Sales & CRM

How to build an outbound email sequence with AI

Sales & CRM3 AI tools7 steps6 friction points

Want an agent to handle this on your live data? See the Starch version →

An outbound email sequence is a series of timed, personalized messages sent to cold or warm prospects to start a sales conversation. Writing one that actually gets replies means nailing subject lines, sequencing follow-ups at the right intervals, personalizing to each recipient's role or pain point, and knowing when to stop. Most operators have to build this from scratch every time they open a new market or launch a product line — which is why it sits permanently on the to-do list.

The workflow looks like a writing problem on the surface, so AI feels like an obvious fit. You're generating copy, varying tone by persona, adjusting follow-up angles — all things a language model is genuinely good at. The output is text, the iteration cycle is short, and the cost of a bad draft is low. That makes outbound sequences one of the first sales tasks operators try to hand off to ChatGPT or Claude.

General-purpose AI tools — ChatGPT, Claude, Gemini — can produce solid first drafts of individual emails, rewrite subject lines, and suggest follow-up angles for different buyer personas. They can take a product description and a target ICP and return a five-email sequence in under two minutes. That part works. The gaps show up when you try to personalize at scale, connect the output to your actual prospect list, and keep the sequence running without babysitting every step.

Skip the manual work → See the Starch version

Sales & CRM3 AI tools7 steps6 friction points

AI walkthrough

How to do it with AI today

A practical walkthrough using ChatGPT, Claude, and other off-the-shelf LLMs — what they're good at, what you'll have to do by hand.

Tools that work for this

ChatGPTClaudeGemini

Step-by-step

1 Define your ICP and value prop in a short brief — two to four sentences covering who you're targeting, what problem you solve, and what proof points you have. Paste this into ChatGPT or Claude as context before writing anything.

2 Prompt the model to generate a five-email sequence: cold intro, follow-up one (different angle), follow-up two (short and direct), breakup email, and a re-engagement email for non-responders after 30 days. Ask for subject line variants for each.

3 Review the output and rewrite the first email by hand. The model's cold open is usually generic — replace it with something specific to the prospect's role or a real trigger event. Use the model's draft as scaffolding, not final copy.

4 Paste the sequence back into Claude and prompt it to rewrite each email for two alternate personas — for example, a VP of Sales versus a founder — adjusting tone, problem framing, and social proof references accordingly.

5 Copy the finalized emails into your sending tool — Apollo, Instantly, Smartlead, or whatever sequencer you use — and manually configure send timing, steps, and exit conditions. This step has no AI assist; it's manual data entry.

6 For each batch of prospects, export a CSV from your CRM or Apollo, open it, and manually inject prospect-specific details into the templates before importing to your sequencer. At low volumes this is fine; it breaks down past 50 contacts.

7 Track reply rates by checking your sequencer dashboard and your inbox. When you want to iterate on copy, start the prompt chain over from scratch — there's no persistent memory of what you ran last month or what performed best.

Prompts you can copy

Write a 5-email cold outbound sequence targeting VP of Sales at B2B SaaS companies with 50-200 employees. Our product reduces sales rep ramp time by 40%. Tone: direct, peer-to-peer, no corporate jargon. Include subject line options for each email.

Rewrite email 2 of this sequence for a founder audience instead of a VP of Sales. Shorten it to under 80 words, lead with a different pain point, and make the CTA a specific question rather than a meeting request. [paste email]

Generate 5 subject line variants for this cold email. Optimize for open rate. No clickbait, no excessive punctuation. Include one that references the prospect's industry directly. [paste email]

Given this ICP — Series A B2B SaaS, 20-100 employees, GTM team of 3-8 people — what are the top 3 pain points that would make them open a cold email about sales enablement? Give me one paragraph per pain point I can use to personalize sequence step 2.

Write a breakup email for a prospect who hasn't replied to 4 previous emails. Keep it under 50 words, be direct, leave the door open, and don't guilt-trip them. Product context: [paste brief].

Reality check

Where this gets hard

The walkthrough above works — until your numbers change, the LLM hallucinates, or you have to re-paste everything next month.

No connection to your actual prospect list — every time you write a personalized variant, you're manually copying names, titles, and company details from a spreadsheet into the prompt.

Outputs vary between sessions; the tone and structure you carefully dialed in last month won't automatically carry over when you start a new chat to write the next campaign.

The model has no memory of which emails performed — you can't ask 'what subject lines got replies last quarter' because nothing is tracked or stored between runs.

Personalizing at scale breaks down fast. Generating one customized variant per prospect works for 10 contacts; at 100 it becomes a part-time job of copy-pasting and reviewing.

No live data from your CRM or sequencer — if a prospect replied or unsubscribed since you last checked, the model doesn't know, and you risk emailing people who've already converted or opted out.

The sequence lives in your head and your sending tool, not in a system — so onboarding a new team member or handing off the workflow means rebuilding context from scratch every time.

Tired of the friction?

Starch runs the whole workflow on live data — no copy-paste, no hallucinated numbers, no re-prompting next month.

See the Starch version →

Starch alternative

The same workflow on Starch

Starch is an agentic operating system. For outbound sequences, that means an agent builds a persistent app connected to your live prospect data, CRM, and inbox — so the sequence runs, tracks, and iterates without you re-running prompts manually.

Install the CRM starter app and describe your outbound process in plain English — Starch builds a pipeline tracking contact status, sequence step, last reply date, and ICP fit score, all in one place instead of a CSV.

Starch syncs your Gmail directly, so email thread history lives inside the CRM automatically. Ask 'who hasn't replied to step 2 in more than five days?' and get a real filtered list, not a manual audit of your inbox.

The LinkedIn Automation app finds prospects matching an ICP you describe, sends outbound connection requests at human-paced intervals through browser automation, and surfaces warm contacts ready for email follow-up — all without touching LinkedIn's API.

Describe the sequence you want in plain English — 'five emails, alternating angles, 3-day intervals, stop when they reply' — and Starch builds the automation and wires it to your contact list. No drag-and-drop builder, no manual sequencer configuration.

Connect Apollo from Starch's integration catalog; the agent queries it live when building outreach batches, so your prospect data is current when the sequence fires instead of being last week's export.

Iteration is a conversation, not a rebuild — tell Starch 'change the follow-up timing to 5 days and rewrite step 3 for a founder persona' and the app updates without you touching a prompt chain from scratch.

Get closed-beta access →

Toolkit