The AI Cold Email Workflow That Replaces a Junior SDR

So a founder I know fired her outsourced SDR last quarter and replaced him with a workflow she built in two weekends. The workflow runs every weekday morning. It produces 80 personalized emails per day, with reply rates higher than the human was generating, at a cost of about $400/month all-in including the AI tokens. The SDR was costing her $4,200/month. The math is brutal and also it doesn’t suck. Here’s the exact stack and step-by-step that she — and now a bunch of other solo and small-team founders — are running. I’m going to be honest about the limits at the end too, because this isn’t actually a magic solution, it’s a leverage one.

The stack

You need four things:

**Apollo or ZoomInfo**: contact database. ($59-$249/month)
**Clay**: enrichment, scoring, AI personalization workflow builder. ($149-$349/month)
**Instantly or Smartlead**: sending infrastructure with deliverability features. ($37-$97/month)
**Claude or GPT API**: the model that writes the personalization. (pay per token, ~$50-$200/month at scale)

Total stack cost: $300-$900/month depending on volume. Compare to one outsourced SDR at $3K-$5K/month or a junior in-house SDR at $5K-$8K/month all-in.

The workflow, step by step

1. Define your ICP and trigger events. This is the step everybody wants to skip. Don’t. Write down the specific company shape (size, stage, tech stack, growth signals) and the specific person shape (title, tenure, seniority). Define 2-3 trigger events that mean “now is a good time to reach out” — a new funding round, a new hire in the right role, a product launch, a public job posting that signals the pain you solve.

2. Build the lead list in Apollo. Filter by your company and person criteria. Export as a CSV or, better, push directly to Clay via the integration. Aim for 200-500 leads per week of email volume. Beyond that, deliverability degrades fast.

3. Enrich and score in Clay. Set up a Clay workflow that pulls additional data from LinkedIn, the company website, recent press, and any other sources relevant to your trigger events. Score each lead 1-10 based on fit. Drop everyone below a 6. Yes, that means you’re paying for enrichment on leads you discard — that’s how you keep quality up.

4. Generate personalization with AI. This is the step that does the actual work. For each remaining lead, send the enriched data to Claude or GPT with a prompt like: “You are writing a one-line opener that references something specific about [lead] and connects it to [your offering]. The opener must be true, specific, and feel like a human wrote it. Do not start with ‘I noticed.’ Do not use any superlatives. Avoid mentioning their funding unless it’s recent.” Output a single sentence per lead, max 20 words.

5. Build the email template around the opener. The opener changes per lead. The body (~80 words) and the ask (~20 words) stay the same. Total email length: under 120 words. Anyone going longer is wasting your prospect’s time and tanking their own response rate.

6. Push to Instantly or Smartlead. Upload the lead list with personalization. Configure your sending: warm-up your domains for at least 4 weeks before scaling; send from multiple addresses to spread volume; cap each address at 30-40 emails per day. Use the platform’s deliverability tools (warmup, spam score check, send-time optimization).

7. Sequence the follow-ups. Three touches max. First touch the personalized email. Second touch (3 days later): a one-line bump with a different angle. Third touch (5 days later): the breakup. Anyone going to 6 or 7 touches is hurting the brand more than helping.

8. Reply triage with AI. Use a separate AI pass (Lavender, or a custom GPT) to classify replies — interested, not interested, wrong person, unsubscribe, OOO. Auto-route the interested ones to your calendar booking flow. Everything else gets a polite response and stops the sequence.

9. Measure the right thing. Track reply rate (target: 4-8%), meeting rate (target: 1-3% of total sent), and pipeline generated. Don’t optimize for opens; the AI-personalization mailbox providers are getting smarter about flagging optimized-for-opens content.

What this saves you

For a team running 400 personalized emails per week, this workflow produces what a competent junior SDR would, at one-tenth the cost, and runs while you sleep. The savings are real. The catch is that you still need to do the strategy (ICP, triggers, value prop, ongoing iteration). Replacing the doing-the-emails part of an SDR job is easy. Replacing the thinking-about-who-to-email part is not.

Gotchas

Deliverability is the bottleneck. Warm up your domains. Use SPF, DKIM, DMARC properly. Send from a domain that’s not your main brand domain. If you skip these, the workflow will produce great content that lands in spam.

The AI personalization quality decays at scale. Above 200 emails per day from one workflow, you start seeing the AI hit the same patterns. Add prompt rotation or take a quality hit.

Replies require a human. Don’t try to AI-write your responses. Once someone replies, a human takes over. That’s what closes deals.

Compliance. GDPR, CCPA, CAN-SPAM. Include unsubscribe links. Respect them. Cold email is legal in the U.S. if done correctly; the rules in Europe are stricter.

FAQ

Does this really replace an SDR?

It replaces the email-writing part of an SDR role. The strategy, list refinement, reply handling, and meeting-running parts still need a human. For solo founders, that human is you.

How long does setup take?

A weekend to build the first working pipeline. Another two weeks of tuning before reply rates settle. Plan for a month before you’re getting full value.

What if I send fewer emails — say 50 a week?

The workflow still works. The cost is lower (~$200/month). The volume is fine for a niche, high-ticket ICP. Don’t scale beyond what your closing capacity supports.

Will this stop working when everyone does it?

Already partially. The “AI-personalized cold email” approach is widespread enough that the bar for what counts as personalized is rising. The teams that win are the ones doing deeper enrichment and tighter ICP work, not the ones generating prettier emails.