How to Build a Cold Twitter Outreach System in 27 Minutes

TL;DR

Build a cold Twitter outreach system by sourcing followers of niche accounts, filtering them in Clay, and using AI to identify qualified prospects before sending tailored DMs. The demonstrated campaign narrowed about 49,000 profiles to 3,883 likely B2B SaaS founders or executives, with GPT-4o Mini classification projected at $4 per 100,000 rows. Read on for the complete sourcing, qualification, distribution, and messaging workflow.

Transcript

Hey, so I was in the middle of building our own internal Twitter DM campaigns and I thought why not make a video showing the end to end process of us kind of starting up a new campaign. The idea for it, finding the leads, filtering the leads with Clay all the way going to um you know like actually sending the DMs and what kind of DMs to send, what ... Read More

Key Insights

• Lead sourcing starts with niche Twitter accounts whose followers are likely to include the intended buyers. The campaign asks Grok for five active accounts with fewer than 50,000 followers that discuss B2B SaaS and attract founders and executives rather than primarily serving indie hackers.
• Scrapely collects both followers and followed accounts from each selected Twitter profile. Repeating this process across five source profiles takes about 10 minutes in the demonstrated workflow and creates a large initial pool that can be exported for qualification.
• Follower count is an initial quality filter rather than proof of relevance. The campaign first required more than 250 followers, reducing roughly 49,000 imported profiles to about 24,000, while the broader recommendation is to require at least 100 followers to avoid messaging likely bots.
• AI classification is more precise than relying only on profile keywords. Words such as "building," "founder," "CEO," or "CTO" can describe agency owners, ecommerce operators, indie hackers, or SaaS executives, so the campaign evaluates names and complete profile descriptions against explicit qualification criteria.
• The Clay prompt classifies whether each person is a B2B SaaS founder or an executive at a real company. It returns a simple yes-or-no result, runs only when a profile description exists, and excludes people whose profiles appear to represent indie SaaS projects.
• GPT-4o Mini is used because the presenter considers it effective and economical for classification. Using an external API key, the stated projected cost is $4 for 100,000 rows, and profiles without descriptions are skipped instead of consuming classification requests.
• Lead distribution should reflect the number of sending accounts. When five accounts are available, a qualified list of 5,000 leads can be split into separate CSV files so each isolated account receives its own portion rather than leaving four accounts unused.
• Campaign copy should use the vocabulary of the target audience. B2B SaaS founders are addressed through demos, signups, free trials, SQLs, MQLs, MRR, and ARR rather than agency-oriented language about appointments or meetings, with tested angles including Loom offers, pay-per-demo proposals, and direct pitches.

Questions & Answers

Q: How do you build a cold Twitter DM outreach system?

Start by finding active niche accounts whose followers include your target buyers, then scrape their followers and followed accounts with Scrapely. Import the profiles into Clay, filter by follower count, and use AI to classify qualified prospects before exporting them for audience-specific DM campaigns.

Q: How do you find B2B SaaS prospects on Twitter?

Ask Grok for five active accounts with fewer than 50,000 followers that discuss B2B SaaS and attract founders and executives rather than indie hackers. Manually review those accounts, then use Scrapely to collect both their followers and the accounts they follow.

Q: How long does it take to collect leads from the source accounts?

The demonstrated process repeats the follower and following scrape across five Twitter profiles. According to the presenter, setting up those scrapes takes about 10 minutes.

Q: How do you qualify Twitter leads using Clay and AI?

Import the scraped profiles into Clay and create an AI classification column that evaluates each lead's name and description. The prompt returns yes or no based on whether the person appears to be a B2B SaaS founder or an executive at a real company, while excluding apparent indie projects and skipping profiles without descriptions.

Q: Why are keyword filters insufficient for Twitter prospecting?

Terms such as founder, CEO, CTO, or building do not establish that someone operates a B2B SaaS company. Those words may also describe agency owners, ecommerce operators, or indie hackers, so the workflow uses AI to interpret the full profile description against explicit qualification criteria.

Q: How many qualified leads did the Twitter outreach workflow produce?

The workflow began with about 49,000 scraped profiles. Requiring more than 250 followers reduced the pool to about 24,000, and AI classification then narrowed it to 3,883 likely B2B SaaS founders or executives.

Q: How much does GPT-4o Mini classification cost in this workflow?

The presenter projects a cost of $4 for classifying 100,000 rows with GPT-4o Mini when using an external API key. The classifier runs only when a profile description exists, so rows without descriptions do not consume classification requests.

Q: What cold DM angles should be tested with B2B SaaS founders?

The campaign tests three angles: offering a relevant Loom video, proposing qualified demos or SQLs, and sending a short direct pitch. The copy uses SaaS terms such as demos, signups, free trials, SQLs, MQLs, MRR, and ARR, and can include the campaign's result of 14 booked demos in 30 days as social proof.

Summary & Key Takeaways

• The workflow begins by asking Grok for five active Twitter accounts with fewer than 50,000 followers whose audiences include B2B SaaS founders and executives. Their followers and followed accounts are collected through Scrapely, producing a broad prospect pool that can later be filtered for relevance rather than trusted as immediately qualified.

• After importing about 49,000 profiles into Clay, the campaign filters for follower count and uses GPT-4o Mini to classify profiles from names and descriptions. The classifier looks for founders or executives at real B2B software companies while excluding indie projects. This process produces 3,883 likely qualified prospects for export.

• Qualified prospects are exported as CSV files and can be divided among multiple isolated sending accounts. Campaign copy is then tested through three angles: offering a relevant Loom video, proposing qualified demos or SQLs, and using a short direct pitch. Each message adopts SaaS language and includes demonstrated outreach results as social proof.