Crypto marketing: 300 Instagram accounts and a targeting graph

Hey guys, I hope you are well. Um, quite happy to be able to make this video actually. I think this will be really interesting for anyone who's been missing this marketing, growth hacking, automation content on this channel because I haven't actually been doing that much of it recently. I think that that you'll really enjoy this video. I'll just give a little bit of background, like context. This work is being done for a client. They've given me permission to record and document the process. However, they they need to remain anonymous and there are like there there basically a few things I can't actually share for for for sort of their their protection.

And then there's a few things that I'm going to actually like retract and I'm not quite ready to make make public yet. But I'm I'm going to share as much as I can. So this is for a crypto client who they're they're like a sort of crypto marketing agency, let's call them. They do quite a lot more than that, but that's the specific area that I'm I'm working on. So promoting early stage launches essentially of tokens. So they came to me and because I have all of the sort of mobile devices that we previously used in our last marketing agency and we also use on Tinder, they asked if we could basically set up as essentially a huge farm for for promoting um crypto projects.

So they were interested in all sorts of of things like Twitter, leaving YouTube comments and upvoting, tuning into watching live streams and leaving comments and and so on on various different platforms. Cuz there's a lot of different avenues that that we we can explore of this. However, we're starting off doing some Instagram accounts for them. So we're we're just doing a a little test. I've set up 300 accounts and we're we're we're doing outreach essentially on on all of these accounts. They're based on mobile devices.

Like I've got a couple of them here. And to start off with we did some pretty interesting like automated content generation. This was like a problem that we really had to solve because we're creating accounts at scale. Like when we're looking to generate about 5,000 accounts, we all we want them to be really high quality with unique content. these are the solutions I came up with to to be able to build high-quality unique content at scale. So every single one of the accounts has content unique to that account.

And this was generated through an API. We were using Bannerbear for this. And it worked surprisingly well. In fact, I'm going to get one of the phones now. So I will try to camera wants to zoom in. Yeah, this is automatically generated content. And as well, if we come on to the profile page, camera doesn't want to focus. There we go. So this one has 12 posts. We're getting all of them up to 24. But you can even see just from looking at this, it's all branded to that specific account.

And I don't want to I don't know boost my own work too much. But I I was really happy with the outcome of them. So the way that that this is actually done is through finding blogs who are promoting crypto-related content, any content related to these accounts. You could apply this to any industry though. We're actually now playing around with a couple of other clients with this as well. We are we're extracting from the RSS feed of particular blogs. And whenever they put up a new piece of content, it's automatically pulled, added into airtable.

And then every single blog has what's called like the metadata. So if anyone who isn't familiar with this, if you go and share a website on Facebook or Twitter, something like that, it will display like a nice image of the website assuming that the website owner has put a nice image there. It will automatically pull that in to make it like just display but very nicely essentially. So, we're just pulling that same information and that image is being used as the background for the content. And then all we're doing is taking the title of that particular post and chucking it on top of that image.

So, it's a uniquely the image is actually generated by us just using curated content from blogs and yeah, from a number of different blogs. So, the quality of the content actually comes down to how good is the content how good are the images on on the blog. But, I was really happy with how these came out and it actually means that we're able to generate automatically new posts every single day with the latest news going on in crypto. I I really think that is quite interesting.

And then another sort of thing that I've been playing around with is a few things actually. So, one of them is like Fear and Greed Index which a friend of mine introduced me to which is a what is the current view towards crypto. Are people fearful of crypto or are they greedy? And you can actually use this to build up quite a good investment strategy. I won't go too off topic here. When the fear when the score is low and people are in fear, it's a great time for you to buy crypto.

When people are getting greedy, it's a great time for you to So, what another way that we were making content is actually pulling the Fear and Greed Index which is just it's just like a measurement and just posting daily saying current Fear and Greed Index is like 28 or 71 or whatever it is. And then another way that we you can do automated content generation was to find There are some sort of crypto meme pages and like trader humor type pages. And rather than just copying and re-uploading their content, what I was actually looking at doing is extracting the the text, so getting the actual joke, and then doing something similar to what what we have here.

Like you can integrate with Google's image API, for example, like the just a a stu- a stupid example would be something like I bought eight monitors as a crypto trader, so now I can lose money eight times faster. It will just be like stupid sentences like this, like trade trader humor. And then all you would need to do is put a background put into the Google API image search API to find an image of a trader setup and then just run it through the API, use that image as the background, put the text on top, exactly like what we did here, and you have unique content.

And you can generate this at at scale. And then the other really good thing is that each image, I don't actually want to go into like the accounts, but each image is branded, so it has this unique account's username on there, and it just looks so professional, and it also means that no one can steal our content, and that applies to all of our all of of the accounts that that we're managing. Yeah, that was the the first thing that we've been doing. We've been making a ton of content.

Um now, the next thing to solve is targeting. I'll walk you through a couple of different things that I've been been working on, and these were like the early stages when I was playing around with it, like just messing around in AirTable, basically. AirTable is not the right tool to do this because it was just me me playing around, but these were like the sort of first iterations, and I will go on and show you the the final iteration that that we have landed on. I'm just going to walk you through the database.

This is a list of target profiles within the crypto space. This is just given quite a high-level overview. If you see it moving around at the moment, that's because I'm currently scraping some stuff. So, uh I'm I'm literally just going to walk you through it. These are target profiles. These are followers. This is a relational database. These are followers of this target profile. They exist over in this table over here. This is The goal of this base This is an old iteration, but the goal of this base is to tell us is this a good quality target or not.

This tells us the total number of profiles that were extracted. This tells us the number of their followers that have an avatar as a profile picture, which is an indication that the followers are a low quality. This tells us the average number of followers that This is getting a little bit confusing for to explain this. This is the average number of followers that exist on the followers of this target account. I hope that makes sense. This is the percentage of them that passed our basic filters.

So, the higher that number is, basically, the higher the more relevant this is to us as a target. And so, I can actually show you what our filters are here. Uh we're filtering out some common Indian names. Like, I actually looked at what are the top 1,000 Indian names. We're also checking make sure that the username doesn't contain an underscore, that it doesn't have numbers in it. Not that all accounts with numbers in are bad. Just I did a test and I found that about 80% of the profiles with numbers in weren't relevant.

A little bit more than 80% actually. So, it just meant it's not worth us spending 80% of the credits being spent on It's just a wasted resource. So, I'm just sort of limiting that like we've got no shortage of profiles. So, I want to filter them down as much as possible. And that field is just being used to generate this percentage. And then this is telling us the percentage of the profiles that are a great target. These are profiles that we're looking for. They match our target criteria.

So, just to explain to you why I worked in this particular in this way. For example, we have this account here who 0.3% of their followers are a good match for us. Even though this is a crypto profile and it appeared to be good starting off with. Even here, look, the average is probably about 1.5% of their targets are are actually relevant to us. Even less in some cases. So, this target is actually more than 30 times more efficient for us to use and extract followers from than it is for us to use the ones that that are down here.

So, this database just gives like quite a high-level overview of almost a quality score of how good are these accounts as an actual target for us. And of course, we just prioritize from the top down because it means we have to extract less profiles, less API requests, and it saves money for the client. can literally apply to any industry as well. It's just currently this is crypto accounts, but it's just based off the filter it filters in in the following database. I'm just going to walk you through how this actually works.

Basically, you will also find that this is it runs as like a network. So, these are all crypto target profiles, and you will find that a lot of them have like duplicate followers. It's not like this has, I don't know, 10,000 unique followers compared to this one. There's a huge amount of crossover. And I'll I'll show you what I mean. This grid view. Okay, so what I'm doing is actually checking for duplicates. So, we extract all the followers from our target profiles, and you can see here we've got an example right now for lyric_nft is following metasweetdreams and cryptosupermarkets, which means there's crossover.

So, this is one account that we have extracted that is following two of our target profiles, which actually makes him a better quality target than someone who is only following one of our target profiles. So, this gives us a sort of quality check. We have an automation running that will just merge these together. It will delete one of the records, and it will add cryptosupermarkets up here. We have a basic filter. I think I actually just showed you this, but it filters out numbers, Indian names, and then some Ukrainian and and Arabic Russian letters, sorry, and and Arabic letters that which is not looking for for at the moment, then we can come into Scrape Lead.

So, I will even show you like follow on follow, I'm sure most people in fact I'm pretty sure everyone should know what follow on follow is. These accounts are all accounts not owned by us, but they're doing follow on follow, which means their followers have come from the same method of marketing that we're using, which means all of the followers here are highly likely to follow back other crypto accounts because that is how these accounts have grown. So, what I can actually do is write in does bio contain crypto or does it contain NFT, and this is going to return me a list of accounts that are doing follow on follow that are in the NFT space or in the crypto space.

And I can then give these list of accounts over to our team back there who are also doing follow on follow, and that it will result in them having a great follow back rate. In some cases, we've been seeing a follow back rate of up to about 80%, which is it's it's insane. Like I have managed French supermodels before using similar marketing and we weren't getting 80% follow back rates. It is just it is really and which I'm talking about models posting in bikinis. Who would have thought that crypto was able that we were able to generate a marketing strategy like that was able to outperform a supermodel posting pictures in bikinis promoting OnlyFans and and we're able to to outperform that.

Anyway, so then I'm just going to take you on to the next iteration and you can sort of see how how things evolved and escalated. So, these are great accounts. These are great target accounts for us at the current point in time. Because our accounts are new, generally they have less than 500 followers. And what you will find is it's harder to Right now, we're only interested in accounts that have less followers than we do as the follow back rate is much higher. So, I have this filter set to posts greater than 24, followers greater than 100, less than 500, following greater than 200, and then just again some basic filtering and on the bio and so on, as we have now actually scraped the bio when they reach this point.

Once they pass our basic filter, we will go and check the actual profile and extract all the info And yeah, and so we could expand this up to like has 2,000 followers, just we're not quite ready to yet. If an account with we own with 400 followers is following accounts with 2,000 followers, it's going to have a very low followback rate compared to we grow these our own profiles up to up to 5,000 followers using smaller targets, and then we increase our targeting to include target profiles that have up to 2,000 followers.

We will see a much much higher followback that that way. So, we have the data already, we've paid for it, it's just not useful to us just yet. But anyway, you can see like the information that we scrape. So, name, full name, link, what what are they linking to on their on on their social media page, number of posts, number of followers, number of following, text in the bio, are they verified, and who are they following. So, this is what I was mentioning as well. Here is some example accounts where this guy is following five of our target profiles, which means he is super involved in the like crypto space, which makes him a great target.

So, it's just giving this roll up all it is doing is counting how many of our target are they following. So, this is just giving like a sort of quality score, if that makes sense. Like, we will still target people with less, but it's just a sort of prioritization mechanism, essentially. Okay, and then let me take you on to the next development, the next evolvement of this. So, we have a list of great targets. So, then what I wanted to do is build a system to find out how come we find more target profiles that are in the same community as these ones.

So, what I did is I pasted them in here and we went and extracted who are these target profiles, the great targets that we have, these are the ones with between 200 to 500 followers, who are they following? Who is in their network? Who are they following in common? What we have here is about 45 thousand target profiles that we extracted and of accounts that This is is I hope I'm explaining this well and this is making sense. We have our good targets that we have established that they are high-quality profiles that we want to start following and we are looking for how can we find more of the more profiles exactly like this?

Who do they have in common? So, I've added them into this airtable and we are extracting the extracting the following of the good targets to check are there other target accounts that we're not aware of that all of these profiles are following, that they have in common? And For example, out of target profiles that that I did, like high-quality profiles I added in here, 15 of them, so 5%, just under 5%, sorry, about 4%, are all following this account, Coin Crypto Geeks, which makes this and sorry, in fact, let me just expand this a little bit more just to explain to you.

So, of course, there were some other ones as well, like Cristiano Ronaldo, 23 of the profiles that we have are following Cristiano Ronaldo, of course, and then we have The Rock, 19 are following him. So, what I did is I extracted the followers of the accounts of the network that they all have in common because we're not going to start targeting The Rock. He has like 310 million followers. So, just that's not showing that The Rock is a high-quality account. It's just showing that he's just dominating.

And what I did is I extracted the followers of the people within their network. And here you can see what I mean. So, I've extracted every single account that those target profiles are following. So, some of them are like even have two in common. Maybe it's just a friend. Maybe it's I don't know. And then the lower down the list we go, these are just going to be like random accounts like friends and family that that they're following. So, it's not interesting for us. What we're interested in is the higher this number is and the smaller the audience is of the target account, the higher quality that account that target profile is.

I hope that makes sense because it means they have a much more targeted, a much more potent audience. It means it's very relevant for us. So, I can put in followers of the profile within that network is less than 100,000, for example. And here Bitcoin uh sorry, CryptoJack you can actually ignore because that was a target profile in our previous database. So, of course there is going to be a high number of accounts that following. But here we have CryptoRelevant, which is one that I wasn't using before.

And we have 16 of our target profiles, so more than 5% of them are following this page, which instantly makes CryptoRelevant a great target for us. We and we should be pulling that in and extracting all the followers of CryptoRelevant and doing analysis in that previous database. But let me even just bring this down even more. So, rather than 100,000 followers, let's bring it down to 30, and to 30,000 followers. And so, these guys are actually doing follow on follow, and they're using similar targets to us.

So, we could actually go and check who are they following, and they've done a lot of the work for us already. But basically, the the smart thing to do here is take this list. The higher the number of the higher this number is, it means the more accounts they have in common, and the lower the number is, the higher quality these are for us. So, it's a case of copying these accounts now, going and extracting their followers, because that is like going to be a really high-quality network for us to use as targets.

This is quite difficult to to explain, but what we're talking about is essentially mapping out a network. So, we started off with putting in a list of crypto accounts, big crypto accounts. We went for and extracted their followers to find who are the highest-quality ones. Then we took the highest-quality ones, small accounts, but they're good targets for us. We put them into this database, and mapped out the network of accounts that they are following, which is what we've built out here. And now we have actually discovered these are the high-quality targets that all that a lot of our high-quality accounts that we're interested in following have in common.

So, we're generating a network and understanding the relation between all of these profiles. It's difficult to explain, but I really Yeah, Now, I'm going to get to walk you into the really interesting stuff. Like I said, in Airtable, before it was just me like playing around. Airtable's limited. Look, I I absolutely love Airtable, as everyone knows already, but it's limited to 50,000 records, to which is the the only only problem with it. I was looking at a way to solve this, because we're basically talking about handling millions, literally ten tens of millions of of records.

We're talking huge amounts of data. So, I played around with some Airtable. There was one called BaseRow, which is like an open-source Airtable alternative. Not bad, maybe like worth looking at, but like the API just wasn't wasn't robust enough anywhere near robust enough for the type of of of stuff that that we're doing. So, I'm going to show you this is Retool, and we've opted to go for this for displaying our data. So, we were originally planning on using like but was unable to connect into Retool.

So, we're using a um Postgres um database. And we're dumping like this is actually just like a little test environment that we set up yesterday. Where we put 21,000 records in there. So, today what I'm working on is pulling all of the data that I've already extracted and scraped out of Airtable and um dumping it into this database. And then just changing our automations around. So, rather than pushing data into Airtable, it now dumps into our like Postgres um database. But, what I've been working on building out, and this not myself actually, we've got like a data database engineer sort of work working on this.

So, what this will actually allow me to do is put in check followers range, and I can say just return me a list of target accounts that have between In fact, I'm just going to change this filter. Let's go by you are able to just like literally drag and drop this range slider. How many followers does the account have? How many following do they have? Do they have more than 12 posts? Post count that Anyway, it's only me using it, so it really doesn't matter. So, what I'm talking about building out is this is like just an insane targeting system.

So, it just like this would actually apply outside of just crypto as well. So, you can put in does like username contains whatever or you can just leave it blank. But, then what I could also do and where I think this gets so interesting. Let's say I've got a client who runs a I don't know, a golf course. And and they want more they want me to find more people to come and use use their golf. What I can do is I can just do some basic filters.

Post count is greater than 24. 24 is quite a good number. It means it's generally a real account. Followers, let's say 400 to 10,000. Following, that doesn't really matter actually. And username contains also doesn't matter. But, then in what I can do is come into and this has analyzed all of the accounts that they are following. And if I type in golf, it's going to check are they following accounts with the name with the word golf in their username or bio, which is let's say they're following like golf shop, golf Portugal, golf whatever.

It's going to return a hit count on this this profile. It will return the list here is following 10 accounts with the name golf. And this would apply to anything. You could also do it for crypto. You can do it for NFT for for whatever, SAS for like literally anything that that you can think of. And then you could also do some other filters like bio contains. We're also doing like the image analysis I mentioned already. So, we can do I only want to find females or I want to find males and unknown gender.

I don't want verified profiles. I don't want private profiles. I want the engagement rate to be between like 3 to 20% would be an example. And I only want profiles who have posted in Portugal, and this is analyzing only the last 12 posts. And they have also posted something between February the 1st to today, whatever today is. And then you could also come in with an age estimate. So we want people who are between the ages of 45 to 70. For example. There's also actually like an ethnicity.

This is just from like profile picture analysis. So it's literally telling you is guessing the ethnicity of someone. Now, I haven't actually built this out yet, but it's it will be something that I sort of plan on using. You will often find like a lot of clients that we used to to work with are quite Yeah, just quite specific. They would often complain about followers who are coming from specific backgrounds and like that they want, I don't know, European or like followers who are based in in the US, for example.

Yeah, and also you could actually use this another interesting use case I thought is you could very easily use this for dating. So, you could add in gender, female of or even male, depending on your preferences. And then going into a range around 500 to 5,000 followers, bio contains like you could even do something like bio contains single. Um but I personally wouldn't actually do that. And then you could type in the location close to you. So if I'm based in London, age estimate is going to be what, 20 to from 20 to like whatever, 29, whatever your preference is.

And this is going to return you a list of women who have posted in London in the past 12 posts uh and you could also actually do something like following contained dating. What would you do? Dating could not dating that's not a good example. You could try and like filter down by finding women or even women who have similar interests to you like you could actually type in like your favorite restaurants or you could type in like places that like some of your favorite Instagram pages and add them in here and it will return you a list of followers who are following similar pages to you.

Could be like an So there's just so many different use cases for for this data which is just yeah is is so interesting. Like we actually have all of this data together already uh and like I said I've got a data engineer. We had it functioning but it just started to crash when we started processing huge amounts of information and so he's currently working on that side of things just to to keep keep it stable. So this is just like an internal tool that I'm literally just sharing it with you that is made that I've made for like these for this crypto targeting.

And I think for the social media agency course like I I would actually look at giving it there as well cuz it's like really useful for that. Yeah what else to add? The automated content generation I I think is so interesting. Like I've I'm actually really enjoying this project. Like I'm on it like full time at the moment. it's in early stages. So it means it's a case of figure figuring everything out setting everything up. And it's been a little while since I have been able to work on a project where like the I I just love working with a client where they just say this is what we want to do.

This is our this is what we want and I'm just left to figure everything out to build everything out to come up with new ways of doing things to it's and when you have a budget to to do and the other good thing is that I realized is it's always good to work with clients who are because it means you don't have to explain why you're doing a lot of stuff which maybe doesn't it wouldn't make sense to a client who who doesn't understand marketing. Um I often find that you'll be like explaining a lot of the basics on or if they're not technical as well, whereas these guys are like growth hackers themselves.

Yeah, it's just nice to work with people who have an understanding and and an appreciation of what you're doing. This is my main focus at the moment is literally building out this database. We are talking huge amounts of sort of data analysis and data scraping which is is what what I love to do. Yeah. Uh it's quite a long video. So, if you have watched it through this far, like literally feel free to leave comments. I'm happy to like answer everything. And also sorry, just it feels good to be able to actually release some more videos like this as well.

Even I haven't been growth hacking or when I've been working on some client projects, it contains confidential information or they're like not happy for me to release something that I've just documented something that I've built for them or whatever the case may be, but these guys were they said it's absolutely fine as as as long as as they remain anonymous. Yeah, I I hope that was clear. It's quite complicated to to explain especially in one video, but yeah, I I hope that was interesting and insightful.