Cold Outreach That Stays Out of Spam

A custom email outreach platform with sequences, deliverability tooling and LLM personalisation

The client paid per seat for several outreach tools and still landed in spam. We built one platform that owns the whole pipeline, from domains to replies.

Marketing technology · Product build

Cold Outreach That Stays Out of Spam
Client
B2B lead generation team (anonymised)
Type
Outreach and campaign platform
Core
Sequences, deliverability, reply detection
AI
LLM personalisation with review
Stack
Next.js · Node.js · PostgreSQL · Redis queues

Email outreachDeliverabilityLLM personalisationNext.jsNode.jsJob queues

What the team needed

The team ran cold outreach on a patchwork: one tool for sequences, another for warm-up, spreadsheets for domain health and a separate AI writer. Per-seat pricing, deliverability hidden behind one score, and no access to their own data. We built one platform they own. Multi-step sequences, SPF/DKIM/DMARC checks before sending, inbox warm-up and rotation with daily caps, LLM opening lines an operator reviews, and reply detection that stops a sequence automatically. Job queues let sending, warm-up and polling scale independently.

Constraints we had to design around

  • 01

    Deliverability is the product

    Brilliant copy in a spam folder is worthless. Domain authentication, warm-up and sending limits had to be first-class features, not settings.

  • 02

    Many inboxes, many domains

    Volume had to spread across inboxes without any single mailbox exceeding safe limits or one bad domain dragging the rest down.

  • 03

    Personalisation that doesn't sound like AI

    LLM output had to be grounded in real prospect data, stay short and be reviewable before sending.

  • 04

    Stop the sequence when someone answers

    Sending a follow-up to someone who already replied burns the relationship. Reply detection had to be reliable across providers.

Architecture and what we shipped

  1. 01

    Sequence engine

    Multi-step campaigns with delays, conditions and per-contact state, executed by background workers on a job queue.

  2. 02

    Domain authentication checks

    SPF, DKIM and DMARC are verified per sending domain before a campaign can start, with clear instructions when a record is missing.

  3. 03

    Warm-up and inbox rotation

    New inboxes ramp up on a schedule. Live sends rotate across inboxes with per-inbox daily caps and automatic pausing on bounce spikes.

  4. 04

    LLM personalisation with review

    First lines generated from prospect fields and public company data, stored as drafts an operator can approve, edit or regenerate.

  5. 05

    Reply detection and classification

    Inboxes are polled, replies matched to contacts and classified (interested, not now, out of office, unsubscribe). Positive replies stop the sequence.

  6. 06

    Campaign analytics

    Sent, bounced, replied and positive-reply rates per campaign, per inbox and per domain, so a deliverability problem shows up where it starts.

How it was delivered

Sending infrastructure first, because nothing else matters if mail doesn't land. Then personalisation and replies, then analytics.

  1. Phase 1

    Sending core

    Inbox connections, sequence engine, domain checks, warm-up and rotation

  2. Phase 2

    Personalisation and replies

    LLM first lines with review, reply detection, classification and auto-stop

  3. Phase 3

    Analytics and operations

    Campaign, inbox and domain reporting, bounce handling and operator tooling

Questions about custom outreach software

Why build custom cold email software instead of using an existing tool?

When outreach is core to the business, per-seat pricing, black-box deliverability and no access to your own data become expensive. A custom platform lets you own the domains, sending logic, personalisation and analytics.

How does the platform protect deliverability?

It verifies SPF, DKIM and DMARC per domain, warms up new inboxes gradually, rotates sends with per-inbox limits and pauses an inbox when bounces spike.

How is AI used in the outreach platform?

An LLM writes personalised opening lines from prospect data. They are stored as drafts, so an operator can approve, edit or regenerate them before anything sends.

What happens when a prospect replies?

The reply is matched to the contact and classified. Interested replies stop the sequence for that contact immediately, and unsubscribes are honoured across every campaign.

How do we start a project like this?

Book a free scoping call. We agree the first version in writing, usually the sending core and one campaign type, and quote a fixed price for it.

Paying for five outreach tools and still landing in spam?

Own the pipeline instead. We'll scope the platform and quote a fixed price for the first version.

Tell us what you're building

Prefer to talk? Pick a 30-minute slot.

sales [at] yarify.tech WhatsApp Telegram