Skip to content

Automated Cold Outreach

"The only advantage I want is ... a starving crowd!"
— Gary Halbert

Even though it's a well-known fact that marketing works best when you reach out to prospects at the peak of their pain, most marketers still take a spray-and-pray approach. When it comes to cold email outreach, they simply build a list of random people at a company and blast out copy‑pasted scripts without considering whether the timing is right or not.

Think about it: if you had internet access and an infinite number of people working for your marketing team with an unlimited budget, how would you do cold outreach? A good answer could be:

"Have them monitor all the press releases, data platforms, business news, social media, and every possible source to spot anyone signaling they might need my services."

In other words, reach out to the starving person when they say they need food. That's the goal. Let's see how we can use AI to solve this in any market.

4 Simple Steps

Here are 4 simple steps if we break down the original idea:

  1. Have an AI eye on the internet to see who publicly gives away cues about needing your services
  2. Extract more information about the prospect for further context/verification
  3. Get the prospect's contact details (email, social media, etc.)
  4. Reach out with a personalized script tailored to the prospect's needs

1. Your AI Watchtower

1.1 Data Sources

First, identify the best news sources for your business. Here are a few ideas to get you started:

  • Social media niche groups/communities (great for B2B and B2C): LinkedIn groups, Instagram hashtags, Facebook groups, Reddit, etc.
  • Press release websites (if you have a B2B business): GlobeNewswire, PR Newswire, Cision, etc.
  • News platforms (again, better for B2B businesses): Yahoo! Finance, Reuters, Bloomberg.

Think of any active online source for your niche. Even if there's a source with 99% irrelevant announcements, it can still be useful—we're looking for that 1%.

Pro Tip: Most websites offer advanced search functions or industry‑specific categorization features. Keep those in mind to narrow down your search.

After you get a sense of these data points, start creating web scrapers or, even better, use their APIs to access that data (always check each website's robots.txt file to make sure your IP doesn't get blocked).

1.2 Read the data

Now that you have multiple data sources, you should separate the signal from the noise. Think of this step as asking ChatGPT to reply with Yes or No to a question. Here's an example:

"Let's say you're a professional recruiter. A possible signal for you is an announcement of a funding round at a company, since hiring often follows fundraising. That's a signal for you that must be separated from the rest."

Now design a prompt for this and test it with ChatGPT. The better you define the separation between Yes and No, the more accurate your automation workflow will be. We'll be using these prompts in the APIs.

Pro Tip: LLMs may produce inaccurate outputs if prompts are ambiguous. Make sure you maximize the accuracy of the model with prompt engineering best practices.

At this point, you want to save the signals in a CSV or Excel file for later steps.

Note: There are many ways to classify a text as "Yes" or "No." Here, I suggest using ChatGPT and prompt engineering for simplicity. In real business settings, be sure to explore different methods to achieve the most accurate results .

2. Understand Your Leads Better

At this point, you have your leads but not enough context about their situation to create a personalized message. Let's address that.

2.1 Use AI tools

Researching topics has never been easier with tools like Perplexity or ChatGPT's Internet Search. All you have to do is think:

"What are some specific things about this lead I can mention in my message to capture their attention?" Here are some examples:

  • If it's about a funding round, look for: investor's name, investee's name, funding amount, funding stage, quotes from founders.
  • If it's about a new drug in development, look for: drug name, collaborating drug developers, regulatory status, clinical phase.
  • If it's about a product launch, look for: product name, unique features, target market, release date, partnerships, competitive differentiators.

To extract information, use chat-based AI search tools. Just like in the previous step, make sure you provide enough context for each search and use prompt engineering best practices to get accurate results.

Pro Tip: It's better to extract single words or phrases for each field instead of LLM-generated explanations. This will make step 4 much easier.

Note: Some of this information may already be in the initial online article you found and saved in the previous step. Make sure you use that too.

2.2 Manual search agents (optional)

Sometimes, it's better to take the search one step further and create your own AI search agent to get more accurate results. This requires additional development, but in most B2B cases, you'll get better data.

There are many ways to create these search agents, but one of the most efficient is orchestrating effective web searches using Boolean operands or external APIs from RSS feeds or data providers. Here's a simple example of extracting funding numbers from different sources:

Search Agent Orchestration

3. Find Contact Information

This is probably the easiest step. If you've used social media crawlers to get the lead, then all you have to do is slide into the user's DMs.

If you've used other data sources like press releases or news, you can easily use Hunter.io, Instantly.ai, or other APIs to automatically get a list of contacts at the company.

Note: For smaller companies (generally speaking, fewer than 100 employees), it usually makes sense to contact the CEO. For larger ones, it's probably better to aim for other C-suite members or VPs.

4. Personalize and Reach Out

At this step, you should have:

  • Contact name
  • Contact email
  • Details about what made that person a lead

All you have to do is create a personalized script with this information and send the email/DM.

For this step, you can get creative by exploring LLM-generated scripts, but I usually like to maintain more control over the script since it's the first touchpoint with the client. Here's my process:

  1. Choose one of the leads you found in the previous steps and create a highly personalized marketing email for them. It'll probably look something like this (probably not the best marketing script, just an example):

    Sample Personalized Script

  2. Send the email. You can use Direct SMTP, email services, or email marketing tools like Hunter.io, Instantly.ai, HubSpot, etc.

What Next?

Congrats! Now you have an automated email marketing system that reaches out to the right person, at the right time, with a personalized message. This is your new lead-gen and marketing outreach machine that costs only $15 per month on a server.

Pro Tip: To keep this running, you can either run the script manually every day, or take a more professional approach by using process schedulers like Celery to run the script at set intervals. Pair this with a simple dashboard, and you'll have the coolest AI marketing agent ever.

There's no limit to this. There's always another data source you can explore to create a marketing campaign. It's best to create as many campaigns as possible to test different possibilities and see which ones are most effective.

  • Let's have a virtual coffee together!

    If you need help implementing these types of automations, schedule a free 30-minute session with me to discuss your AI challenges and explore how we can work together.

    Book Free Intro Call