Find the Hidden 3%: Turn Dead Outreach into Revenue with Claude Artifacts

00:00:01 - Bayram:
On the topic of AI in sales. And today it's going to be practical again. Like, we had two lectures, now two workshops, and the previous one was focused on N8N. By the way, let me know if you were part of it. We learned how to design the meeting prep and meeting evaluation aspects of sales meetings, account executive meetings, and now.00:00:31 - Bayram:
I want to show you another simple but very powerful functionality that Cloth AI provides, which is the cloud artifact. But before we jump into that, I want to explain a couple of things around sales data and cold outreach data, specifically, because I think if you have a framework of understanding how to think about.

00:01:02 - Bayram:
Sales process and how to think about data that is generated by the sales process, then it's fairly easy to understand how to prompt AI to do the stuff you want you want it to do. Okay, so I'll share my screen, my iPad screen in a moment. Hold on. Now, this is.00:01:34 - Bayram:
This should be the right one. Let me know if you can see my iPad. And I would love if you mute your mics just to make sure that background noise or talks. Do not interrupt us. Awesome. That's great. So basically, there's this.00:02:03 - Bayram:
A great book by Intel CEO, one of the Intel CEOs back in the days. And in the second chapter of that book, basically he describes what is the minimum set of metrics to evaluate any process, both manufacturing processes and people processes. And obviously, cold outreach is the people process, where essentially we have.00:02:34 - Bayram:
People or maybe some automation AI agents that perform outreach on behalf of you or your colleagues. And you want to make sense of that process. You want to control and manage that process. So the minimum set of metrics is this. So let's say this is some kind of process. In our case, it's going to be a cold outreach process. So essentially what you need is.00:03:05 - Bayram:
Is three things. Basically every process is associated with some kind of input. So you want to make sure that you can measure the amount of input that went into the process. So for the case of cold outreach process, what do you think is the input? What is the input to the cold outreach process? You can unseen the chat.00:03:36 - Bayram:
This way we can. Leads? Yes. Essentially it's number of leads or a list of leads that we want to touch. Right. Could be leads that we extracted from a database or we use some service like Apollo or something. And then basically this is what comes into the process. Then in scope of this process, we send emails.00:04:06 - Bayram:
Or we send LinkedIn connection requests, we get responses. And essentially, what's the second minimum metric that we should track is the output, which makes total sense. And essentially, what is the output of the cold outreach process? What are we trying to achieve? What is the goal of the cold outreach process? Yeah, in cold outreach,00:04:38 - Bayram:
Probably it's not sale per se, but rather like a meeting or something, or some kind of positive interested response. So I will focus on meetings because in B2B sales it's primarily meetings, but in B2C it could be a purchase, like a sale. That's right. Or it could be just a click clicking on.00:05:10 - Bayram:
If it's an email outreach. So maybe it's just a click and visit of your blog post or something. But essentially, yes, for the example of B2B, it could be a meeting, or I would say an interested lead or a warm lead. We have two things, and the third aspect of this is basically something else. Who can guess.00:05:41 - Bayram:
What's the third metric that we need? So basically it's just three metrics, the input and output. And if we divide output by input, we get the conversion rate. But there's something more important here, something that's very important when you want to control the process. Think about people who, for instance, you manage people who send emails to cold leads, and then they say that, hey, for instance, last week we had 10.00:06:14 - Bayram:
Warm lists or 10 responses. What's the third aspect of this process? That's very important to manage the process. Or you can take another process. Let's say it's a process of serving food in the restaurant. People come in into the restaurant. That will be an input. People come out of the restaurant of shop. That could be an output. But there's something else. There's something else.00:06:46 - Bayram:
Any thoughts, Any ideas? Well, not. Not really. Some processes may lack the money aspect, per se, so something else. So just. Just listen to what I said. I said. Last week we had this. This week we had that. So the. The third aspect is time. In, in this case.00:07:15 - Bayram:
It's the time it takes for input to become the output. Because let me give you a couple of examples. For instance, if it's called outreach via email, probably the average time given lead spends could be days. But if it's a process of serving food in the fast food restaurant, it could be minutes.00:07:46 - Bayram:
In the case of, for instance, an RFP process, it could be months. What's important is time, because we want to understand how much time given input spends in a given process. Because if we have these numbers, then we can not only manage the process, we can actually predict the process. Meaning that if I know that the average time.00:08:16 - Bayram:
Time that elite spends or before responding is, for instance, two days. And I know that the conversion rate is 10%. And I know that yesterday we sent 1,000 emails. Then can you tell me how. How many warm leads will we have tomorrow? Well, obviously you can, right?00:08:48 - Bayram:
If this was yesterday, then that means, and the average time is two days, then that means that in two days we will see the output of this process, which is 10%, which is basically hundred worm leads. That's why time aspect is important. And essentially what he says, this is enough. Like any process, if you want to manage, if you want to make it predictable, you want to measure input and output to the process and you want to.00:09:19 - Bayram:
Measure time it spends. And all of the other analysis is basically like derivative of these specific analysis. We want to analyze the conversion rates, we want to analyze the time and things like that. Now to the cold outreach process, Essentially, if you're doing any type of cold outreach, what's the average conversion rate or what's the average.00:09:50 - Bayram:
Say interest rate in your campaigns share. What's your average? Is it 10? Is it 30? Is it 1? So what's the average conversion rate for the cold outreach process? 2.2%. Yes, on average. Our benchmarks.00:10:20 - Bayram:
For instance, if we're using email or LinkedIn, it's around 2 to 4%. This is a very healthy percentage. It seems like you have exactly in the space. This is very healthy. Getting to 3% on average is a very healthy metric. But what's interesting is that.00:10:50 - Bayram:
In these 3%, since we send different kind of emails at different times to different people, essentially what's possible is that, for instance, some combination actually is 10%. For instance, some segment of our target audience converts much better or much faster, for instance.00:11:21 - Bayram:
Sorry, for instance, one day. But there could be a segment or micro segment that converts only with 1%. And it takes, for instance, 10 days. So essentially what we want, we want some help, some assistance to avoid making decisions on the average.00:11:52 - Bayram:
3%, but rather dissect that those 3% into, for instance, segments like this. Basically a segment to double down. Right? Because if we identify a segment that converts much better and much faster than we want to focus, we want to double down on this kind of segment or the sec, the second type of analysis.00:12:24 - Bayram:
We want to perform is segments to avoid. Well, essentially, any cold outreach process costs us money or time. Time could be converted to money. So if we identify a very like, maybe not the 1%, but say 0.1%, then we want to avoid these. Sorry. Yes. So.00:12:58 - Bayram:
Yeah, mute. Okay, I muted. Awesome. So please mute your mics because this. This just disrupts and interrupts us. Okay, so getting back to it. So the second segment that we want to. Second type of analysis we want to perform is basically segments to avoid, essentially, if we're paying or spending time.00:13:24 - Bayram:
On this. We want to save on that. The opportunity cost is so low that we don't want to do that. And very often there's a third aspect. There could be some things that are unexpected, like some insights that you had no idea about. And this just feels not in sync with the yes, of course there will be a recording not in sync with.00:13:56 - Bayram:
Essentially what you expected, some insights that you know, differences, some some outliers in your data compared to, for instance, what would be expected if you are part of my previous seminars, I told you a story about the video message that we used for for for cold outreach through LinkedIn and video message actually performed.00:14:26 - Bayram:
Really bad. It was the worst type of message in terms of performance. And this was very unexpected for us. We expected that video would convert actually much better. Then we realized what's wrong with that video and with the message that complements that video. But that's another story. Essentially, the analysis we want to perform is segments to double down.00:14:56 - Bayram:
Segments to avoid and segments to insights that are unexpected we had no idea about. So that's the framework that we have for any process. And just believe me, you can apply this not only to the cold outreach process, basically any process, manufacturing, service businesses, you can. Basically, if you have this framework, then you can adapt systems, systems of records to make sure that you.00:15:26 - Bayram:
Count these numbers and then leverage AI to help you analyze this data. So now to the second part of it. How do we leverage AI to analyze this data? Well, let me switch to. Okay. Okay. You should see my computer screen now. So how do we do that? That's where.00:15:56 - Bayram:
I want to show you one simple but very powerful feature that I expect all of the AI assistance will provide in one way or another. And actually, there's an earliest sign that this is the case and that can actually replace the data. Scientist, my friend, calls this wipe analytics. There's wipe coding to code in this wipe state. So he calls it wipe analytics because you basically.00:16:27 - Bayram:
Why you don't need the data scientists to reveal those insights, to reveal those segments that we discussed. But actually, you can then tune the prompt and use it for any kind of analysis that you like to any kind of data. Maybe your personal finance data as well. So let's start with the cloth. So essentially, what cloth?00:16:57 - Bayram:
Recently added is the concept of artifacts. Do let me know if you tried building at least one cloth artifact before. Let me know in the chat, but in Claude I'm using a desktop version, but actually the web version looks similar. Let me show you. You have a section, this one, that's called artifacts, and that's exactly.00:17:27 - Bayram:
The section that we're going to use. What is an artifact? Artifact is basically a mini app, a mini website that can do what you want to do, for instance, visualize a data or for instance, visualize some educational material, as we can see here for the Molecule Studio. Or some kind of converter, you see Python to JavaScript Converter.00:17:58 - Bayram:
QR code or for instance, I was part of Hackathon recently and we needed to sign a consent form. The organizers, they shared the consent form in the Word in the Word document, as a Word document. And it was very hard to actually sign the form from your mobile device because you had to open the Word export to PDF, then sign by PDF, blah, blah, blah. And they didn't use Docus.00:18:29 - Bayram:
Sign or something. So what I built is an artifact that basically lets you read the consent form, click the checkboxes that you agree to this consent form, and then why I click basically sign the form, get the signed PDF, and that PDF you could send to organizers. But what's more interesting that about maybe a month ago, Anthropic introduced an additional.00:19:00 - Bayram:
Very powerful capability that basically lets you call Claude inside your artifacts. So before, artifacts were more or less static. Like you could just visualize something, but you couldn't add brains to these apps. Those apps were pretty static, pretty dumb. But now you can actually call Claude from within the app from within the artifact. And this gives you a very powerful.00:19:30 - Bayram:
Capabilities. You probably guessed that for the kind of analysis that we want to perform, we actually want an artifact where we could upload our data and then make a call to Claude. And Claude would actually analyze the data and respond with those interesting segments that we talked about, like the segment to double down on or the segments to avoid and things like that.00:20:01 - Bayram:
So I want to show you how this is done, and this should be. If you can follow along, that would be great. So I would just need your feedback that you're following along. If you open the artifacts, just click on new artifact and let me know if you're there. Just a plus sign or something in the chat if you can follow along. If you're a desktop, you can actually perform these things.00:20:30 - Bayram:
Next. Okay, so let's just give probably a minute. So essentially, again, it's in the cloud. In the left part, you can see the menu section called Artifacts. So that's what you want to use. Okay. I think just Merdan is following along. That's fine. So here you can see that we can actually build different things like games, templates, quizzes. And this is actually very.00:21:01 - Bayram:
Useful anthropic shared a couple of like a research a couple of days ago about the way educators use anthropic cloth in their teaching rooms in their classrooms. And it seems like actually artifacts is a huge like element in their in the AI powered way of teaching because they use artifacts to build some cool visualizations or quizzes or projects to.00:21:31 - Bayram:
To explain different concepts, but we will just start from scratch here. And if we choose start from scratch, this is where Anthropic Claude would just ask us what type of thing we're looking to build, what's the main purpose and things like that. I actually prepared the prompt, which I'm going to share with you, and of course I'll share it.00:22:03 - Bayram:
With this webinar recording and materials. Basically, this prompt does the following we instruct the CLAUDE to build a cold outreach dashboard because we want a reusable kind of artifact that lets you upload the CSV with cold outreach data and then call the CLOTH to analyze that data and visualize the results. It's very important that we.00:22:33 - Bayram:
Have an instruction here that a certain built in way to call CLAUDE AI is used because this instruction is very important, because without it, sometimes CLAUDE forgets that it has this capability. So you want to provide explicit instructions to do that, then basically we need to specify the actual prompt that will be called after this called outreach Data is app.00:23:03 - Bayram:
Uploaded, the cloth will be called with that prompt. And this is exactly the kind of prompt we want to call Claude with and the kind of information that we want to get. So essentially, what we want to get is what are the top segments, those that we want to double down on, the avoid segments, those that we want to avoid, and some immediate actions and insights that Claude thinks are important.00:23:33 - Bayram:
Here. So I'm just gonna copy this to Claude and execute this. So what happens now is that Claude actually builds an app similar to Lovable Bolt and all the others. But what's great about this app that there is an option to call Claude.00:24:04 - Bayram:
Built in option with any prompt, and I will show you that in a moment. But basically this is a very powerful thing because I think about this as thin clients, basically, when the server is actually your LLM, in this case anthropic, you just built a very thin client, which is a visualization that we're going to get.00:24:33 - Bayram:
In a moment and you instruct the server what kind of information or what kind of analysis you need, and then it does the thing. And that's very powerful thing because essentially via prompt, you can instruct Claude to do anything you want, anything, you know, Claude can do, and instruct that and have that built in to your into your web app. And there's an added bonus.00:25:03 - Bayram:
That essentially, when you share, you can publish these artifacts and share them with your teammates and things like that. When you share that, you don't spend your money, your tokens on this analysis. Well, cold outreach is a pretty extensive analysis. It could be thousands or tens of thousands emails sent. And you don't want to spend that money on this analysis for every user.00:25:34 - Bayram:
So what cloth artifact allows you to do when you share, it instructs the user to sign in via cloth. And this way, that person's subscription, that person's tokens will be used to generate the insight to basically execute that prompt. Of course, maybe in some circumstances this is not what you want, but in many circumstances, this is exactly what you want. But the most interesting part.00:26:05 - Bayram:
Actually happens here. So you see, even if you don't understand software or programming languages, this is what I'm sure you will understand. What happens here is basically, it's exactly the prompt that we shared and that we instructed here. It basically says that, hey, analyze the data, blah, blah, blah, then it inputs the data that we uploaded.00:26:37 - Bayram:
And it just provides some instructions how Claude should format the response. Because this format or the structure will be used to visualize the results. And if you don't understand, you can always just ask to explain. So you can just select this and ask to explain the most interesting part.00:27:07 - Bayram:
Happens here is basically, as you can see here, it tries to call anthropic and essentially this is the bad way of doing this. But I want to show you because again, it didn't realize that it should use some built in capabilities. So we will instruct it to do that. But let's say you have the same issue.00:27:38 - Bayram:
And we will just upload a sample CSV file. I have a guest for you. So you see, this is a sample outreach data file that I generated with the Cloth or chatgpt. As you can see here, we have company information, industry size, contact name, title, blah blah blah. And what's important we have whether there was a response, whether there was a meeting book.00:28:08 - Bayram:
And what was the time? Remember those three aspects that I talked about that every process should be tracked with? So we have how many leads we reached out to? We have 48 here, but that's fine. This is just a sample. Then we have how many of them actually replied and booked a meeting with us and the time it.00:28:39 - Bayram:
Took. So this is the sample data we want to upload to our dashboard. And here there's some basic information about the data, how many records, columns and things like that. But our most important and most interesting stuff is the analyzing with Claude, so let's wait a bit.00:29:10 - Bayram:
And see if it works. By the way, if you're following along, let me know if you got to the point where basically you have the dashboard and you can upload the CSV data and you're waiting for the CLAUDE to perform the analysis. So obviously it's taking time, but.00:29:40 - Bayram:
Here we have a response, and it's exactly the kind of actually, it worked. And this is interesting. Probably anthropic changed something. That's great because it used to be the case that when this type of calling anthropic didn't work. But it seems that they addressed this, which is great. And this is the exact kind of analysis that we want to have. So essentially, we want to know what.00:30:12 - Bayram:
What are the top performing segments? And if you review the data that I uploaded, then you will notice that essentially this segment, basically the SaaS VPs at Series B at basically 50 to 200 employees, they have a very high reply rate, actually 100% and a booking rate. Of course the sample is small and we can upload much more data. I'll show you that in a moment.00:30:43 - Bayram:
But overall, this seems like a very important and interesting outlier, the kind of segment we want to double down on. Let's double check just to make sure that it's here. Essentially what it was saying that SaaS companies, Series B VPs, they have a high response rate, 100% response rate and 80% booking rate. So essentially we would expect that these.00:31:13 - Bayram:
Four here. They have both the response rate, but maybe one of them doesn't have the booking rate. Let's see. Yeah, and that's basically it. You see, three of them both replied and booked a meeting. One just booked and three booked and replied, which gets to the kind of data we have here, 100% reply rate and.00:31:43 - Bayram:
80% booking rate. And again, what's great about this is that it doesn't really matter what our data looks like, because this, this data is sent to the LLM and LLM actually knows and decides how to perform the analysis. But the only kind of prompt, the only important part is what is the format to respond.00:32:14 - Bayram:
To provide this analysis. And then what web app does is basically formats, visualizes that data and we can see some of these segments. Now the second part, the segments to avoid. So basically, if we send the pitch video emails to companies over a thousand employees, well, it's lower. It's basically 0% reply rate across nine attempts.00:32:44 - Bayram:
So what I realized that sometimes you don't have the brain power to actually think, compare your rates to the benchmarks and things like that. What's great about Claude and others is that basically they will even make an analysis like a data scientist would do. But what I like about it is that actually that analysis is sometimes much, much better than the kind of analysis I get.00:33:14 - Bayram:
Whether from some data scientists, because Claude has any. LLM has a concept and information and knowledge about all of the domain of cold outreach, and it can actually make some decisions and some conclusions about the kind of actions we should do, the kind of benchmarks we should look into. And here are the insights.00:33:46 - Bayram:
And let's say I don't like the way this presented. Let's say I want to add some. Yeah, the API endpoints. Yes, that's right, man. But let's say this feels a little bit like just a report, analytics report. Let's add some visualization. Well, here, basically, you can do whatever you want. You just want to instruct Claude.00:34:16 - Bayram:
So let's say please visualize the data. For instance, a pie chart of the breakdown by industries. Breakdown of pie chart. Breakdown by industries. Number of leads. Pie chart with number of leads.00:34:45 - Bayram:
With a breakdown by industries. And we can obviously instruct and this will be added to this artifact, as we can see now and essentially. Now, let's wait a bit.00:35:20 - Bayram:
By the way, just FYI, sometimes Claude Synette does bugs. But if there's an issue or a bug, you will see a small button that similar to what we have in Lovable and Bold, that says something like fix it or something. You can just click on that and it will eventually fix it. Okay, let's try again.00:35:47 - Bayram:
Okay, we see that some of the data was immediately analyzed because this didn't require analysis by an LLM, because, you know, having a pie chart by, by industry, you don't need much brain power. This could be done automatically. So we have leads by industry, as you can see. And we, we have reply and booking rates by industry where we can actually see.00:36:16 - Bayram:
See what industry performs better and whether we want to focus or double down on some specific industry or some specific combination of those. And now we have that analysis that we talked about. Again, same kind of conclusions. And last but not least is being able to actually share this dashboard for this. We have this great thing of publishing an artifact.00:36:47 - Bayram:
And essentially we get the link and you can see that this link basically looks like similar and you can share this link with any of your teammates and let them upload any kind of data they want to perform the same analysis, perform the same information, or.00:37:17 - Bayram:
The same stuff you want to achieve. It could be, for instance, you have company wide way of writing emails. The prompt could be something like hey, write a cold outreach message text Cold outreach email text. Using these guidelines, then instead of sharing the prompts or sharing.00:37:49 - Bayram:
Some other way of doing this. Basically, you can just build an artifact, publish it, share it with your teammates, and if they have the subscription, then anthropic subscription, then they can basically generate any number of leads. Of course, N8N the tool that we reviewed on our last webinar is much more powerful. But what I like about this one is that this is built in. This doesn't require.00:38:17 - Bayram:
Any kind of specific knowledge. It's way more faster, simpler, but with the same kind of way powerful. And you don't need to know what API is. You don't need some API, keys, tokens and such. You just sign in with the cloth and you just use it. And that's, I think, one of the most powerful things that we could do here. And that's essentially what I wanted to share today.00:38:47 - Bayram:
Again, just to reiterate, any kind of process could be measured using three and banished and predicted with three metrics input output in time. So if you want to analyze cold outreach, then basically you want to measure those. We had those in the CSV form. This is 5,000. Let's actually try to analyze.00:39:18 - Bayram:
It and see. Yeah, of course. It will actually probably fail. Let's see. Let's see if we can do that. Okay. Yeah, it's. Yeah, it's too. It's too big. But I think you get the point. Basically, what we can do is.00:39:48 - Bayram:
Analyze any kind of data the way we want, and artifacts allow us to create the dashboard and share it without any programming language knowledge and without basically any data. Scientists Because I was always frustrated that I need to wait for days or hours to get a new kind of report or a new kind of dashboard, and many times that report lacked the business data.00:40:18 - Bayram:
The main understanding of the process that we're analyzing. That's where I think LLM beats every data scientist. So you can call it vibe analytics, but what I think is that these kind of tools will replace sales ops people and data scientists, sales analysts. If we're talking about sales process and we can do and get insights faster, we can share those insights and we can make these.00:40:49 - Bayram:
Dashboards much more flexible, as I demonstrated to you today, than hard coded predefined types of analysis that we used to in the pre LLM era. So that's basically happy to answer any questions you may have. You can actually unmute or you can just use chat for that.00:41:23 - Alua Baikadamova:
Hi, can I summarize this lecture? So you use how to say you use chat GPT these prompts to to do to use. So you use them in your sales projects, right?00:41:50 - Bayram:
Yes, that's right. Essentially I'm using LLMs. In this case it's Claude, but Claude has this built in ways to build apps and share them. I use LLMs to analyze our sales data and share the dashboards for the other teammates or with customers in our case. Sometimes you want to share some of your data with your customers. That's why this could be used to share with your.00:42:20 - Bayram:
Your customers as well. But yes, this is the summary and essentially what I think every big AI assistant like Cloud, Gemini and ChatGPT, they will make it very easy to build this kind of apps. And I think the lovables and bolts of this world face a great threat because overall, it's very easy to build these apps.00:42:51 - Bayram:
But what's great about it is that it's much simpler. You can reuse that, your login, your, you know, basically your user account at Claude and ChatGPT. So for those assistants, this is a way to actually, like, you know, increase your engagement rates and retention rates by keeping you in the ecosystem of this AI system. You probably noticed that with the release of ChatGPT,00:43:23 - Bayram:
They did a lot of Talk how great ChatGPT is for generating web apps. I think one of the reasons is that the web app generation is a very hot topic in terms of funding and et cetera. But it's relatively easy thing for an LLM for a powerful LLM, and that's why I expect these guys to build this kind of tools. And I want to just show you.00:43:52 - Bayram:
How easy it is to build these dashboards and get the analysis and basically maybe in many cases replace the data scientists, the product analysts that you currently use to get the insights you want to get support to give correct and readable names. Yes, Dmitry, that's a good question. Yes, it's important that column names are.00:44:22 - Bayram:
Self explanatory, but sometimes you may want to add additional context into your prompt. So, for instance, sometimes if, for instance, the column names are not self explanatory, then I would expand the prompt to explain what kind of data is in each column. This way LLM does a better job of formatting it the right way and of course, building the right kind of.00:44:52 - Bayram:
Insights. But like most of the times, if they are self explanatory, it will understand. But if you notice that it can understand, then just add some context about the columns. Is there any way to give access to that for certain your own so with your own tokens to spend? Not yet. Not yet. This is not possible through like through artifacts at.00:45:22 - Bayram:
The moment. I think, Claude, in one of the releases, they mentioned that they may add this, but you gotta understand their business strategy. The business strategy is to use this as a form of a viral kind of mechanism, you see? So something like Windows and apps or iOS and apps. So the more people build artifacts, useful artifacts, the more other people, teammates.00:45:52 - Bayram:
Etcetera. Will want to join the ecosystem to leverage the power of those web apps. You see, that's why I think it's in their best interest to not do that again. You got to understand that. You got to basically understand that there's a huge difference between the business people who need insights and analysis.00:46:23 - Bayram:
But don't know coding and programming skills. And this looks very simple and easy and accessible to them and developers that build products for others. That's when you probably want to have your own tokens to spend. But we'll see. Maybe I think I saw somewhere on Twitter or maybe in the blog post release that they might add that in future, but not at the moment. Okay, so.00:46:52 - Bayram:
Sounds good, then I guess that's it. Hopefully the simple but powerful, powerful thing will help you get insights faster, build dashboards faster and share them and educate your teammates about the segments to double down on and segments to avoid. And basically increase the reply rates and the booking rates for your cold outreach and any other process that.00:47:22 - Bayram:
That you're managing or predicting. Thanks very much, and you have a good rest of the week. Bye.