Apollo, Crunchbase, BuiltWith, job-board feeds |
| Orchestration | Scheduling, routing, calling the model | Claude Code (scheduled tasks and MCP), Make, n8n |
| Reasoning | Research summaries and drafting | Claude, other frontier models |
| Sequencing | Sending, throttling, reply handling | Lemlist, Instantly, Smartlead |
| CRM | Source of truth, suppression | HubSpot, Attio, Salesforce |
Browse independent reviews in the Alex Stack Suggestion directory. If you want to build the pipeline in Claude Code, the free Outbound in Claude Code playbook covers a daily ICP lead, enrichment and sequencing engine, and Outbound in Grok shows the same idea with scheduled Grok tasks. For the connection layer, see MCP for marketers.
Writing messages that earn replies
The agent drafts; your rules decide what's allowed. Give it a message spec:
- Length: 40–90 words; one idea; one question.
- Structure: signal → relevance → specific offer → low-friction ask.
- Proof: one verifiable fact about their company; never invent.
- Tone: direct and human; no hype, no flattery, no fake familiarity.
- Prohibited: claims you can't source, "I noticed your recent growth", fake urgency, misleading subject lines, pretending a message is a reply.
Example (hiring signal):
Subject: Your RevOps hire
Hi Sam — saw you're hiring your first RevOps manager. When that role lands, the first project is usually cleaning up lead routing before the Q1 push. We've built a short routing audit that flags the three biggest leaks in a week. Worth a 15-minute look?
Every number or claim in that message must trace back to the research notes. If it can't, the verifier rejects it. For reply-ready starting points, the free 50 cold email and LinkedIn DM templates and the Signal-Based Outreach Playbook are good scaffolds.
Personalisation that isn't creepy
- Reference business events, not personal details.
- Use public, professional information only.
- If the research is thin, send a simpler, honest message rather than inventing relevance.
Deliverability: the constraint that decides everything
Signal-based volume is lower, but deliverability rules still apply. Google's sender guidelines set the bar:
Source: Google email sender guidelines
Practical checklist:
- Use separate sending domains from your primary domain, and warm mailboxes gradually.
- Set up SPF and DKIM (and DMARC) correctly; Google recommends 2048-bit DKIM keys where supported.
- Include one-click unsubscribe headers on marketing and subscribed messages and a visible unsubscribe link.
- Keep list quality high: verify addresses, suppress bounces and unsubscribes across all tools.
- Watch spam complaint rate in Postmaster Tools; pause a mailbox that trends toward 0.1%.
- Cap daily volume per mailbox and ramp slowly.
- Never send a batch you haven't reviewed.
The legal basics (not legal advice)
- United States. The FTC's CAN-SPAM guide says the law "makes no exception for business-to-business email." Requirements include accurate header information, non-deceptive subject lines, a valid physical postal address, a clear opt-out, and honouring opt-outs within 10 business days. Each violating email can carry penalties of up to $53,088, per the FTC's current figure.
- EU, UK and other regions. Rules on consent, legitimate interest, data sourcing and objection rights differ by country and by recipient type. Take local legal advice, document your legitimate-interest assessment where you rely on it, and honour objections quickly.
- Data sources. Know where your contact data comes from and your right to use it.
- Platform terms. LinkedIn and other platforms restrict automation. Don't rely on tactics that violate them.
Measuring signal-based outbound
| Metric | Why it matters |
|---|---|
| Meetings booked per 100 contacted | Core efficiency metric |
| Positive reply rate | Quality of signal and message |
| Pipeline and revenue per signal type | Which signals deserve more investment |
| Time from signal to first touch | Speed advantage |
| Complaint and bounce rate | Deliverability health |
| Approval rate and edit distance of drafts | Quality of the agent's output |
Tag every account with its signal type in the CRM so you can compare hiring vs. funding vs. job-change vs. tech-install results after a quarter. Double down on the winners, retire the rest.
A 30-day build plan
| Week | Goal | Tasks |
|---|---|---|
| 1 | Foundations | Define ICP and offer; pick two signals; build a 200-account target list; configure domains and authentication |
| 2 | Monitoring and research | Set up signal checks; build the research prompt and fact-sheet format; test on 20 accounts |
| 3 | Drafting and verification | Write the message spec; add the verifier step; run a human review of 50 drafts and tune |
| 4 | Controlled launch | Send 100–200 approved messages; track replies, complaints, meetings; review and decide what to scale |
Common mistakes
- Signal without relevance. Reaching out because a company raised money, with no link to your offer, is just a prettier spray.
- Letting the agent send unreviewed. One hallucinated fact can burn a prospect and a domain.
- Scaling before the basics. If a 100-contact test doesn't produce replies, 10,000 won't.
- Ignoring suppression. Unsubscribes and bounces must sync across every tool.
- Single-signal dependence. Combine signals (hiring + tech change) for stronger intent.
- No feedback loop. Without outcomes by signal type, you can't improve.
Related reading
Frequently asked questions
Signal-based outbound means contacting accounts because something changed — they hired for a relevant role, raised funding, a champion changed jobs, or they installed or removed a technology — instead of emailing a static list. The signal gives you a true, specific reason to reach out now.
The four most dependable are hiring (a role that implies the problem you solve), funding (budget and urgency), job changes (a past champion or buyer moves company) and technology installs or removals (a stack change that creates or kills a need). Test each against your own win data, because the best signal depends on your product.
Yes, as a drafting step, provided every factual claim is traceable to the research notes and a person approves each batch before sending. Fully autonomous sending at volume risks deliverability damage, factual errors and brand harm.
Rules vary by country. In the US, the FTC states that CAN-SPAM makes no exception for business-to-business email: you need accurate headers, non-deceptive subject lines, a physical address, a clear opt-out and you must honour opt-outs within 10 business days. The EU, UK and other regions have separate consent and legitimate-interest rules, so take local legal advice before running campaigns there.
There is no universal number. Deliverability depends on domain reputation, authentication, list quality and complaint rate. Gmail requires spam rates below 0.3% and recommends staying below 0.10%; senders above 5,000 messages a day face stricter rules. Start with low volumes on warmed, authenticated domains and scale only while complaint and bounce rates stay low.
Free resources & related work
Want a second opinion on your GTM?
30 minutes, no pitch — just a direct conversation.
Next in AI & Agents
The Ultimate Guide to GEO and AEO: Getting Cited by ChatGPT, Gemini, Claude and AI Overviews
What generative engine optimization (GEO) and answer engine optimization (AEO) really are, what the research and Google's own documentation say works, how big AI referral traffic actually is, and a step-by-step plan to measure and improve your brand's visibility in AI answers.
Keep reading
All AI & Agents guides →
The Ultimate Guide to GEO and AEO: Getting Cited by ChatGPT, Gemini, Claude and AI Overviews
What generative engine optimization (GEO) and answer engine optimization (AEO) really are, what the research and Google's own documentation say works, how big AI referral traffic actually is, and a step-by-step plan to measure and improve your brand's visibility in AI answers.

MCP for Marketers: The Complete Guide to Connecting AI to Your Stack
What the Model Context Protocol is, which ad platforms ship official MCP servers, what marketers can actually automate with it, the security risks nobody mentions, and a four-stage path from "ask" to "act" that keeps humans in charge.

Agentic GTM Workflows: 25 Claude and Claude Code Automations for Marketing, Sales and Ops
A practical guide to building agentic go-to-market workflows with Claude and Claude Code: the difference between workflows and agents, five design patterns, 25 concrete automations across marketing, sales and ops, and a maturity model for rolling them out safely.