Skip to content
Selected work

Autonomous sales outbound

A sales pipeline with nobody standing in it.

Client
Maxime · French industrial group
Sector
Industry · LED displays
Duration
4 days · May 2026

The problem

The sales team spent dozens of hours a week on LinkedIn: finding prospects by hand, qualifying them, copy-pasting into the CRM, writing every message one at a time. Volume was capped at what one person can process in a day, and personalisation quality dropped as volume rose — exactly the wrong way round.

The architecture

  1. 01

    Apify

    targeted prospect extraction

  2. 02

    GPT-4o

    qualification + icebreaker

  3. 03

    Lemlist

    campaign triggering

  4. 04

    n8n

    end-to-end orchestration

  5. 05

    Supabase

    central state, 30s sync

Every stage is independent and replayable. If one link fails, state stays consistent in the database and processing resumes from the last confirmed checkpoint, with no duplicate sends on the campaign side. The control center shows live pipeline state rather than a summary computed after the fact: the sales team sees what is happening, not what happened.

The product

Anonymised screenshots — client names masked, amounts and volumes fictional

Three automations, one place to look

The control center brings together continuous LinkedIn monitoring, a campaign trigger, and opportunity detection across the sector's trade shows. All three write to the same database, so there is exactly one place to look — the condition for a tool like this to actually get used.

The generated icebreaker isn't a merge field

The model receives the prospect's profile and the product context, then writes an opener grounded in something verifiable from that profile. The difference from a mail merge shows up in reply rate, and more importantly in the fact that a prospect who replies doesn't feel processed like a spreadsheet row.

API consumption in plain sight

The dashboard shows live credit balances for the third-party services the pipeline depends on. An autonomous pipeline that stops for lack of credits is a pipeline that fails silently: the gauge exists so that can't happen without warning.

Stack

  • n8n
  • Apify
  • OpenAI GPT-4o
  • Lemlist
  • Supabase
  • Next.js

The outcome

  • 4 days

    from brief to production

  • 100%

    of the cycle without human intervention

  • 30s

    dashboard synchronisation latency

  • 3

    automations in production: monitoring, campaigns, trade shows

Need something similar?

I read every message myself and reply within 24 hours. No automated quote, no form disappearing into a void.

Start a project

Other work