AI cold email works by feeding an agent real data before it writes anything. The workflow is four steps: scrape a prospect database, enrich each record from the prospect's own website, have AI write one personalized email per prospect, then send through cold-email software. Alec Saluga's last seven days on one client: about 1,700 prospects and 112 opportunities.
Key takeaways
This is one agent, not a fleet. Alec Saluga's point is that you do not need something complex to get real results.
Seven days on one client: a sequence to about 1,700 prospects, roughly 3,500 emails, and 112 opportunities.
For context Alec gives on tape: a 3% reply rate is typically considered good, and half of those replies might be telling you to stop emailing.
The personalization is what works. The agent reads the prospect's own site and references something real, like how long the firm has been around or a case it just settled.
Apify supplies the data (an off-the-shelf scraper store), and Instantly does the sending so you are not burning your personal inbox.
How do you use AI for cold email?
Most AI cold email is worse than no cold email, because people point a language model at a name and a company and ask it to be charming. The result reads exactly like what it is.
The version that works inverts the order: get real data first, then write. Alec Saluga demoed this live on Session 006 and it is deliberately simple. In his words, this is not 25 agents, this is more like one agent, and it goes to show you do not need something incredibly complex to get real results.
What results should you actually expect?
These are the numbers from a screenshot Alec pulled right before the session, covering the last seven days with one client.
The context matters more than the headline. Alec's own baseline: a 3% reply rate is typically considered pretty good, and half of those replies might be telling you to stop emailing them. That is what a normal cold-email operation looks like, and it is why the opportunity rate here is worth examining rather than just admiring.
One honest caveat: the client is not named, and the 3,500 figure is Alec's estimate on tape ("it's probably 3,500 emails-ish"), not a reported number.
Metric
Last 7 days, one client
Prospects in the sequence
about 1,700
Emails sent
roughly 3,500
Opportunities
112
Typical good reply rate, for context
about 3%
Where does the prospect data come from?
Data is the real bottleneck, not the writing. Apify is the answer Alec points people to: a store of off-the-shelf scrapers you can run without building anything.
You can scrape Facebook pages for emails, TikTok accounts for analytics, Google search results, or, in Alec's case, a lawyer and attorney lead scraper, because one of his businesses sells to attorneys.
The agent handoff is the part worth copying: you give the agent the scraper ID and an API key, and it runs the scraper itself and pulls the data in. The agent is not just writing, it is fetching.
How does the AI personalize each email?
The scraped dataset is the starting point, not the personalization. The enrichment step is where the agent goes to the prospect's own website and pulls out something real: how long they have been around, a case they just settled.
Then the email references it. "I see you've been around since 1997." "I saw you just settled this case for this amount, we can get you five more of those."
It is the same instinct as the blue-bubble text message from earlier in the session: the extra layer of effort is what signals a human did the work, and that is what moves reply rates.
“it can go on the website and pull out some stats from the law firms, like, hey, I see you've been around since 1997”
Alec Saluga · 52:39
How do you send at scale without burning your inbox?
The finished emails push into Instantly, which is the cold-email software behind the screenshot. Its job is deliverability: sending at volume without torching your personal domain and inbox reputation.
That is the whole loop. Pull in data, write personalized emails tailored to both the prospect and your offer, push into a sequence, get meetings on autopilot. It just runs.
What is not built yet, and Alec is explicit about this: the agent that replies to the responses and sets the meetings. That is the next layer, along with pushing it all into a CRM. Worth knowing before you assume the loop closes itself.
The repeatable order of operations
If you take one thing, take the sequence. It works because each phase feeds the next.
Scrape the prospect database
Start with data, not copy. Use an off-the-shelf scraper rather than building one: Apify has a store of them, from Google Maps and search results to industry-specific lead scrapers. Alec found a lawyer and attorney lead scraper because one of his businesses sells to attorneys.
Give the agent the scraper ID and an API key so it can run the scraper and pull the dataset in itself. This is what makes it an agent rather than a writing tool: it fetches its own inputs.
The dataset alone is not personalization. Have the agent visit each prospect's website and pull one real, specific detail: how long they have been in business, a recent result, something only someone who looked would know. This step is the entire difference between this and spam.
Write one email per prospect and push it into the sequence
The agent writes a hyper-personalized email tailored to both the prospect and your offer, then pushes it into cold-email software that sends at scale without burning your personal inbox. Then it runs.
On one client over seven days: a sequence to about 1,700 prospects, roughly 3,500 emails, and 112 opportunities. For context, Alec Saluga's own baseline on the same call is that a 3% reply rate is typically considered good and half of those replies may be asking you to stop emailing.
What tools do you need for AI cold email?
Three pieces: a data source (Apify's off-the-shelf scrapers), an agent to run the scraper and write the emails (Alec demoed this with Codex), and cold-email software to send at scale without burning your personal inbox (Instantly). That is the whole stack.
How do you personalize cold emails with AI at scale?
Do not ask the AI to be charming with just a name and a company. Have the agent visit the prospect's own website and extract one specific, real detail, then reference it. Alec's example: noticing a firm has been around since 1997, or that they just settled a particular case.
Is this a complex build?
No, and that is the point Alec makes explicitly. This is one agent doing one thing, not the 20-to-50-agent systems shown elsewhere in the same session. It pulls in data, writes personalized emails, and pushes them into a sequence.
Does the agent handle replies too?
Not in this build. Alec names the reply-handling and meeting-setting agent as the next thing to implement, along with pushing everything into a CRM. As shown, a human still handles the responses.
Tools used in this post
Every tool here has its own page with pricing, who used it live, and honest alternatives.
49:19 Alecwe started the sequence for about 1,700 people. It's likely they've gotten two emails, it's probably 3,500 emails-ish a week. And from that we've gotten 112 opportunities.
49:19 AlecYou're typically getting maybe 3% reply rate is pretty good.
50:22 AlecHalf of those 3% might be telling you to stop emailing them.
50:22 Alecwhat we do is we scrape that database and we pull in all the info about the prospect and we write these hyper, hyper-personalized emails
51:31 Alecwhat Apify does is they have some off-the-shelf scrapers here. They have a whole store filled with scrapers.
52:39 AlecYou basically give them the scraper ID and an API key and it can go and use the scraper to pull in data.
52:39 Alecit can go on the website and pull out some stats from the law firms, like, hey, I see you've been around since 1997
53:31 AlecIt basically allows you to send emails at scale without burning your personal email.
55:29 Alecthen you implement the agent that's replying to the responses and setting the meetings
Who wrote this
Every AI4NTP post is written by an operator who was in the room when the work happened.
A former B2B salesman with no technical background who self-taught AI. He builds and deploys AI-driven marketing and websites, and has grown a following of over 15,000 teaching AI adoption.
Founder and host of AI4NTP. He sold his first company from his college dorm room, and as a fractional CMO has helped scale multiple businesses past $50M in ARR.
