This Clay Workflow Gets 10x Better Cold Email Results (Full Setup)

Most cold outbound fails because people are spraying and praying. Today, Nefal is going to show his clay workflow that scrapes job postings, analyzes company data, and writes personalized emails to hiring managers only for a few pennies per lead. This is how the top 1% of cold emailers are doing outbound right now. So, let's get into it. All right, man. So, I'm interested to to learn more about this outbound workflow you have. So, over to you. It' be good if you could just share your screen and and walk us through it. um and interested to hear all the the nitty-gritty details about it. >> Sure. No worries. Well, um just get my screen and up. So, um so yeah, I'll just uh I'll just kind of show like the end of in mind. So, just a high level what this flow is. It's basically helping recruitment agencies uh target uh and find more hiring managers to become customers, right? Um you know, recruitment is generally typically like a very hard niche to to do cold email in. So anywhere you can get an edge um be different um is, you know, is going to help plus the volume, right? So we're, you know, we're using kind of a few variables here. um which you'll see actually in Bison it's just one uh one variable but within clay we're able to yeah use multiple multiple variables in in one so for example like here this is one variable uh which is which is the vacancy itself this is a part of the JD right which we we analyze to see what's the most important um kind of criteria and then we've made it a bit more relevant just to essentially um kind of call out what niche or kind of company uh or industry they're in, right? Um and then finally, we have like a rotating uh subject line, sorry, a PS line that would uh customize depending on the job and and the kind of the the the company itself. So, um so yeah, that's that's that. Um, so yeah, I'll first of all just show I'll start from the beginning and then and then you can kind of see where where it ends up.

So you go start with LinkedIn, right? So this is where you'll just pull out any um kind of job uh search, right? Um and then from here you just take the link, right? And then what we're using for this is Ampify. So, Ampify, for those that don't know, it's a it's a marketplace for scrapers and it allows you to scrape um LinkedIn jobs essentially. Um and then when you're you you you deploy the the task and essentially what that would allow you to do is push all of that data into Clay. Now, you can see here that there's all these uh records here, right, which is basically the the job search. And then from LinkedIn, we get all of this information such as your company employee size, job description, company name, all this kind of stuff, right? The basic info. Um, we also you also get the JD in here as well, right? Um, so if you if I just open up one of the the JSONs, so you get um, yeah, the job description text here, right? So that's where we're going to use as context for uh, you know, one of the variables. Now >> just a question Nefar. It's all right.

This I see this ICP check check box. This to me looks manual. What what's the the the logic behind that? It seems like a check and balance, but just interested to hear a bit more about that. >> Yeah. So this is um this is a bit of a basic one, but essentially this uh this uh client here, they only want to target companies that are below let's say 2,500 employee size, right? So if it if based on the employee size right they're below that then it would get a check. You can see here that this is uh above that so it wouldn't get a check. That's the only thing. Um another but another thing that we are checking as well this is a bit more yeah a bit more sophisticated is actually checking if the company that posted the job is an actual like end customer potential end customer or is it another recruitment firm right? because obviously we don't want to be contacting other recruitment agencies uh you know and and and that's going to be a waste of time right so so yeah we have those checks in place and then um at one point we was using Apollo right to find the contacts um obviously you know with Apollo's recent um you know challenges right um his scrapers and that type of thing we're using now just the the clay function uh people search function in clay Um so essentially you'll do a people search right against these domains and then from there you'll get the contacts right so this is where you'll get and just to show for for people that um aren't familiar with place people search you just kind of like Apollo you would put in the company um yeah kind of the uh the job titles so this is the job titles of the hiring managers right um in into into clay and then you'd reference the domains from the previous table and then and that's basically it, right? And also your location.

Now this is where it gets a bit more interesting. So your first of all what the first step is to just actually get valid emails, right? Cuz obviously we don't want to be um enriching contacts that don't have valid emails. It's just going to be a waste of API credits. So we're just using a simple waterfall um of ICPS, prospio, and better context. Um and then here we're looking up uh the previous table, right? Um so then essentially um so in a sense, yeah, you know, you're able to actually let me go SDR one because that was the one previously. >> You've got you've got so many tables in here. I see seems very overwhelming. Yeah, because obviously every table is a different uh it's a good question actually. Every table is a different hiring manager and department because you're scraping the uh you're finding contacts based on the or first of all you're finding jobs within a specific department right like sales right SDRs >> but then you want to find people who could be the the bosses of those SDRs which is obviously the yeah the CRO's VP of sales so you want to make sure that when you're finding people they're aligned so then it's not you're not contacting you know different department >> um so that's why yeah we have multiple tables Um, one other thing is, yeah, so like as I, as I was saying, the lookup the lookup jobs, that's where we're pulling in all the job descriptions. You can see here that some companies have multiple vacancies, right, which is really cool.

Then you can use in the step two, you can use like reference that like another job, right, in in their company. Um, so yeah, we have all that all of that as context. And then here, this is where we started doing the enrichment. So we're um this is one of the variables, right? So this is looking at you know what actual what type of company that they are. Um this is actually normalizing the job title cuz sometimes you know you get like SDR1 UK Europe you know um so it's it's not doesn't sound normalized right natural. So we're naturalizing that >> normalizing company name. We don't actually use that to be honest but yeah we just do it anyway. Um and then yeah and then this is when we're taking so that this one v1 job this requirement is actually looking at the first vacancy that we see and then pulling out one uh you know one requirement that we can reference right so here deep expertise in salesforce and CRM reporting right in that SDR right they're doing just pulling out something from the jade that that we think is going to be important again >> so sorry to keep interrupting could Can you show us the prompt that you're using for this um >> this one here?

>> Yeah, sure. Sure. >> And are these are these custom prompts you build internally or do you have like um uh like a prompt generator that creates these? Like what do you have any sort of method to the madness? Um yeah, so we I I basically had a conversation with Claude um which is a back and forth um conversation um and um essentially I yeah started with one one prompt right um that I wanted to do um and then I would share the output that I got from you know from Clay put it back in into claw and say look this is what I got I want more like this. Can you edit the prompts that way? Um, so you can see here that we're just looking at the we're just sharing the context, right, of um, what we want them to do. And and the key thing is, right, you can see here that we're not generating the whole email in one prompt. I'm just trying to get the best variable for the email, right? Um, so that's the key thing with like prompting and using CLA. like just do give them one job at a time and then you'll get the best output and the cheapest as well.

>> Um >> interesting. There's quite a big prompt. You can tell that there there's been a few iterations of like mistakes and then you go back and say no, don't do this, do that. Um I think that this is really the key because most people don't go to this length and and and they're unhappy with the output, but it's because they haven't their prompts aren't good. basically aren't aren't descriptive enough. >> Yeah, exactly. And it's all about examples as well, right? If you're giving them like enough examples, right, then they get the idea, right? Um, so yeah, you can see here there's a lot of examples of inputs, outputs. Um, and you can see here like literally like the cost is >> like zero, three zeros and five and five and a five and a two, right? Um, so >> do you have any opinions on um what models are good for what use cases? So for context, I'm we're not running any cold outbound at the moment. So you start to get rusty after a while, but I do remember when being in clay, you could use 40 for example, and then you could use anthropic and you would get there would be variance in in the output. Do you have any like opinions on what models are good for certain use cases or is it just you know use the cheapest model for for everything?

Yeah, I would um I would say if you're looking to do like what I'm doing here, which is because there's two options, right? You can have like web searching availability, right? Which is where you actually need to let's say if you're if I'm trying to analyze a company just to see within if they're within ICP, then I would obviously use web search. And at that point, I would probably use like a 41 4.1 mini, right? um in the past 40 and and to be honest there's yeah the 40 mini can work as well but I find just the results of 4.1 mini just to be a bit better um and also it's like yeah it's cheaper as well it can be cheaper in some sense um but um yeah so for like I like for company research IP research I would definitely yeah use like a 4.1 mini with like um what I'm doing here which is obviously I'm using text or data that I already have in the clay table and obviously manipulating that to get an output. You can use the cheapest model for that. So like for 40 mini, I'm using 40 mini here, 41 mini. Um yeah, rarely I would go to like 40 unless I have like I don't know like a 100 rows and I'm trying to like get the best output for these 100 companies, right? Um for research, right? And that's using like web scraping abilities as well. So, um, yeah, if you're doing cold out by that scale, you don't really want to venture above 4.1 million to be honest because then it doesn't make sense. You just make a really expensive table.

>> Yeah. Yeah, it makes sense. >> So, so yeah. So, um, so yeah, that that's basically one of the variables. Um step two uh we're like I said if they do have another vacancy we are actually writing the like the whole step two um you know um in uh in one prompt actually. Um so actually let me see what what we yeah we're still using for many here because again it's just we're just manipulating the data that we have. So there's no research bit research bit um research needed or web scraping needed. Um but yeah, so we're referencing like the the pre like the vacancy number one and then yeah and then obviously they have a vacancy they have another job so we're references referencing that as well right um but then if if let's say they don't have a job right then I'm just using a formula and just doing like a really basic follow-up um just it's basically just a nudge and then I'm a aggregating that um into one column. So then again in in email bias and it's is it's really you know it's clear. Uh and then and then yeah and then from here we're just generating a PS line, right? And this is from a just like a a a lob of text that I've just inputed in here of like different case studies that that this company has um has generated over time and yeah and it basically would choose the best one depending on uh what company we're speaking to.

demo generating a three-word subject line or or you know all small caps. Um um and then this one is an interesting one. So sometimes um we would like in the past we've contacted someone who let's say is hiring for the sales director and then we would contact the sales director, right? So it looks like oh you're getting replaced type of thing. So this is just a check just to just double check that the person that we're hiring is a level above the vacancy type of thing, right? >> Okay. Yeah. >> Uh so just avoid any awkwardness. Um and then Yeah. And then that's and then you get kind of like the full uh using clay to like actually just create the the sub the the um the custom variable as well. So yeah. So you can just use this as the custom variable as you saw earlier. So that we just so it has like Yeah. So it's really good for deliverability cuz you know every email is different. And then um and then this one here, last bit, is just basically just to refine um the grammar actually cuz sometimes when you're doing like this Frankenstein kind of way of like mixing up variables and all that kind of stuff, right, into like one email, grammar can be a bit off. So it's it's just doing a grammar check at the end and just naturalizing some of the some of the the copy as well.

>> Nice. Nice. There's a lot that goes into it. It's like almost like a a Lego set. Um yeah, so a couple of questions. Um because I can imagine the these are tend to be the questions that are on people's mind. So going back to the lead enrichment, how come you've selected those vendors in the waterfall enrichment? Um because there there's no shortage, right, of of data providers? have they just you found them to have the highest success rate or >> um yes it's a good question. So um it's whatever there's part part of it is whatever APIs uh the client has. So you know we just try to be as efficient as possible. So if they have a subscription with >> a specific tool like ICPS then then yeah we would um then would you would just use that right if they're paying for it already. Um the better context is an interesting one cuz they they are a waterfall enrichment. So they would use multiple providers to get you the result, right? So >> okay, >> essentially that's an interesting one.

So you're kind of like having yeah like maybe 10 providers in in one kind of API call type of thing. Um but I mean me generally like the the bigger the the the one the daily drivers that we use a lot lead magic >> is one um we use uh find email I always think find email is a solid like number two as well um and yeah and I think there's some up and cominging ones as well like that are similar to better contact which is better and rich which again they have >> uh um you know partnerships with like multiple provide data providers and they can provide you like a very cheap um uh yeah a very cheap credit uh system right where yeah it's it's probably one of the cheapest on the market to be honest um got it and then so you you mentioned you're using email bison I've I've never used a product obviously the generally I know this is a generaliz generalization but generally people that are doing these sort of workflows it's like smart lead or instantly If you're more enterprise and you're going to use some of the the bigger tools that have emerged over the last couple of years, what what's the the justification for using email bison over those smartly and instantly?

>> Yeah. So, yeah, I still you know personally also use like like on the story we still use instantly. I predominantly use instantly actually. Um and then smartly we have for like you know a couple clients emo bison. Um, yeah, essentially this this client, well, this client's actually my friend, right? So, kind of helping him out at the same time. But yeah, he he I recommend I recommended him email bias and because um uh the the amount of volume that he's going to be doing um for his campaigns is quite large. So, if you're if let's say you're sending over um 100,000 emails a month, but then you're doing it across like multiple clients, for example. You're just paying one one fee. That's uh yeah, that's just, you know, it's like $500 a month or whatever it is, right? But that's just one fee, but you don't get charged per lead, right? Whereas instantly or smartly, you get charged by how many leads you have stored in your workspace, right? So there's that like benefit where it's yeah scalable. Um it's cheaper as you scale. Um but also like they have a private uh server so you're um you have a dedicated server an IP for your sequencer right so it's you're not sharing an IP with like others which which kind of smartly than instantly do um and yeah and so it's yeah deliverability I found is has been better on inst on on sorry on email buying as well >> yeah which if if you're sending tens hundreds of thousands of emails that that that makes a huge difference. Um and and a campaign like this, would you view this as like an evergreen campaign?

You just let this >> run because there's uh you know uh job adverts aren't ever going to stop, right? So, do you just leave this running in the background or is it a campaign that you'd have to supplement with other campaigns to ensure that you're contacting enough leads per month for for this client? >> Yes, it's a good question. So, um I would say you it can be evergreen the way we've set it up. Um there's yeah it's it's it can be or sorry it's not evergreen at the minute. Um you just have to have like an API called interbison um to to have the leads uploading evergreen. Um I would say you could if you were to have like n or or make a flow into this then you can make it evergreen. Um, so yeah, but the the only the only caveat with this setup right now I would say is uh Clay's people search uh functionality isn't as superior as like an Apollo or sales now obviously is where a lot of the data is.

So um because of that you're to if you're going to get more if you want to get like better contact data then part of the step when you're find finding contacts has to be manual. Um so that's why it can be a bit convoluted. Um unless there's an API for a scraper which I haven't found I haven't found yet but um but yeah. >> Okay that makes sense. Um okay so separate to this workflow let's say I'm somebody who is like if we had a a scale of like uh go to market engineer let's say right you're going to be over here on the extreme of you know probably spending hours building these workflows creating like a real outbound engine whereas there might be people at the other side of the scale where they're using tools like smart lead instantly but they're not maybe building advanced things like this maybe it's just you know scraping leads from sales nab and just uploading them to the sequencer and and and minimal relevancy or personalization.

Are there any for for the for that group of people? Are there any um workflows that that you think they should tackle first? Like any workflows that you've just seen be absolute bangers that have worked consistently across like varying clients? >> Yeah. Um yeah, I would say right one flow that I can share right now is um a really well I would say it's it's simple enough um where you if if you're if you're able to pull a list let's say from an account list so you just need a a list of accounts from let's say store leads or even Google maps or whatever um you could create a campaign uh using just clay right uh finding in the context. So do you want me to share that one as well like whilst? >> Yeah, go for it. Yeah, if you think Yeah, for sure. >> Um so so for example um this is using um yeah this is using store leads right as a um as that you know as that source of truth of of accounts right. So yeah whatever source that you have you could you could use that. Now, um you could probably skip this part cuz this is very specific to to what my client is is um is request is is requesting, but yeah, like I said, they they want they want to only target customers that are within like a 20 mile radius of their location, right? So, um yeah, we're doing that kind of we're finding the actual company address. We're actually seeing is it okay is it within 20 m? If if yes, then cool. and then would push to this next table here. Um now but yeah if you if you weren't if you didn't need that then I would just skip to this stage where first of all I would always validate right um is this company that I'm targeting is it my ICP and in this case right it's e-commerce. So what we're first of all we're doing was we're checking is this company within e-commerce right um it's it's a client prompt um so a simple one to be honest um but yeah it's just checking if they have things such as like c by now right that type of thing so just keep it very simple um and then from there this is probably the >> are you are you um your screen isn't shared by the way it sounds like you're talking through >> oh okay so maybe Um again, >> yeah, I I um yeah, I agree with what you're saying. Basically, lead qualification is like because I think that that's the mistake a lot of people make is they're just um reaching out.

It's spray and prey, right? They're just reaching out to a bunch of people. >> Um and that's why they see low response rates and they think cold outbound doesn't work. But if you can reduce the size of that pool but reach out to relevant people then um well you're going to get higher response rates but also your deliverability isn't going to get killed because more people are replying which is a positive feedback loop. Um so yeah I'm totally on board with that. >> Yeah for sure. like um and that's and that's yeah just just going to your point like if I was someone who I just like I had some you know some ability in clay or whatever and I just had like a account list this could be a flow for them. So yeah like I said you just pull an account list into clay um and then you just check yeah do that account qualification either e-commerce or not or just edit it to wherever ICP is. Now even this I would I would even skip this. This is very specific to like uh e-commerce, but what I'm doing is I'm actually um pulling out one of their products on their website just to just to call out, right? Um so so yeah, that that's that. Um but I would say like yeah, you could skip that if you're not doing that. You just skip that step. And then and then here, this is where you just find the context, right? So um I'm pushing all the qualified accounts into another table here. And then what I would say is this is where you're finding the this is like the the key prompts here. So in this case what we're doing was we're trying to find using the internet to find um these personas right. So you would put in whatever personas that you want and the key thing is output it results in JSON. So then if um so then let's say where's one of the results here?

I have three general managers, right? Confidence is very high. So then I can push these into another t into the next table. So I have essentially a a table with all contacts right that I need within these companies. And then at this stage you're just essentially yeah kind of as you can see these are all the comp people and then you're just doing again the simple waterfall uh to find find those work emails validating it and then you're just pushing it to um your sequencer. So like I know it's complicated but it's probably just like two steps really interesting. Are you seeing any patterns with regards to calls to action or offers one ones that's tend to to have I know it's the offer is going to vary across company and industry right but there any patterns you're seeing. Yeah, I I would like when we're on boarding a new customer, like I would, you know, you can always tell if like this this campaign is going to rip, like we're going to get positive replies. Like this the one the clay table I showed you just now like we knew like it was just going to fly. Like we just knew because they were offering like you know three um like free months up front type of thing um for like you know for um for their services for their products. So, and that value of the value of that is a minimum of like 3K, right?

>> Mhm. >> So, it's like that's like if you have an offer where uh yeah, if you have an offer that people would pay for and you that's Alex from home like people will pay for that's always going to do well in cold in cold market. Now, of course, not every client has that, not every company has that. So if you don't have that then I would say the next best thing is providing like insights. So because a lot of the time people aren't in buying mode right now right they're in information gathering. So if you want to increase your positive reply rates how can I provide information that's really valuable that people might pay for right? So that could be like okay can I share you a twominute video on how we've helped CMOs um you know reduce churn rate by 50%. right in the healthare space right so it's like a very specific CTA for that particular particular person and it's pure insight right and so yeah that that's what I would say is like the second best makes sense yeah yeah yeah and would would you guys normally construct an offer so like say you on board a client would you usually help them construct an offer for these outbound campaigns or is it common for them to come to you already having an attractive offer Yeah. Well, it it's it's a it's a mix, right? Some some some clients, like I said, will have like a great offer already and it's just we just add our coding more finesse and expertise and then we'll kind of put that on steroids.

In some some cases, it they're earlier stage, right? Um they don't have that product market product market fit. So, yeah, they come to us for testing, right? Right. And that that's what Kodi was really good for is being able to test like at scale what what is landing for various industries and testing different offers around that and um and yeah and and I think at the very least you know you should be doing that to just give your client as much insights as possible. >> Nice. Nice. Yeah. Yeah. You know um that's why I always like cold outbound is you can especially with the tools available today. I've done it for like a couple of businesses where literally within one day, two days, you can know if you have something that's attractive or not. Uh, and if not, then you can, you know, make changes and then know within another day. Um, there aren't many. I mean, you can do that with paid ads, but how much are you going to pay to to get the same level of insights? >> Yeah, exactly. Yeah, it's just going to be too expensive. So, yeah, cold email definitely the the most efficient way to do that.

>> All right, man. Well, I appreciate you running through these workflows. Um, anything else you want to share or we can we can finish it there? >> No, no, all good. Um yeah, appreciate you having me on and um yeah, we'll uh look forward to see speak to you