case study · automation · 2025

Cold Outreach Engine

How a messy multi-tool cold-outreach setup became one clean n8n flow — saving 40 hours a week and tripling reply rates.

Client
Tastewise
Industry
SaaS · B2B
Role
PM + Builder
Services
Automation
Year
2025
Duration
4 weeks
⚡ AUTOMATION · n8n + AI · Q2 2025
40h

Saved per week,
ongoing since
April 2025

brief · context

A growth team running on five separate outreach tools.

The client was a B2B SaaS in the food intelligence space, selling into mid-market and enterprise restaurant brands. Their growth team had built up a cold outreach motion over two years — and like most outreach motions that grow organically, it had become a five-tool franken-stack with three separate enrichment subscriptions, two email-sending platforms, and a spreadsheet glueing the whole thing together.

The team worked, but it cost a fortune to run, took two full-time hours a day to operate, and the per-lead enrichment bill alone was burning through six figures a year. They wanted a leaner version of what they had — same volume, same quality, less duct tape.

I was brought in for a 4-week scoped engagement to audit the existing setup, design a replacement, and ship it.

challenge · what was broken

The growth team was operating the system, not running it.

  • Five separate tool subscriptions, three of them duplicating enrichment work.
  • Two hours a day of manual data movement between platforms (CSV exports, paste, repeat).
  • No single source of truth for "did this lead get a reply?" — answers lived in three inboxes.
  • Per-lead enrichment cost averaging $2.40, paid even on duplicates and bounces.
  • AI personalization was a separate manual step, gated by one team member's bandwidth.
  • Sequence performance impossible to attribute — was it the copy, the timing, or the list?
approach · how it ran

Four weeks, four phases, no surprises.

01

Audit

Mapped every step of the existing flow. Logged true cost per lead, time per task, failure points.

Week 1
02

Sketch

Designed the replacement in n8n. Showed the team a working prototype on day 9 with three real leads.

Week 2
03

Build

Production flow built and tested with the team's actual lead lists. Side-by-side run with old system.

Week 3
04

Migrate

Cut over to the new flow. Documented every node. Recorded a Loom for handover. Old subs cancelled.

Week 4
solution · what was built

One n8n flow, four nodes, replacing five tools.

The new system is a single n8n workflow that handles enrichment, AI personalization, sequencing, and reply tracking end-to-end. Triggered by a Google Sheet drop, it routes leads through four key components — and writes everything back to one source of truth.

Component 01 / Enrichment

Smart enrichment (cost-aware)

Checks an internal cache first, then routes to the cheapest enrichment provider that has data for the domain. Skips known duplicates and bounces before spending a cent.

n8n Apollo API Hunter API Postgres cache
Component 02 / Personalization

AI-personalized first lines

For each enriched lead, generates a personalized opening line referencing their company's recent press, launches, or hiring signals. Uses GPT-4 with a tightly-scoped system prompt and a few-shot library.

OpenAI API GPT-4 Few-shot prompts
📤 Component 03 / Sequencing

Multi-step sequences

Sends the personalized opener via Smartlead, with two follow-ups timed by behavior (opens, clicks, replies). Pauses sequences automatically on out-of-office detection.

Smartlead n8n logic OOO detection
📊 Component 04 / Tracking

One source of truth

All replies, opens, clicks, and conversions write back to a single Postgres table powering a Metabase dashboard. The growth team finally has attribution they trust.

Postgres Metabase Webhooks
results · the receipts

Faster, cheaper, and actually trackable.

40h
Saved per week, ongoing
3.2×
Reply rate vs the prior tool stack
$0.18
Per-lead enrichment cost (was $2.40)
5→1
Tools replaced by one workflow

Beyond the headline numbers: the growth team got their afternoons back. The cost savings funded their next hire. Most importantly, they now have attribution data they trust — every reply traces back to the exact sequence, list, and personalization variant that produced it.

Ahmedabad → everywhere · Q2 · 2026
Hey

If you’ve got something half-built, half-broken, or just half-an-idea, send it my way. We’ll see if I can help. No deck required.

- Akash