AI for Sales Webinar January 22nd

This session was a full walkthrough of the AI sales stack I actually use, from a plain $20 ChatGPT or Perplexity subscription up to automated pipelines built on Apify, Claude Code, and Instantly. It is for anyone who sells, from solo founders on a tight budget to teams already running campaigns at volume. Everything below happened live, with real leads, real campaigns, and real dashboards.

Sales is not only sales (6:13)

I opened with the idea that frames the whole session: sales is an entire funnel, not just the conversation where you close. Before you ever talk to anyone there is lead sourcing, then data cleaning, then research. I do not know how many hours I have lost in Excel and Google Sheets deduplicating emails, and research is brutal at scale for a list of one hundred, two hundred, or five thousand prospects.

After that comes the contacting, the meeting, the business proposal, and the closing. Two things most people underuse: your meeting transcripts are immensely valuable because AI can analyze them and become a coach for you, and proposals can be templated or automated far more easily than people think. The webinar focused on the front of the funnel, because that is where the time actually goes.

Getting the most out of a $20 subscription (9:44)

Before touching any prompts, cybersecurity 101. In ChatGPT, go to Settings, then Data controls, and switch off "Improve the model for everyone" if you are on a paid plan. Perplexity has the same thing under personalization, called AI data retention, and Claude has it under the Privacy tab as "Help improve Claude". I personally do share my data with Claude because I am a big fan of Anthropic, but that is personal preference.

Then three use cases that work in any of the big tools:

Decision makers at one company (11:36). I took Antilogy, a client whose website only lists a generic contact address, and ran a deep research asking for the names of the decision makers. It found all of them, plus one email. That is the trick: company emails always follow the same structure, so if you find one email you found all of them, and you just ask the model to extrapolate the pattern. No email validator needed, one deep research run.

Many companies at once (15:49). Ask deep research for the CEOs of consulting firms, marketing agencies, or law offices in a given location, Paris, the US, Mexico, anywhere. You get the companies, the CEO names, and often their contacts, then you can run deep research on each company individually. If you cannot afford email enrichment services yet, this is a fantastic budget option.

Due diligence on one person (18:18). Before a discovery or closing meeting, run deep research on the person to find hooks, points in common that establish equal ground in the first minutes of a call. I ran it on myself and it found where I studied. Deep research exists in ChatGPT, Claude, Perplexity, and Gemini, and honestly they all work fine.

On top of that, ChatGPT's agent mode is great for local businesses (19:27). I used it to build a list of florists for my uncle, who wanted to work at a flower shop: "go to Google Maps and bring me back a list of 20 flower shops in Paris." You can run several in parallel, one per business type or per city. For $20 a month, unbeatable value.

Enter Apify (22:14)

A lot of people talk to AI as if it will solve everything by itself. It will not. What solves your problems is knowing which tools to use and how to connect them to AI. Apify is a platform for gathering data from the web: LinkedIn, Instagram, YouTube, Amazon, Facebook. People love to complain online, and a Facebook community full of "I wish another product did this" posts is a lead source.

The big lead databases are expensive: Apollo costs hundreds of dollars for a few hundred leads a month, and tools like FullEnrich are pricey too. On Apify, a Leads Finder actor takes plain natural language, "founders in El Salvador", you click Start, and you download the results. The catch is that the raw export is dirty: duplicates, missing emails, stale data. Cleaning that by hand takes a lot of time, which is where the next part comes in.

Building the funnel live with Claude Code (25:31)

Claude Code is an agent that became my best friend six months ago. Mine is connected to my Google account, my Fireflies, and my Apify account. Live on the call, I had it launch the Leads Finder remotely: C-level, small and mid-sized companies, El Salvador, professional services. It came back with 99 results including verified emails.

Then, one prompt at a time: download the data and clean it, removing duplicates and anyone without an email, which left 76 people. Next, use the Perplexity API to check whether each person appeared in recent media coverage. Since Perplexity is natively connected to the web, it surfaced interviews, CEO appointments, and funding news, and narrowed the list to 11 people with genuine intent signals. That compression is the point: at 10,000 raw leads the same ratio gives you 1,000 leads with extremely personalized intent signals all at once.

Terminal output is hard to read, so I asked Claude Code to build an HTML dashboard over the three databases. Then the key move: tell it to save the workflow, so every future lead pull repeats sourcing, cleaning, research, and display automatically. That is roughly 25 to 33 percent of the sales funnel automated in 20 to 25 minutes, going slowly on purpose.

Timing and intent from LinkedIn comments (29:48)

Sales is also a matter of timing. When you were fifteen and wanted to go out with your friends, you did not ask while your mom was angry. Same with leads: the right message to the right person at the right time.

On LinkedIn, people openly talk about their problems, and posts with lots of reactions, comments, and reposts show you exactly what they care about. Apify has scrapers for all three, so I gave Claude Code a post link, had it pull 19 comments, then analyze them with the Anthropic API to find who was genuinely curious. Three people stood out: one had DMed the author directly, one asked what the system would cost to implement inside their company, which is extreme intent, and one expanded on the concept with her own perspective. A database of 1,000 leads who interacted with a post yesterday costs about $10, so your cost per lead is extremely low. This exact approach got me two meetings, simply by opening with "I saw your comment on this post and thought it was interesting."

Watching who views your profile (45:39)

Then I demoed one of my favorite builds, made for a sales coach with 40,000 LinkedIn followers and around 2,000 people viewing his profile every week. No one can follow up on that volume by hand.

Instead of paying for the extremely expensive LinkedIn API, Claude Code controls the Chrome browser through MCP: it navigates to LinkedIn, switches the viewer filter from the past 90 days to the past 14 days, scrolls, and downloads the HTML. Then it runs cookieless Apify scrapers, so your own account never does the scraping and is not at risk, and qualifies every viewer against the ICP with another AI pass, ending in a dashboard. The live run extracted 66 viewers, 58 identified and 7 anonymous, since private-mode browsers cannot be captured, and enriched 51 companies. No one qualified this time, which is normal: it is tuned to my client's ICP, not mine. The application took two days to develop, which is the other sales KPI I care about: speed to value.

Cold email that does its own admin (51:14)

For this webinar I ran cold email campaigns through Instantly, and one of them brought 29 positive replies, 6 from England and 8 from El Salvador. Managing cold email is a full-time job, so I built a Claude Code workflow that runs after every campaign finishes, and it goes far beyond "this person said yes":

• The roughly 95 percent who never reply get moved into a list stamped "contact in three months". They never said no, so they get recontacted.
• Out-of-office replies are sneaky gold: people put phone numbers in their signatures. The workflow analyzes each email thread, extracts phone numbers and LinkedIn profiles, adds the person to my CRM, and sets the follow-up date to the exact day they are back. I recover 60 to 70 percent of those phone numbers without touching any enrichment service, and the data is clean because it comes straight from their emails.
• Positive replies go into the CRM with next steps written down.
• Aggressive nos go to a block list, and bounced addresses go to a bounced list.

I ran it live on the England campaign and got the full breakdown: leads contacted, unique replies, 21 sending accounts. Run a 10,000-person campaign over the holidays, when 20 to 30 percent of people are out of office, and you get thousands of fresh phone numbers to cold call for free.

Never automate the human part (59:49)

By now sourcing, cleaning, and research are automated. Contacting is where I draw a big ugly heart on the funnel. I am one hundred percent against automating LinkedIn messages: in the age of AI we are all drowning in generated spam, and people crave human contact more than ever. Email is the exception, since you cannot send that volume by hand anyway, as long as the context is as relevant as you can make it.

From meeting transcripts to proposals (1:00:19)

Fireflies records all my meetings. I had Claude Code pull the transcript of a discovery call with Mark, who wanted CEO coaching, and extract his exact pain points, down to the precise lines he said. Then it built a business proposal around those verbatims: his vision, the six use cases he wanted to explore, his success criteria, the program, deliverables, pricing, and next steps. One prompt, in French, and already clean enough to be presentable.

If you are on a budget, look at Typst, a free document language that generates very good-looking PDFs, comparable to LaTeX. Combine transcript extraction with Typst and your proposals write themselves in minutes. With the meeting and proposal stages covered, we automated about 80 percent of the entire funnel, and we did not even touch follow-up, where deals die because a human simply forgot.

Resources and next steps (1:10:17)

I shared my Notion page with all the session recordings, every tool mentioned, student programs to get these tools for free, Apify included, and a step-by-step Claude Code setup guide for Windows. I also opened bookings for 30-minute calls if you want to discuss working together. These sessions run every two weeks and are getting more niche over time: AI for marketing, AI for sales, AI for operations.

Q&A highlights (1:12:08)

How do you stay organized enough to resell these workflows? Three answers. My framework lives in the claude.md memory file, built up over time, and Claude Code updates it automatically when it hits a problem. Keep frameworks specialized: one for SEO, another for workflow automation, because you would not put a plumber's brain in a lawyer. And to share a workflow, have Claude Code turn it into an application with a visual UI and basic authentication. Everything I showed, except the browser scrolling, is 100 percent just code underneath.

What about vibe coding? For enterprise clients, a single developer vibe coding is a no-go because of data security; a team of five to ten building robustly together is different. For SMBs the stakes are lower and speed matters most, so they profit hugely from it. My longer-term worry is skills erosion: in a few years developers may no longer know how to code by hand, and we will be hunting for the dinosaurs who can still review system architecture.

Should a business development lead build alone or with the internal tech team? It depends on the complexity of the use case and how much speed matters. Bring your team a PRD, a product requirement document, and ask for something that works within a week rather than something bulletproof. Execution is extremely cheap now; distribution is the limiting factor.

Best setup on a $20 budget? Do not pay for Claude Code. Use Antigravity or Gemini CLI, because Google's plans are extremely generous right now. I pay $100 a month for Claude Code and consider it the best money I spend, but on a tight budget it does not make sense. I run Claude Code inside Antigravity myself, simply to keep a visual on the files being created.

I exported my LinkedIn contacts, now what? Use the LinkedIn Profile Scraper No Cookie actor, at $3 per 1,000 profiles, to enrich the database, then have Claude Code qualify everyone against your ICP. For emails afterwards, a service like AnyMailFinder works.

Browse all past and upcoming sessions on the webinars listing.