Full transcript
0:01What is going on everyone? This is
0:03probably one of the most crazy agents
0:05that I have spent the time building and
0:08we are creating personalized lead
0:11magnets at scale. So what that means is
0:14if I contact 3,000 people, 1,000 people,
0:17whatever, it's going to create me a,000
0:20lead magnets. To do this, I basically
0:22built deep research without spending the
0:25200 bucks that you would need to do with
0:27OpenAI. And we're going to use three
0:29tools to make this purr in a go to
0:32market motion. These three tools are
0:34triggery, clay, and n10. Now, we're
0:39going to break it down into three steps.
0:40Three steps per tool. First up is
0:43triggery. So, what we want to do is we
0:46want to find people that are talking
0:50about topical things that are relevant
0:52to our business. I.e., we want to find
0:54thought leaders. So to do this, I'm
0:57going to find anyone who's talking about
0:59cold email, maybe social signal, uh, and
1:03maybe like intent
1:05data and has been talking about this
1:08more than three times in the last 14
1:11days. The idea being is that we want to
1:14find those people that are talking about
1:15things that are relevant for us. So what
1:18I could do is I could go and start
1:19tracking Adam Robinson. Great
1:22individual, a lot of followers. I want
1:24to reach out to to him. So, if we then
1:27go and find him inside of the engagement
1:30uh suite that we have
1:33here. Um let's load him up. Here we go.
1:37So, you can see we're we're tracking
1:39him. We've pulled a lot of leads um for
1:42Adam. And the idea being is we want to
1:44take the engagement that Adam is getting
1:47and send this data through to
1:51um a clay table, right? And so we've got
1:5318,000 leads here from monitoring him.
1:55We want to send this data through to a
1:57clay table. So very simple. We're just
1:59going to head to here. We're going to
2:00set up a clay web hook and we're going
2:02to auto push data through from Adam into
2:06a clay table. And this is like an
2:09example here of some data coming
2:10through. The cool thing if you're using
2:12a tool like triggery is it gives so much
2:14data that we don't need to use any clay
2:16credits involved in here. So, we've got
2:18their name, full name, domain, location,
2:22the engaged post, so you get the full
2:24text of the post
2:26itself. Um, and then the LinkedIn URL,
2:30so on and so on. So, what we're going to
2:32use here is first of all, we're going to
2:34use a formula field. This is going to
2:36check to see that they're relevant ICP
2:37for us or not. And in this use case, I'm
2:40doing it based on location.
2:43If that is true and they're a good ICP,
2:45then we go through to this thing called
2:47like a post category. And so we're going
2:48to use AI to read the post that the
2:51individual has engaged with from Adam
2:54and check whether that person uh or
2:58whether that post is talking about
3:01something relevant to what my business
3:03could potentially offer, i.e. like how
3:06it could solve a potential pain. So in
3:09this use case or in this instance I've
3:10said hey read this LinkedIn post and I
3:12want to ensure that this post is
3:14relevant to the pain my business solves.
3:16If this post talks about intent data
3:18signal prospecting marketing strategies
3:20ABM strategies so on so on um let me
3:24know if that is true. So if it says it's
3:26true it then sends it through to this
3:28kind of lead magnet research agent so to
3:31speak. So here it says um an employee
3:34with a job title of X from this company
3:37X with this domain X like this post X.
3:41Your job is to research why this person
3:44has engaged with this uh piece of
3:46content based on who they are, their job
3:48title, so on and so on. And then give me
3:51a topic or a research paper to write
3:55based on why they have engaged with that
3:57piece of content themselves.
4:00So what will happen is the AI is going
4:02to basically go run a bunch of research
4:04and then start to give me like a
4:06response. And so this example, this
4:08topic is how can AI powered
4:10personalization in B2B outreach be
4:12leveraged to optimize website visitor
4:14conversion rates and sales
4:17efficiency. So what we do from here is
4:19we have two options. We can either send
4:21this through to a tool like
4:24um hey reach or smart lead or instantly
4:27or something like that or what we can do
4:31is and if we do send it through to these
4:34tools we can say we can reach out to the
4:36prospect and say hey prospect I can see
4:38that you engage with content around XY Z
4:41I've actually built a research paper
4:42which kind of dives into this subject a
4:44little bit more. Do you want me to share
4:46that with you? PS go follow Adam
4:49Robinson. He talks a lot about this. The
4:51reason I do that PS line is they always
4:53reply saying, "Hey, I actually really
4:54follow Adam Robinson." Knowing full well
4:57that they do follow him. It's just a
5:00kind of like a deliberate ploy to get
5:02like a response and create that kind of
5:04trust uh straight
5:06away. Then we can basically monitor if
5:09they say yes, they want the lead magnet.
5:10And then we can trigger this web hook
5:13here. So once this gets triggered, it
5:16fires it through to this um this agent
5:20essentially. And this lead magnet agent
5:23has five different agents inside of each
5:25other. First up, we have this query
5:27builder. This query builder takes the
5:30topic and then searches for five
5:32different subtopics. So breaks it
5:34basically down into five areas that the
5:39research leader then needs to go and
5:41start researching. And so the idea being
5:44is that we are pre-prompting an agent to
5:46then run research through using um
5:50perplexity um to gather the initial bit
5:53of research which will then send through
5:56to this project
5:58planner. This project planner will then
6:01basically go and create a kind of
6:04structure and um pre-prompts for the
6:08research assistants. So what I mean by
6:09that is it's going to create the title
6:11of the article, the subtitle, the
6:13introduction. Then it's going to create
6:15the chapters. So this this chapter title
6:18is the shift uh to signalbased
6:20prospecting. Here is the prompt. Next
6:22chapter is building your foundations in
6:24triggery. Here is the
6:28prompt. Then what it does is it sends
6:30that including the prompts through to
6:32this team of assistants where they then
6:34start to run research per chapter. But
6:36the key part here is and you can either
6:40take a screenshot of this now or I'm
6:43going to put the link and all of the
6:46prompts that I've used in the comments
6:48below guys. So you'll have that you'll
6:50have access to um this agent plus the
6:53clay table should you wish to have it.
6:56Um but we are basically saying okay hey
7:00we are writing a chapter for the title
7:03and um the subtitle the previous chapter
7:08was X the next chapter is X the current
7:10chapter is X and here is the prompt and
7:13what this basically means it gives the
7:15agents context so if they're writing
7:18like six paragraphs or six chapters they
7:21know what's coming up they know what
7:24they have previously written so So that
7:26there is no overlap in them writing this
7:30kind of topic area um or this this kind
7:33of um booklet or research paper or
7:35magnet whatever you want to call
7:37it where it then basically sends this
7:40data through to this editor here. Um
7:43this editor then starts to combine all
7:46of this data together and kind of
7:48formats it in a way that is like
7:49appropriate to use. I.e.
7:51Um, it can write in the language of that
7:54prospect. It can make it sound like it's
7:57coming from me. It can use citations,
7:59which is kind of key because we're using
8:02perplexity that will give citations.
8:04We're also using a trained LLM here to
8:07um have a look at like what we like the
8:09triggery knowledge hub. And the other
8:11thing that you'll notice is we use
8:12anthropic throughout this. We're using
8:143.7 reasoning. So, it's their kind of
8:17like their reasoning model. And the
8:19reason I use Claude, not GPT, is Claude
8:23just, you know, just outperforms kind of
8:25any other model that I've used when it
8:26comes to kind of creative writing and
8:28kind of like ideiation based um
8:31workflows. So I'd highly recommend you
8:33use them. Um I'm using their thinking
8:36process uh through open router and then
8:38just a usual kind of 3.7 through the
8:41node in NATM.
8:44But going back to this flow, once the
8:47editor has done what it's needed to do,
8:49it then creates the document, updates
8:52this document, share like makes this
8:54document sharable, gets the share link,
8:56and then sends this data back through to
8:59Clay where we then have the share link
9:03and all of the information on that
9:04person. So, we can then send this link
9:06to that individual with the copy uh with
9:09the copy, but also with the lead magnet
9:12attached. And that is how we are going
9:15end to end creating personalized lead
9:18magnets at scale. And they are
9:21incredibly good. And I'm going to leave
9:23um a document below showing an example
9:26of one that it wrote based on how you
9:29could leverage triggery. And that's
9:31really really cool because it tapped
9:32into our own knowledge base to kind of
9:34write that um lead magnet as
9:36well. So if you like this, subscribe,
9:40give this a thumbs up, follow along. I'm
9:42gonna be posting more and more agents
9:45every single week.