Cold Email Automation: The Make + Instantly Pipeline Behind 2,000 Emails a Day

The short answer: our cold email machine was one Make scenario between our lead source and Instantly. Every lead was checked for its email provider, verified with NeverBounce, given an AI-written opening line, then pushed to Instantly, which sent from warmed-up inboxes. It contacted up to 2,000 people a day with a reply rate around 5.5%. The AI was the cheap part. Deliverability was the real work.

I built this first for our own prospecting, then deployed it for a client selling digital products. Here is the actual scenario, module by module, and what each part was there to protect.

What did the full pipeline look like?

Three stages, each one a separate piece:

1. Find the people. A precise LinkedIn Sales Navigator search (job title, industry, company size, country), then an email finder (we used Anymailfinder) to get each person's work address.
2. Clean and prepare each lead. This is the Make scenario below. It decides whether a lead is safe to email, writes the personal line and hands it to the sending tool.
3. Send and follow up. Instantly sent the sequences from several warmed-up inboxes and caught the replies.

Most of the result came from stage 2. A great email sent to a bad address does nothing, and a few hundred bounces can burn a sending domain for months.

How does the Make scenario work, step by step?

1. A webhook receives the lead. Each new contact from stage 1 arrives in Make with its name, role, company and email. A filter lets it through only if the previous step succeeded.

2. We look up who hosts the email. An HTTP call checks whether the address runs on Google Workspace, Microsoft 365 or another provider. That matters for deliverability: emails sent from a Google inbox to a Google address, and from Outlook to Outlook, tend to land in the inbox more often. So each lead was later matched with inboxes on the same provider.

3. A router splits the leads into three paths.

• Path 1: verify, then personalize. The address goes through NeverBounce. Only addresses marked valid continue. OpenAI then writes the opening line, a JSON module cleans the output, and an HTTP call adds the lead to the right Instantly campaign.
• Path 2: personalize directly. Leads that were already verified skip the second check (no point paying twice) and go straight to the AI line and Instantly.
• Path 3: log for review. Anything that doesn't fit, like an unknown host, goes into a Google Sheet. Nothing uncertain gets emailed.

4. Instantly takes over. Instantly sends the sequence, rotates between inboxes so no single one sends too much, and keeps the inboxes warm in the background.

Why verify every email before sending?

Because bounces are the fastest way to lose a domain. The cold email industry treats a bounce rate under 2% as the line not to cross (Hunter). Meanwhile B2B lists go stale fast: Instantly measured an average decay of about 2.1% per month, so a list loses a big share of its valid addresses within a year (Instantly).

Scraped and enriched lists are the worst offenders. Without a verification step, a normal list would push you well above 2% bounces. With NeverBounce in the middle, only addresses confirmed as real ever reached Instantly, which is what kept our domains healthy at this volume.

How did the AI personalize 2,000 emails a day without the cost exploding?

We did not ask the AI to write whole emails. The email itself was a fixed template with variables (first name, company, role). The AI wrote one thing: the opening line, from the person's profile.

Two reasons:

• Cost. One short line per lead uses a tiny number of tokens. At 2,000 leads a day, that difference is everything.
• Control. A tested template keeps the message, the offer and the legal footer identical for everyone. Only the first sentence changes, and that is the sentence that makes people keep reading.

The JSON step after OpenAI forces a clean, predictable output, so a strange answer from the model never breaks a send.

What were the results?

• Up to 2,000 people contacted per day.
• A reply rate around 5.5%, when B2B cold email averages sit around 2–3.5% (lemlist, Apollo). Anyone who runs outbound knows that's a very good rate.
• More clients signed than classic manual prospecting brought us, first for our own agency, then for the client.

What would we change if we built it today?

• Handle replies automatically. We'd connect Instantly's reply detection back to Make: an "interested" reply gets a booking link and lands in the CRM, a "not now" gets a reminder in three months, an unsubscribe is removed everywhere.
• Use a cheaper model for scoring. Small classification models now cost a fraction of a cent per lead. That makes it affordable to score every profile before it's even verified, and to sort replies.
• Move the heavy version to n8n. When volume or the number of steps grows, n8n bills per run instead of per step. We rebuilt this pipeline that way for a higher-volume project (n8n vs Make).
• Respect each market's rules. Easy opt-out in every email, and the B2B rules of each country (CNIL in France, PECR in the UK, CAN-SPAM in the US). The full checklist is in our AI lead generation guide.

This is the system we build for clients: our automated lead generation service, with the email automation that handles the sequences and replies. If you want one for your business, tell me about your offer and I'll tell you honestly if outbound fits.

Keep reading

• AI lead generation: the full pipeline and the rules
• n8n vs Make: which one for a pipeline like this
• AI chatbots for business: the inbound side of the funnel