How We Built a Lead Automation System That Saved 25 Hours Per Week

## The Client and the Problem Our client runs a B2B software company in the UK. They were generating around 80 inbound leads per month through their website, LinkedIn ads, and a couple of industry directories. The problem was what happened to those leads after they arrived. The SDR team was manually: 1. Copying leads from the website form into HubSpot 2. Researching each company on LinkedIn and their website 3. Writing a personalised first-touch email 4. Scheduling a follow-up reminder if no reply after 3 days 5. Moving the deal stage in HubSpot after each touchpoint This was taking roughly 25 minutes per lead. At 80 leads per month, that's 33 hours — nearly an entire working week — on administrative tasks that don't require human judgment. ## The Audit Before building anything, we spent a day mapping the process in detail. Two things became clear: 1. Steps 1, 4, and 5 were purely mechanical and fully automatable. 2. Steps 2 and 3 required some intelligence but followed consistent patterns — meaning AI could do most of the work, with a human in the loop for review rather than full authorship. This distinction matters. The goal isn't to remove humans from the process — it's to remove humans from the *low-value* parts of the process so they can spend their time on the parts that actually require them. ## The Architecture We Built The full system runs in n8n with four main workflows: **Workflow 1: Lead Intake** Triggered by a new HubSpot contact (created from any source). Enriches the contact record with company data via Clearbit, scores the lead based on company size, industry, and role seniority, and assigns it to the right SDR based on territory. **Workflow 2: AI Research & First Draft** Uses the enriched data to call GPT-4 with a carefully crafted prompt that writes a personalised first-touch email referencing the company's specific context — their recent funding round, their tech stack (pulled from BuiltWith), or a specific pain point common to their industry. The draft is delivered to the SDR via Slack with a one-click "approve and send" action. Review takes 30 seconds rather than 15 minutes. **Workflow 3: Follow-Up Sequencing** Monitors the HubSpot deal stage and email open/reply status. If no reply after 3 business days, triggers a follow-up from a pre-approved sequence. If the email is opened but not replied to, triggers a different (softer) follow-up. All of this is invisible to the SDR unless they actively want to see it. **Workflow 4: Deal Stage Automation** Updates HubSpot deal stages based on email events (opened, replied, booked meeting) and Calendly booking webhooks. The SDR's HubSpot view is always up to date without them touching it. ## The Results After 60 days in production: - Manual admin per lead: down from 25 minutes to 4 minutes (review and edge cases only) - Time saved per month: 27 hours - Lead response time: down from average 4.2 hours to under 3 minutes - First-touch reply rate: up 18% (faster response + more personalised messaging) The SDR team's feedback was universally positive — not because they feared being replaced, but because they were finally spending their time on conversations rather than data entry. ## What Made It Work A few things were critical to the success: **Human in the loop by design** — We never fully automated the outreach. The SDR still approves each email. This maintained quality control and kept the team bought in. **Obsessive testing before launch** — We ran 200+ simulated leads through the system before going live to catch edge cases (unusual company structures, missing data, etc.). **Graceful degradation** — If any step fails (Clearbit returns no data, GPT times out), the workflow falls back to a simpler path rather than failing entirely. The lead always gets processed; it might just get less enrichment. **Tight HubSpot hygiene** — Automations are only as good as the data they run on. We spent a day cleaning the client's HubSpot before we built anything. --- If you're spending more than 5 hours per week on tasks that follow predictable patterns, there's a good chance we can automate them. [Let's talk →](/contact)