AI for Outbound Webinar March 19th

This session is about outbound: building a complete AI powered pipeline that sources leads, cleans the data, researches every person, and preps the outreach, all in under thirty minutes and for about sixty dollars a month in tools. It is built for founders and operators who know they should be doing outbound but keep avoiding it because the manual version eats entire days. The structure is the usual one: a quick win you can apply immediately, a live build that raises the complexity, and a Q&A at the end.

Why outbound, and why people avoid it (2:07)

Outbound is something every single company and every single person should be doing right now. It is also genuinely hard, partly because a good outbound pipeline takes real time to implement, and partly because a lot of people are simply scared of putting their services in front of the world.

Quick background on me: my name is Miguel, I studied astrophysics, I have been teaching people how to use AI since the release of ChatGPT, and now I run my own company doing AI education and implementation. I do a lot of sales automation because that is where people see results immediately. And here is the pattern I see constantly: people tell me "what I do is not working" when what they actually need is ten or twenty times more volume before they can even judge whether it works. So the whole session is about bringing things to scale with essentially two tools.

The outbound funnel in four steps (5:55)

Outbound is made of four things. Lead sourcing, which is just a complex phrase for "where are the people you want to talk with". Data cleaning, because lead lists always arrive as huge messy databases. Research, meaning going through each person's profile to verify they are actually worth contacting. And contact, the sending itself, which takes forever and, if you brute force email, can burn your domain. Trust me, ending up blacklisted by Google is really not fun.

The AI powered version of that chain (7:12) keeps exactly one thing constant: the contact is done by you. If we automate everything end to end but send the incorrect message to the incorrect person through the incorrect channel, nothing works. If you automate something that does not work, the result does not work either. So automate sourcing, cleaning, and research aggressively, and only automate the outreach itself once you know your message converts on email, LinkedIn, SMS, or WhatsApp (8:33). Automating the actual outreach is the last step you ever automate.

Demo 1: sourcing 100 CEOs with Apify (9:22)

Apify is my favorite platform for everything related to lead generation, because leads are everywhere: Instagram, Reddit, YouTube, Google Maps. If you sell to local businesses, Google Maps is literally your best friend. People love to talk about their problems in public forums, and if you find the people asking for help, you can sell your services to them.

For the demo I used the Leads Finder actor, where an actor is just Apify's word for a service. You give it a job title, location, company size, and keywords, and behind the scenes it queries huge databases of people whose information is being resold. I typed CEO, kept Mexico as the location, asked for one hundred of them, and clicked start (11:40). A few seconds later I had one hundred leads with company name, company website, full name, job title, and industry. My previous runs on the same actor pulled three and a half thousand people each.

Here is the catch: this information is always dirty (12:56). Some rows have no email, some have no last name, some are missing the phone number. You could fix that in Excel, but I never want to look at an Excel sheet again in my life. So I downloaded everything to a folder and moved to step two.

Demo 2: cleaning the data with Claude Cowork (13:51)

I cleaned the file with Claude Cowork, mostly because it is trendy and I used to criticize it a lot, and it turns out it is actually quite good. The key difference from Claude Chat (14:48): Cowork has access to the files on your computer, and it does its work by writing code in the background. That is the secret sauce they do not really advertise. Creating Excel files, cleaning reports, consolidating information, all of it happens through code.

I pointed it at the folder, asked what was inside, then asked it to clean the data and build a dashboard. It removed duplicates and blanks and exported a squeaky clean CSV of every lead that actually has an email, in literally seconds: eighty five clean leads out of the original hundred (18:00), plus an interactive dashboard breaking them down so the list becomes immediately actionable. Maybe I only want to contact the people in Jalisco or Nuevo Leon; now I can see exactly who they are.

One bonus that is not lead specific (16:12): this works on any dirty data on your machine. My favorite pile of dirty data is my downloads folder, and yes, you can have Claude Cowork clean that up too.

Demo 3: researching people with deep research (18:51)

Once the list is clean, the next job is research. If there is someone you really want to talk with, you can use Claude or Perplexity to run deep research on that person and surface things from their recent life or career that give you a hook. You can also look for intent signals, which just means evidence that someone is ready to buy, because they are interacting with your competition on social platforms.

I ran a deep due diligence search on myself as the example (19:22), and the amount of information this surfaces is startling. I have found the names of people's kids by accident running this kind of research, because you never know where your data lives online. A concrete client use case (21:07): we push for podcasts, so we research who appeared on a podcast recently and personalize the invite around what they said. The limitation is speed: deep research takes about three to five minutes per person, so at this level it does not scale. That is what the advanced demo fixes.

Demo 4: LinkedIn outreach with Claude Chrome (22:08)

To close the loop I imported the clean leads into a Google Sheet and told the Claude Chrome extension: this is a list of leads in Mexico, find the people living in Guadalajara, go to my LinkedIn profile, and send them connection requests. It started going through profiles one by one, and I stopped it before it actually sent anything, because I do not want strangers added to my LinkedIn right now, and because, again, I do not recommend automating outreach until your message is proven.

The cost breakdown (26:40): Apify at thirty nine dollars plus Claude at twenty, so about sixty dollars a month to grow your LinkedIn network. And since the same list contains emails, you can plug it into a cold email system like Instantly and go multi channel: emails and LinkedIn requests to the same people you actually want to reach.

The full pipeline: Claude Code plus Perplexity (27:19)

For the advanced version I switched to Claude Code, which is already connected to my Apify account, so lead sourcing and cleaning now happen within about two minutes of conversation. Then I asked it to research every lead and their company through Perplexity, looking for recent news I could use as a hook for a campaign, and to do it fast.

This is why I say prompting is dead (30:27). I am literally just having a discussion with my computer. Claude Code decided to write a fast parallel script, and when I asked it to explain parallel like I am five (31:19), it said: without parallel you mail one letter at a time; with parallel you hire twenty friends, each grabs a letter, and they all go to the post office at the same time.

The results (32:25): out of ninety five leads, it found news on thirteen people, completed in fourteen point six seconds with zero errors, from one sentence. The hooks it surfaced were real: a livestream with ninety eight thousand views, a piece on e-commerce trends in Mexico, a new security product announcement, a speaker slot on the longevity industry, someone who just became CEO and founder. Each finding serves one of two purposes: a personalization hook, or a filter to remove people you do not want to talk with. Then one more sentence created a Google Sheet with everything (34:14), and Claude added hook summaries and confidence levels for each piece of news without me even asking.

Where leads actually live, and the outbound math (35:01)

Apify sourced these leads from a database, but leads come from everywhere: LinkedIn posts, Instagram engagement, YouTube comments, Reddit threads. Everyone talks about their problems in public. Whatever the source, you clean it with the same code from step two, research it with Perplexity or a service like EXA, and then decide how you want to reach out.

That last step matters most, because in the age of AI everyone is looking for human connection (37:03). If you ever get a message from me, it is actually from me; the only outreach I automate is email, because I cannot send that volume manually, but the intent is still mine. And the math makes the effort worth it (38:01): one hundred connection requests a day, roughly twenty accepts, maybe five replies. That is five potential people you could be doing business with, from a system built in under an hour.

Q&A highlights (39:01)

Is there a simpler way to connect Claude to Google Workspace? Yes. In the chat window, hit the plus symbol and open connectors: Drive, Gmail, Google Calendar, Google Sheets, Canva. And here is the underrated part: anything you connect to your Claude application, Claude Code gets access to as well, no MCP servers to mess with. OpenAI just shipped a very similar update, so this pattern is becoming standard.

Does LinkedIn provide an API you connect Claude Code to? (41:50) It exists, but it is extremely expensive. The workaround is that I never connect my LinkedIn account to anything; I use Claude to interact with LinkedIn the way a person would, clicking buttons in the browser. And if you want LinkedIn data without touching your account at all (42:54), Apify can scrape the reactions, comments, and profiles on any public post, extremely cheaply.

What about LinkedIn restrictions? I have never heard of someone hitting restrictions using the Claude Chrome extension, which is honestly surprising, and it is why I allow myself to use it as much as I do.

Updates, and a PostHog bonus (44:18)

All the session recordings live in the webinar library on my website, including the ones I keep off YouTube because they show sensitive data like LinkedIn profiles and emails, and the webinars page has the calendar links to join live every week. I have also been building my A3T community for individuals who want to learn AI with me directly; the content is still being finished, so message me and I will send a half off discount.

One last thing worth stealing (46:56): I built my website with Claude Code and recently connected PostHog, the service that shows you how people actually use your site. I handed that data to Claude, it noticed people rage clicking on a page, and I found five or six real bugs that way, then asked it to fix them all. I did not one shot that website; it took several days of upskilling. But analytics read by AI is a feedback loop I will be running from now on.

That was the March 19th session. For the rest of the series, every recap and recording is in the webinars listing, with a new live session every week.