Cold Outbound Workflow: Research, Personalise, Sequence with an n8n Sales Agent
An n8n sales agent for outbound does the research and writing a good SDR would do for each prospect, at a volume a person cannot sustain, inside limits that protect your domain. The workflow builds the list from your ICP, researches each account (site, news, hiring, tech), writes a first line and a short email grounded in that research, sequences with reply and bounce detection, and routes replies to a rep.
Written by Max Zeshut
Founder at Agentmelt · Last updated Sep 11, 2026
The problem
Outbound is either generic at volume (ignored, and damaging to your domain reputation) or personal at a volume of ten a day. SDRs spend most of their time on research and copy, not conversations.
What changes when it runs
Every email in the sequence references something true and specific about the prospect. Volume stays within deliverability limits, replies are detected and handed to a person within minutes, and the team measures reply and meeting rate per segment and per angle instead of guessing which messaging works.
Trigger, then 8 steps
Trigger
Schedule Trigger (daily batch) + reply webhooks
A daily run selects the next batch from the list; the sending tool's webhooks report replies, bounces and unsubscribes in real time.
Build the batch
HTTP RequestPulls prospects matching the ICP from Apollo or Clay that are not in the CRM, not suppressed, and not contacted in 90 days. Daily batch size is capped per sending mailbox.
Research each account
AI AgentReads the company site, recent news, job postings and tech stack (via enrichment and a web-fetch tool) and writes a five-line research note: what they do, what changed recently, a plausible trigger for your offer.
Write the first line and email
AI AgentFrom the research note and your messaging playbook, writes the opening line and a sub-80-word email for the chosen angle. Structured output enforces length, no adjectives about your product, and one clear ask.
Quality gate
IFEmails are checked for specificity (must reference the research note), banned phrases, and length. Failures are rewritten once, then dropped from the batch rather than sent generic.
Load into the sequencer
HTTP RequestPushes the prospect and the personalised steps into Instantly, Smartlead, Lemlist or Outreach, which handle sending, warm-up and inbox rotation.
Detect replies and classify
WebhookEvery reply is classified: interested, question, not now, wrong person (with a referral), unsubscribe, out of office. Sequences are stopped on any human reply.
Hand to a rep
SlackInterested and question replies go to the rep with the thread, the research note and a suggested response; 'not now' is scheduled for a follow-up; referrals create a new prospect.
Report by angle and segment
Schedule TriggerWeekly: sent, replies, positive replies, meetings by segment, angle and mailbox; deliverability metrics from the sequencer; the losing angle is retired.
Data it touches
- Prospect data (Apollo, Clay, LinkedIn Sales Navigator exports)
- Company sites, news, job posts (web fetch)
- CRM for suppression and logging
- Sequencer (Instantly, Smartlead, Lemlist, Outreach)
- Messaging playbook and angles
Guardrails
- Daily volume per mailbox is capped and warm-up is respected; the workflow never sends outside the sequencer's limits.
- Emails that fail the specificity check are not sent; generic filler is worse than silence.
- Any human reply stops the sequence immediately; unsubscribes are honoured across all lists.
- Suppression lists (customers, open deals, competitors, opt-outs) are applied before research spends any credits.
Personalisation that survives the quality gate
'I saw you are the VP of Sales at Acme' is not personalisation, it is a mail merge. The research step produces a note with something that changed — a new office, a job post for the role your product replaces, a product launch, a quote from the founder — and the copy step must reference it. The quality gate checks that it did. Emails that cannot be made specific are dropped. Reply rates on batches that pass this gate are typically several times those of template outbound, and the domain reputation stays intact because the volume of low-quality mail is zero.
Deliverability is the constraint, not the model
The model can write a thousand emails an hour; a mailbox can send 30–50 a day without harm. The workflow treats the sequencer's limits as hard constraints, rotates across mailboxes it is given, and reports bounce and spam-complaint rates weekly. If those move, volume drops automatically. Outbound automation that ignores this burns the domain in weeks.
Angles as experiments
Each batch is tagged with the angle — the problem you led with — and the segment. The weekly report shows reply and positive-reply rate per angle per segment. After a few weeks the workflow retires losing angles automatically and the team writes new ones. This is the part that compounds: the messaging gets better because the data says what worked, not because someone had a feeling.
Tools in the stack
| Tool | Role in this workflow |
|---|---|
| n8n | List building, research loop, quality gate, reporting |
| Claude | Research notes and copy, with structured output |
| Apollo / Clay | Prospect data and enrichment |
| Instantly / Smartlead / Lemlist | Sending, warm-up, deliverability |
| HubSpot / Salesforce | Suppression and logging |
Want this running without building it?
Automation workflow
$297/month
We set up, host and maintain this workflow on n8n and connect it to your tools. Setup included, cancel monthly, you keep the JSON.
Custom build
$4,000–7,000 one-time
Your systems, your rules, your edge cases. A one-off build on Claude and n8n, delivered with documentation and a walkthrough.
Covers up to 2,000 new prospects a month with one sequencer; data and sequencer subscriptions are yours. LinkedIn steps, multi-language, or a custom research tool set are a custom build.
Frequently asked questions
Is this compliant with anti-spam law?
B2B cold email is lawful in most jurisdictions with a clear identity, a working unsubscribe and honoured opt-outs, which the workflow enforces. Regions with stricter rules (some EU countries, Canada under CASL) are configured with consent or opt-in requirements per segment.
Which sequencer do you recommend?
Instantly or Smartlead for volume and inbox rotation; Lemlist for smaller, more personal programmes; Outreach or Salesloft if your team already lives there. The workflow feeds any of them.
Can it do LinkedIn steps?
LinkedIn automation carries account risk and is offered only as a custom build with conservative limits, or via a sequencer that supports it natively.
Case study
AI Sales Agent for B2B SaaS Company: 3x Pipeline Coverage in 90 Days
How a 40-person B2B SaaS company used an AI sales agent to automate outbound prospecting, generating 3x pipeline coverage and reducing cost per qualified meeting by 60%.
Calculator
Put your own numbers in
Estimate the cost of the current process and what this workflow changes.
The pillar
AI Sales Agent
Research leads, personalize outreach, and book meetings—deploy in minutes, no code required.