# 65 The Marketing Industry Solved Measurement. It Was Always the Wrong Problem. | Peter Grafe

Peter Grafe: And actually carried out a global survey with global survey interviewing not customer Tesla customers but like for example just uh, the normal population for and then started to understand how do they actually perceive the brand Tesla. And what we found was that there were three main points.

Vishesh Duggar: I'm your host, Vishesh building Neo voice control for your entire computer. No typing, no. No clicking. Curious. Check out Vishesh space. I am joined today by Peter Graffi, founder at Blue Alpha, the AI ah, native performance marketing hub. A data scientist who built Tesla's marketing measurement framework from the ground up. He discovered the industry had solved the wrong problem and left to fix it. Blue Alpha helps growth stage companies close the gap between measurement insights and the decisions that actually move the needle. Hey. Hi Peter. Thanks for joining me on the podcast.

Peter Grafe: Thank you for the invite. Pleasure to be here.

Vishesh Duggar: Awesome. I would first like to thank Scott, uh, Persinger from Biz Trip AI who introduced us. He's been a guest before and yeah, we chat on and off and so yeah, good to. Good of him. M to actually make that intro.

Peter Grafe: Yeah. Fun fact. We actually like we used to be in the same office, right? In the same co working space here and we moved out since then and I just invited him to our office party which is happening uh, end of this month.

Vishesh Duggar: Yeah. So you met him at the coworking space or did you know him from before?

Peter Grafe: No, I met him at a co working space in. Well, it is. The concept is pretty cool because it's called Frontier Tower in San Francisco and it's an old rework and they bought the entire tower and they turned it almost into a vertical village where every floor had like a uh, specific theme. And then you had like an AI floor in the robotics products floor, Biohacking, growth hacking like makerspace and then also yet on the third floor like these office cubes, you know, where you like it's not co working but like it's like shared offices on the same floor like wework style. And we had like two cubes down there and Scott also had a cube and that's how we met because we're actually top the first three office, uh, or the companies that moved into that place. So we're like the, the OGs, you know. And now the OGs graduated out of the place and we're moving out and think they're moving out in like uh, in a month.

Vishesh Duggar: Yeah. Nice. Cool new offices.

Peter Grafe: Nice new offices.

Vishesh Duggar: I read somewhere you um, spent some time in Leuven as well. Leuven in Belgium?

Peter Grafe: Well, I did write my master thesis in the company, uh, at a company that was based, uh, in Leuven, I was actually. I was physically. I was in. In. In Rotterdam.

Vishesh Duggar: Ah.

Peter Grafe: So during my master's, which was data science masters, first data science masters, they actually ever did, like, at the Erasmus, uh, university was like, the first cohort had to write our thesis tied with an internship. Right. So when this, like, mandate was basically set by a university or a program manager, it was like, hey, you need to have an internship, and with the internship, you need to write your thesis, which I loved because it was, like, practical and you could actually, like, apply stuff. I started to apply a bunch of different companies and actually got an internship at, uh, Tesla. And I was like, Tesla, pretty, Pretty cool, you know, like, not. Not bad. And then I was under the impression that I could write my thesis at Tesla. And after I got the internship, they basically told me, like, no, you cannot. So then it was a bit, like. It was a bit of a situation, uh, because I still had to have a thesis internship. Right, Right. But Tesla was like, because of data privacy and regulations, all that kind of stuff, like, it will be very hard to, like, do that. And then I was like, well, I don't want to give up the internship at Tesla. So I actually took a second internship while I still had uni, Right. University, like, courses, plus the Tesla internship to actually fulfill my, uh, master's, like the master's thesis internship. And it was definitely crunch time. I would be lying if I did an incredible job in my master thesis, because I didn't. Um, at some point it was clear that I would get hired, uh, by Tesla. And then the master's degree just became more secondary. Right. And, like, Tesla moved into. Into the focus. But I wrote my thesis on multiple sclerosis, um, and where you can start to simulate actually, like, certain treatment effects. And. Yeah. Was interesting.

Vishesh Duggar: Yeah.

Peter Grafe: But it was. It was crunch time.

Vishesh Duggar: Yeah. I don't know how much time you got to spend in Leuven. It's an interesting town. I, surprisingly, I. I've spent some time in Leuven as well.

Peter Grafe: Really?

Vishesh Duggar: All places. We had one customer based out of Leuven. And so the first trip I actually stayed in Leuven. And then I realized this is not a town to stay in. And so the next trip I actually commuted. Yeah, interesting place. You spent some time there or.

Peter Grafe: No, no, no, no. I. I was actually still in Rotterdam. But the company Ecometrics, that I just, like, did the internship with, was based in or wrote the thesis with that provided the data. Right. Was based in, um, in Belgium, but Unfortunately, or fortunately or unfortunately, I never made it actually to Leuven.

Vishesh Duggar: It's a nice, quaint little university town, but yeah, it can get boring. Yeah, I don't know how big the

Peter Grafe: Belgium viewership is, you know that you

Vishesh Duggar: may be upset now, probably very little, but I mean I love the country. I've been there multiple times, so no hard feelings. Belgium. So you spent four years building measurement systems at Tesla that no one outside the company ever saw. Right. And at what point did you realize you were sitting on something you could commercialize?

Peter Grafe: Yeah. So the journey is actually, it wasn't like four years building marketing measurement systems. So I started as a data scientist at Tesla and I was in the centralized data science team that was serving all of the EMEA um, countries with dashboarding, reporting, data analysis, but also like data science models like for example, predicting what kind of clients are going to convert from a lead to an actual order. Or, or once we started to deliver um, cars, can we actually start to predict um, cancellations in deliveries? Because if we would have known beforehand that a customer is not going to take delivery of a car, there was like a whole chain of events. Right. Because you schedule the appointment and then if they're for example, like cancel right before then we almost have like a drop of like a couple days where we need to like schedule someone else to pick up their car. And at Tesla we're like end of the quarter was like crunch time because we wanted to beat like our previous um, um, record in deliveries like every single times. That kind of information was super crucial. I've done a lot of topic classification as well where we scraped external forums and to move like internal um, survey data. Ah, especially like open questions of like hey, why did you, why did you churn, why did you like leave uh, the brand? All that kind of stuff. And this was like very early um, when ChatGPT just came out and topic classification was like the, the main, um, the main application for these type of things found some pretty interesting insights that helped for example shape the roadmap for the Autopilot team in terms of like what kind of features we actually had to prioritize or what kind of problems um, users found as well as just helping us understand how. And this was particularly in the European market. People thought about the brand Tesla and what was like the purchasing discovery process and what was like top of mind, like for example range anxiety or how to properly charge your battery. All of these insights were then actually like incorporated in our sales advisors, um, sales flow to give them the right tools at the end that they can actually bring us up when talking to clients. So this was me working, um, in this data analytics and data science team in Amsterdam. And then a time came where Tesla went from Tesla, uh, how Tesla builds cars. Maybe it's just like some background information is like if you go to BMW and you say, I want to order this car, they built that one car. So the thing is you put an order in and that car is then being manufactured. Tesla does not work like this. Tesla builds cars that are then being matched to orders, right? So it's like a supply and demand, uh, equation and there's like a whole, um, demand or like demand forecast. And the whole supply forecast, right. And when I joined, which was in 2020. 2020, 2021, um, Tesla, ah, only had limited supply and halfway through the quarter was already like sold out, right? So people, there was actually too little supply for too much demand. And people had to wait quite some time for their cars. Not because they still had to be like, not because their order didn't exist, but because there was simply like no supply. And then what then happened was obviously, um, the gigafactories. And we also had Giga Berlin, right, that like big gigafactory in Berlin started to ramp up and started to produce a lot of more cars. And the demand did increase, but just didn't increase as strongly, for example, as the supply. And then all of a sudden we found ourselves in a bit of a, uh, demand gap situation, you know, where we had a lot of cars on the ground at the end of the quarter that were not delivered, which you don't want because these cars just like sit there. And then you need to pay like parking costs and all that kind of stuff. And Tesla obviously kept continuously shipping new updates, new feature, um, new like versions of the models as well. So we had to like get rid of them, um, to look for new alternatives of like growth. Right? Like, what else can we do? And the topic marketing kept coming up but was like a highly, Was almost like a taboo because like Elon very famously says, if you just build an incredibly great product, you know, you don't need like for example, marketing. And that statement by itself is partially true because I think you need to build an incredible product. And to this day I believe, like, Teslas are like such an amazing product also for the price that you get. And at any given point in time I would choose a Tesla over even a BMW or Mercedes. And now probably the German viewership is going to be on me. But this is just what I actually believe, I truly believe in the product. But at the same time you also need to have publicity or like that people need to be aware of your brand so they actually buy it. And Elon is just an incredible marketer by himself of like, everyone knows the brand Tesla. Do you know the CEO of BMW or Mercedes or Volkswagen? No, like I don't. Right. Like, so the thing is, it's like he was just such a polarizing figure that went from he's going to like save the world with sustainability that during COVID that entire image obviously like shifted which actually had an impact on sales and deliveries in particular in EMEA and also the U.S. so we were faced with this problem of like, how can we accelerate demand? And actually carried out a global survey with global survey interviewing not customer, Tesla customers, but like for example, just uh, the normal population and then started to understand how do they actually perceive the brand Tesla. And what we found was that there were three main points that were completely or three main core beliefs that were completely different from what the product was actually standing for. And these boiled down to like pricing, um, which ultimately affordability. Right. Safety and also usability of the product. And there was this belief that Teslas are still these like super high end expensive cars, which is like not, not true. Right. Like you can buy a Model 3 now for low 30s, which is, it's not cheap, but it's definitely an affordable car.

Vishesh Duggar: And it's not like premium.

Peter Grafe: Yeah, it's not premium. Right. It's not a 70,000 or 80,000 or $100,000 car. A very affordable model. And then the second misconception was safety. And people thought that Teslas are still these like unsafe cars because like during that time there was a lot of, there were recalls that actually happened to every single car, uh, manufacturer as well. Right. And there was also all these like videos of burning batteries and people actually had a feeling that Teslas are not safe, which is like completely false. Like the Model Y is one of the safest cars in the world, period, Full stop. You know, like, it's just like huge m. Misconception. It was actually, I think crowned in 2020 as like literally the safest call. And then the last point was more usability and product experience of like, oh my God, there's so much tech and actually no one knows how to use it. And once you actually have driven a Tesla, it's incredibly intuitive. And this goes back to the point that I will actually want to choose a Tesla over a Mercedes or BMW because the tech, for example, within the Cars is just not as good as a Tesla car. And the feeling itself of driving a Tesla is also incredible. And it has come a long way. So we faced with these three misconceptions and then I was like, what can we do in order to bridge the conception of or the misconception, for example, of that people thought about Tesla and ultimately we landed on paid media. Paid media is incredible because this is a very easy way on how you can push a certain narrative on other people at scale. And then obviously this kind of was a direct clash with Elon hates paid media, as he like very famously says that, uh, Tesla is never going to do like paid media. It does work, uh, all that kind of stuff. Interestingly, at the same time, or actually like before that, like while we were like going through this process, one of another, uh, Elon company, which is like SpaceX and Starling, actually started to already experience with paid media and started to see some success with it. And we actually used that to also start this initiative of like, let's get a growth team at Tesla, which ultimately meant like, we're going to run ads at Tesla, which was super slippery slope. You know, it was like, we got to be super careful with like the creative that we put out. And most importantly, can we prove impact of everything that we do? And every single dollar that we deploy needs to yield like incremental value, right? And excuse me, incremental value, meaning we need to achieve something net new which didn't exist before, right? Like get more leads into the funnel that then actually convert into gross orders. And we put together this task team and I was asked whether I want to lead marketing, uh, data science team, which ultimately meant answering the question, does this work right? And can we prove that this works? And I say lead the team. It was actually just me, you know, it's like typical test, typical Tesla fashion. Just like we'll put one person in there and um, see how it goes. And again, had no marketing background actually before that. And this was like year three of my Tesla journey of four years and very quickly realized that everything that was going on in marketing in terms of measurements and how to attribute and how to actually count orders was just wrong. And there was this one very concrete example is where we launched our first campaign in I believe it was Texas. And it was like YouTube and search and search campaigns are like, if you now go and click buy a BMW, for example, you see like a sponsored ads, right? That's the search campaign. And our Google reps told us like, oh, Google search, like the Sponsored search is like so incremental and you're getting such a cheap cost, you should put more money there. And I was like, hold up, stop. You know, like, how do you prove this first of all, like, can you prove this? And the answer was very unsatisfactory. Uh, and I basically couldn't explain to me why I should believe their results. Right. And then combined with the fact that Tesla is such a well known brand, we actually had the internal belief that search might not be incremental at all. And just because people are looking for Tesla anyways and they see now a sponsor that and they click on it. Right. So the conversion would have happened regardless, but we didn't have a way to actually prove it. And that's ultimately what I then ended up doing is that I built the entire like marketing measurement framework, which are like a lot of data science and machine learning models that we've deployed in emea, APAC and North America to actually flesh out what is the true incremental value that each of these campaigns add. Right. This was the first step and what we found with this particular campaign was actually that search was not incremental at all. And YouTube was the main driver. Right. All the YouTube advertisement that we put out, but no one clicked on the YouTube advertisement, so you couldn't like track it because when was the last time you clicked on a YouTube advertisement? Right. But people saw the ad and then went online, searched, saw Tesla, clicked on the search, sponsored ad, and that's why it got attributed to this. Even though this was actually not the source, but the source was YouTube and we're able to show that YouTube was actually driving most of it. And this then later foundation that actually also started to build trust with our leadership and finance team to actually start giving us more money. Right. And based on this like measurement system that I basically built for Tesla. And then so this is like the origin story of like how I got pushed into marketing and how we realized that there is actually like a problem in the market which is like an attribution problem at first, but ultimately the bigger problem that sits on top is like almost a decision making problem of what to actually do with these insights that you generate that need to be passed on to the marketing team and that there's like too many fragmented tools that you actually need to stitch together.

Vishesh Duggar: Interesting. And so how did you end up tracking people who are just seeing ads? Was it a survey or.

Peter Grafe: Yeah, so ultimately the, there's like two concepts, data science concept that we used. The first one is called a, uh, Marketing mix model, which basically is a probabilistic model where you can infuse every single data signal into right? And then you have one KPI that you actually want to, you have one outcome that you want to measure the impact on. Like this outcome could be number of sales created in this company from your first party data, right? So that's a marketing mix model. And the difference of that approach is that if you go to for example Google or go to a Meta or go to a TikTok or Snapchat or an X or whatever, they all have their own systems on how they measure impact, right? And what then happens is if you see an Instagram ad, you click on it, right? You go to the website, you don't do anything, but then you come back and you search for it and you click on a Google Ad and then you buy. All of a sudden Google says credit uh, to me. Meta says credit to me. And if you look at the both platforms you see like, oh, I got two orders. But then you check your actual order base and it's like one, one, right? So one, um, right. And that problem we solve by just looking at your like what is your first party data? Was your, was the true source of truth of like how many orders did you get? And then look at how much money have you spent over time? How many clicks, views and impression have you generated over time? Have you created variants? And what is the impact of for example pulling up or pulling down a specific channel on your first party data and then using that as attribution? And that's ultimately the marketing mix model. This is like one concept and the second concept is called incrementality testing which is do you know what an a B test is? Right? So you have like, yeah, it's like think about it as an a B test, but an a B test on a uh, um, location level where instead of saying you have two populations and you randomize half or like randomized sample of half of the population, for example, experiences the treatment and the other half doesn't. Now you actually do that more on a GEO level, for example, breaking it down by states where you can say, I am only showing YouTube ads in California and I'm using all the other states to actually build a counterfactual. And again you apply um, a uh, machine learning model which is like at the end of the day everything is a regression to some degree, right? You have a machine learning model that then starts to uh, assign certain weights to build actually a counterfactual that mimics the Sales, uh, for example of California. But then you don't show anything there and then you can actually compare over time. Now I'm doing something else in California and of the other states have built this counterfactual. Is there an uplift? Yes or no.

Vishesh Duggar: So just like an a B test, you do A there and B somewhere else. You can kind of compare. Yeah. Interest.

Peter Grafe: Yeah, just that ah, B usually is don't change business as usual. Right. And like A is usually you apply a treatment of like spending more, spending less. Yeah.

Vishesh Duggar: And so while you were building this at Tesla, one of the advantages you probably had is you know, massive first party data. Uh, you know a brand that generates organic demand without advertising and a budget of I don't know what the budget was like, but I'm assuming it's maybe millions. What actually breaks when you lose all that infra. So you now you don't have access to that. Do your customers have the same parity in terms of the breadth of data, first party data that you had access to at Tesla? And if not, what breaks?

Peter Grafe: Yeah, no, that's a, that's a great question. I think the, we actually didn't use user level data at Tesla. Right. The thing is I was, we're interested in aggregated data of like over time, where and what, you know, over time. So when for example, 1st 13th of May 2026 in San Francisco X amount of orders. So that level of aggregation you actually find in almost every other company as well deciding factor was like how much volume do you actually see? Because we need some level of volume in our metric that we want to measure in order to get convergence of the models. Right. But the volume doesn't have to be 20,000 or 10,000. Like it's enough for example to have 500 conversions per week or like and then have like something around like 2000 conversions for example per month or even lower, you have 1000 conversions per month also still works to actually get like a strong model that can give you an insight. So and Tesla obviously had a lot of data and a great distribution and that doesn't actually prevent you from like successfully deploying these models and starting to like generating insights and actions out of it. What scale gives you, it gives you more a uh, larger playing field to test multiple things all at once. And if you just have like less volume, it's just like you gotta have almost placed bets sequentially after each other and then you measure. At Tesla, uh, for example we were running like 40 different tests all at the same time and it was Just possible because we could, we just had enough volume also as we started to break down our different locations, you know, in order to get actually statistical significance.

Vishesh Duggar: So time to an experiment probably gets elongated and that in turn probably limits the progress because you can't run enough experiments.

Peter Grafe: Yes, for example. But I think these two things also go in tandem because if you're early, you haven't made enough, like you uh, you didn't have enough time to make a lot of mistakes, you know what I mean? And like have a lot of inefficient spend. So working together with us is also very easy of like, hey, this is how we should actually plan. This is how we launch a channel. That's how we make sure it work, works. Once it works, you lock it in and then you like grow, right? Or is the other players like you are actually a very established player. And then we usually enter and we're like, okay, there's like inefficiency happening here, here, here, let's start to reduce this and then start from here and then build it up again. Right? So the thing is, it's like it actually, it's almost like a misconception that these models that we've built at Tesla are only reserved to like the Fortune 500 companies, right? And like startups, for example, like our ICP serves as like a uh, series B startup that for example, we start to serve with our tools that have like high growth and still want to understand what is working, what's not working.

Vishesh Duggar: Maybe this is a good, good time to take a pause and talk about what you guys do. What's a uh, crisp definition or elevator pitch for how you help your customers.

Peter Grafe: So what we are ultimately doing is we're building the, the marketing org of 2050, right? And that marketing organization of 2050 is agent led growth. So you have multiple agents that all perform specific tasks, but you have human led strategy. And that interplay is actually very important because what AI has done, what AI has done an incredible job is it has like flatline, for example, how fast you ship, how quickly you can surface, for example, insights as well, right? But it still needs context and it still needs domain expertise and it still needs strategy. And the strategy part is very often still sits within the human, right? Or needs to be extracted out of the human into the system to then perform an action or an execution. And in particular if you look at uh, marketing agencies right now and like how they've developed is that previously there was a lot of fragmentation, you know, across the different like platforms and that's where the moat that agencies actually had was the execution and the allocation of resources across all of these different fragmented pieces. That fragmentation doesn't exist anymore because data availability is great and AI actually like closed that bridge. Right. So now the real mode is intelligence paired with like having a machine that can actually execute in real time for you. And that's ultimately what we're building at Blue Alpha, which is the future marketing org of like 2050 where we are starting for um, we're starting with performance marketing. And performance marketing means everything where you spend money on like platforms in order to generate more money. Right. And this can be billboards that you see out of home. This can be your TikTok campaign, this goes to your YouTube campaign or if you buy linear TV or like connected TV spots for example. So everything where you just spend, spend money on. We actually have in total seven agents that start to like cover the entire like marketing stack across like planning, budget allocation, creative execution, testing, competitive insights actually. Right. So competitive insights. And on top of it is still the human that actually uses all of these agents to surface information and then infuse it with context in order to actually like implement. And you can call it almost like AI native service of like having this like base or of agents that perform intelligence for you that then the humans uses in order to make like faster decisions better. Yeah.

Vishesh Duggar: So it's not just measurement. You also help with creatives as well. So the entire stack pretty much, yeah.

Peter Grafe: So on, on creative is what we, we don't tell you this is how your creative should look like. So we, we're not a uh, creative asset, um, creation company. But what we do do is companies have like thousands of creatives that all run right and they're like embedded into the campaigns. What often happens are like concepts like creative fatigue. And what that means is you had an ad running too long in a specific campaign and how often people are watching it and how people responding and starts to go down. Right. And this is a concept they call, it's called a click through rate, for example.

Vishesh Duggar: Right.

Peter Grafe: How many people are actually watching ads until the end? It might start at 10% and then at some point it drops down to 2. And then if it starts to drop down to 2, what we usually see is that the cost that you incur for like um, driving more impressions for example actually starts to go up. And the platforms don't tell you that this is happening.

Vishesh Duggar: Right?

Peter Grafe: Because they obviously still want to spend money. So there's like this huge wasted effort of like or wasted money. Problem happening of people spending money on ads that don't work. And what our creative agents do is they run actually every single day and they start to flag these things proactively to help you understand. Okay, I actually need to start to create my um, an to rotate my creative. That's the first step. And then the second step is that you can start to cross reference this across like different platforms because sometimes you use the same creative. And then ultimately the last step is to start to analyze what are the hooks, what are the messaging, how do they work. Right. But it's again, it's still surfacing information. And we're not like a uh, creative agency and so on that actually starts to produce the creative. We're like the information intelligence layer that surfaces all of these insights that then gets passed on for example to a platform or an agency that actually uses those to create better creatives.

Vishesh Duggar: Interesting.

Peter Grafe: And same for competitor intelligence, for example, right? Like understanding where your competitors, uh, are, uh, what channels are they on and so on.

Vishesh Duggar: Is there any third party data available on competitors?

Peter Grafe: Yeah, a lot. So for example, Google and Meta have ad transparency libraries. And what that means is every single ad that is aired is uploaded there. You don't know how many clicks, views, impressions together. You don't know how much money they spent, but you know what someone uploaded when, right. And you also know how long it actually like stayed up there. And you also understand what is the messaging, what is the content like, how do they position themselves. And for these, you just can connect to them via uh, API for example and you can start to like pull that data in and then start to compare certain things. And then there's other um, websites that are called like built with for example where you can actually that scrapes the entire HTML stack of your website. And a lot of times when you store, for example launch a new channel, you install pixels, right. Like for example Meta works with a pixel that is installed on your website to help them with their own algorithm to like for example increase conversion tracking to do better targeting. So you can actually use websites like this to understand how does the marketing stack look like of a specific company and then compare that to like your own company, company and based on this can actually build quite some competitive intelligence just with like publicly available data.

Vishesh Duggar: Yeah. And this library that you spoke about, the transparency library, is it just a library or do they have a, like a GUI version or a UI version?

Peter Grafe: Well, it is a UI version. Yeah. If you Google right now, Google Ads library, like Google Ads Transparency center, um, you can Google like you can handle search for your own campaigns or competitors.

Vishesh Duggar: Yeah.

Peter Grafe: And I think the is like these, these seven agents that together actually make the solution round. Right. And like you have companies that only do competitive intelligence, but only competitive intelligence doesn't help you as much because it's just like one part ripped almost out of context. Right. Same with creative fatigue. I start to know like these creatives are starting to like fatigue and I need to rotate them. But what you actually need is on top, you need to umbrell is the measurement system which we actually built first where every signal that you get you can actually like route through that system to get to understand is this moving the needle for me? Yes or no. And if you don't have that umbrella, all of the rest doesn't matter. And same with like content, um, optimization. What are we optimizing content for? Like how do you understand whether this is actually working or not? Right. So the thing is it's almost like you start with the umbrella which is the measurement which is like holistic across every single channel. And that's why we built first also like obviously at Tesla. And then you start to like push deeper down of now going into like okay, creative signals. You need a signal here. You need to understand what creative fatigue means. Boom, you got an agent. Same on competitive intelligence. Right? Same on just like what is happening on the platforms. Whoever like a reporting agent. Uh, that's the seven agent actually there just general reporting agent. And then all of this comes together plus a planning agent that actually helps you understand what is everything that I've done and how does the future look like all in one tool. Right. Paired with like our service um, as well that we bring to our customers to actually deliver the full end to end stack solution for performance. Uh, marketing.

Vishesh Duggar: Yeah, that's an interesting solution. You wrote that you know the triple M or the, what did you call it? Marketing mix model.

Peter Grafe: Marketing marketing mix media mix model. Yeah.

Vishesh Duggar: And the incremental testing.

Peter Grafe: Mhm.

Vishesh Duggar: Can I tell you what to do next? The industry has been selling measurement as the answer for over a decade now. Um, like can you walk me through a specific moment? You realize the problem wasn't measurement.

Peter Grafe: Yes.

Vishesh Duggar: And if not measurement, what was the problem?

Peter Grafe: Yeah, so this is a great question. So measurement is part of the problem but not the full solution. Right. And in particular in very uh, concrete example is the insight that you will get out of our models is you should spend more money on meta. Right. Or you should spend more money on meta campaigns that prospect for new people. Uh, that insight by itself you can actually not action on because it doesn't tell you which campaign do I need to fix? Which campaign do I need to spend more? Is my creative working? Yes or no? Right. These are all like the measurement models that we built. They don't answer that because they're like statistical probabilistic models that need a certain level of aggregation to actually like work. Because I need to have continuous inputs, um, over time in order to get a read spend that turns on and is on for four weeks, then gets turned off and then I have another spend, um, bucket. No model will actually be able to give you a read on this. So that's where we need to have like some level of aggregation. And when we first started out, and I also realized this already at Tesla, was the insights that I was like surfacing. The marketing team looked at me and was like, okay, what do I do with this now? And we actually had to build a decision making algorithm on top of this that took the insights but then transformed it into something that the marketing team could actually understand. And this is this like measurement to action gap where I think most of the companies break these days still because everyone has a measurement provider, but no one actually acts on their measurement insights because they just don't understand what to actually do with it. Right. So and we, I already understood this at Tesla. Uh, and it was as simple as we're measuring incrementality and if we don't see incrementality after two weeks, we're cutting the campaign done. Um, you know what I mean? Like we're cutting everything in that campaign or we have a certain goal that we want to hit in terms of like customer, uh, acquisition cost. If we're below that customer, uh, acquisition cost with our measurement, we're increasing spend until we're hitting that target, right? And once we hit that target, we leave it and then we leave it. And then we had like a, ah, retest cycle for example after like three or five, four, uh, or five months. And it sounds simple and it is and it was on purpose, right? But it's just starting to get actually like decision making frameworks in place that everyone starts to follow because only then you can actually start to understand what is working, what is not working. And then of course it's like a, it's an evolving process of starting to fine tune this. But what we see a lot with people right now is they start something after a week they start to freak out because like stuff is not like looking how they like like how they thought it would look like. And they just kill everything again, right? And this is like not a process how you can actually build continuously learning and continuous decision making. Because of marketing, you need to let some time pass in order to actually get a full read, right? So for example, if you turn on a new campaign, you cannot expect that within a week for entire, like, um, that you just generated 100% more sales. Right. Depending on what campaign you launched, how you launched it. These things need time. And we're talking about weeks, right? Like usually three to four, maybe four to five weeks, something like this. And what often happens now in marketing is that people start to freak out, out almost and start to question everything and then turn everything off again. Which is actually the complete opposite of like, let's learn over time. Let also let the campaigns and the platforms, they have learning algorithms which also take time, right? And then also the models also need to learn over time. So it's like almost trust the process, right? Have very clearly said, um, what do you want to achieve and where, like, what is your. Like, like CAC measurements? What is your threshold? And then actually like run the test until the end. Use the insights that you get. Um, the insights and learnings that you get to then make a better decision. For example, if the next test that you run, right.

Vishesh Duggar: You need to have some sizable spend to get quality data to make M decisions.

Peter Grafe: Exactly. And also sizable spend overtime, right?

Vishesh Duggar: Yeah. You're saying measurement isn't even the hard problem. Like you said, why has acting on the measurement been so hard?

Peter Grafe: Hard? Because the.

Vishesh Duggar: Yeah, sorry, how does, uh, the. How the decisions you recommend are easier to act on.

Peter Grafe: So measurement is hard, right? Like measurement getting right is hard by itself, but then making them actionable is even like, harder. I think it's just very. Because the models that we've built, our data science model, and these are data science outputs that don't fit into a marketing, like historical marketing, historic marketing world, right? It's just a different language that you basically speak. And the trick is to. How do we turn this insight into a language that, for example, everyone understands? And there's like two paths to this. There's AI is doing an incredible job at that, right? Like, take this insight. Explain it to me so that I understand it. Explain it to me like I'm five, right? Explain it to me like I'm a marketer, right? Like, for example, like, these are, uh, incredible applications for AI. So this is an incredible bridge. And then the other bridge is just general education of like, this is a new system and this is how we actually like start to implement it and having good systems in place on how to actually build trust and educate people so that some point they can do it themselves. And that's ultimately also the AI native service part, right? Of like, you have all the tools, you have all the insights. Now it's about going out there and also like explaining to your customers how this works. Right. I think another thing which is that which you just mentioned, this right now is there's too many dashboards. No one acts on dashboards. You know what we actually believe, and the Blue Alpha platform is also built in such a way, everything, every single agent is an mcp, right? A, uh, model context protocol from Thorpic. And that MCP you can actually load into your Claude, into your codecs, into your like favorite AI workspace. Because we truly believe that in particular Claude and Codex will be like the future workspaces of people. Right. I start my day in Claude, right? Not everyone does it right now, but you will start to feel like over time there will be such a high adoption, for example, that people will do like if they wake up and Google something, for example. Now you go into, uh, cloud and search for something within your chat bar. And what we want to achieve is that you can chat with all of the insights that we generate also within cloud, right? But then what actually stays on the platform side is almost like system of record and planning and strategy, and you want to actually be able to push back information in your workspace back into the BlueAlpha platform. So it's not just a quote unquote like normal mcp, which usually is always just to retrieve or pull data or read data. It's actually bidirectional that you can write as well, right? And the mcps that we're building is you can do your work in Claude, you put together an analysis, but now you want to find consensus with your leadership and start executing it. We actually push it to the Blue Alpha platform into the planning module, where it starts to create a proposal that says, peter, put together this proposal on how to fix Google Ads. You sit together with your management, you iterate on it and then you can click execution. And the execution is also automated because you connected to all the ad platforms and then you can start to track it. And I think it's interesting that everyone was competing about, oh, we're building the operating system for humans. I think it's actually wrong. I think you need to start thinking about how to build the operating system for agents. Because if the operating system for a human is like Claude, you know, it's more about like how do you do agent to agent communication. And that's ultimately where we see what we believe also is the future.

Vishesh Duggar: There was a time where people were talking about being an API first company and so on. And now we've reached being an MCP first company. Um, switching gears, you, your co founders are both ex Tesla data scientists. At some point you had to build a sales function. Uh, what did you get wrong the first time?

Peter Grafe: What do we. Well, I think what do we get wrong? A lot of things, you know, like I think it was a lot, of course, um, how to run, I think how to actually run an actual sales call. Right. Like what are the things, how do you qualify a lead? I think a lot of mistakes are being made is like sales is actually less talking and more listening. Like truly understanding what the pain points are of the people that you like talk to. I think that has been a big learning. Um, and then just generally sometimes you think people. And it's never closed, it's closed when it's signed, you know what I mean? And people can say whatever they want until it's not signed, it's not closed, you know. And I think mistakes. Hiring mistakes, no, of just hiring people for sales positions that weren't a good fit. I think those were, I would say like, if you like talk about painful because there's like obviously monetary value like attached to it. I do sales. Stefan is like our um, our cto, like in this product. Incredibly smart guy, you know, you should also have him on. I mean you can go like very, very technical also to like the individual AI system because um, we're obviously just scratching them like very high level. But for me the biggest learnings in sales where it's really hard work, you know, and you really need to crunch, especially in the beginning when no one knows you, you know, you need to find your wedge in. It's like how to run an actual sales call and then ultimately how, um, to build a go to market function. Right. And on go to market. This varies also by industry that everyone's in. But in marketing there's a lot of noise because everyone tells you that they know everything and they can do everything right. Because it's just that doesn't probably happen in robotics. Right. But this is just what the problem is more in marketing there's too much noise. And for someone to actually that is in the market, it's very hard to understand what is actually true and what is not true. And I think the hardest things that we had to figure out is what is that? Go to market M, uh, motion to break through the noise. And what we just found to be the most successful is there's obviously certain signals when people are like, more, uh, ready to buy, right? Whether this is funding, whether this is someone took a new job. You know, this is trial and error basically of like, figuring out what works for you. But then what we worked, what we found out worked the best for us is like leading with a lot of upfront value to actually, like build trust. Right? And that could be, for example, competitive intelligence, because most of the data is like publicly available. Or you need to like, connect to APIs and starting to like, create benchmark reports, comparison with charts and just like leading with value first of M. Helping clients understand, like, we actually know how your business works without ever having worked with you. And then starting to like, for example, ease in. And, and the Tesla, uh, background helps. Sometimes it helps, sometimes it doesn't. Sometimes people don't like Elon and then doesn't help. But, uh, usually this is also like a strong social proof sign that helps. It plays in our favor that we actually are not, you know, the term industry blind. You know, if you spend a lot of time, for example, in marketing, sometimes it's like, okay, we approach things very differently from like, first principles because marketing has not been like, fully our background. And actually marketing is also a huge data science problem at the end of the day because it's like causal inference. You know what I mean? It's like art and science. Right? But the science part is pretty big. So it's like data science. But I think those have been the hardest things. And yeah, we're still crunching it out. And I think there's also other challenges that will await us on our journey that we still need to conquer.

Vishesh Duggar: And so is this your first time selling?

Peter Grafe: Yes, I was. I did not sell before, but to quote my dad, my dad always tells me that I could sell an Eskimo bridge. No. So maybe, maybe I was. Yeah. I think also selling is, it's almost an art, you know, of

Vishesh Duggar: a lot

Peter Grafe: of conviction and like, very deep knowledge in the product that you, like, sell, right? And you need to be like, fully convinced that this is the right thing and you really need to live. Live it. That's what, that's what we do. We're living it, you know, and then I think shaping the edges is just like, again, how do you run a sales call? You know, like, what are the things, how to qualify a lead, how to follow up, uh, what Are best practices there?

Vishesh Duggar: Yeah, makes sense. You mentioned about intent signals. Is there anything specific you're using for getting those intent signals for yourself? Like what kind of tools have worked for you to identify those intent signals?

Peter Grafe: I mean we have. Yeah, I love talking about this. So we have a uh, two by two matrix in our. Which is like ranked by awareness and intent. And we use Heyreach for LinkedIn automation. We obviously run LinkedIn ads. We run Google Ads as well. Then we have a newsletter. We send outbound emails as well. We post um, every single day on LinkedIn as well. Did I miss something? We also have a physical product that we send out. We have our own creatine gummies. We designed our own creating guys because they give marketers power. So we also sent these out to high value prospects. And all of that ties back to HubSpot, right? Our CRM. And we track everything, right. So things. If you send an email, someone clicks on the email, they go to our website, they reveal themselves and we actually understand what they do on our website. And that then starts to trigger for example a ranking of scoring on like awareness and intent depending on which sites they visit to then trigger, get enrolled in like LinkedIn ads to stay like top of mind or trigger personalized outreach of. Hey, you visit um, a case study of ours and you went to our product tour. I actually know who you are. Let me write you a personalized email of like is there anything you want to jump on a call for example or lead with additional value that we've created from some of our benchmark reports for example to then get the call in. And that works well. But it's like a whole go to market system that we've built in the background.

Vishesh Duggar: And you said you have your own gummies. Yes, like, like sugar gummies?

Peter Grafe: Yeah, no, those are ah, our creatine gummies.

Vishesh Duggar: Oh, that is interesting.

Peter Grafe: You know, so we, we shipped our own creatine gummies. That's nice. Your blue boosts grow harder. So this is, this is like a word play because it's like tastes like growth and it's mmm as well. You know, um, that's a good one. And we have like, we have thousand package uh, pouches. We have a thousand pouches. Yeah, yeah.

Vishesh Duggar: For people listening in audio. You probably need to see the video.

Peter Grafe: They do need to see the video.

Vishesh Duggar: Interesting. Um, and so like you ship them to people who like as a cold outreach or people who are at point some it's.

Peter Grafe: Well first you need an address and you need to understand that there also Are in that location physical. Which is not always the case because like now with remote work, a lot of people are just remote. But, um, we, yeah. Find the addresses of people that we had usually calls with and then we send it, um, for example, to them. And we actually had deals closed because of that, you know, so the thing is, it actually does work because it's just different. And then you all of a sudden have like, for example, blue Alpha gummies in your office, you know, which is like.

Vishesh Duggar: Yeah, I like the whole. Mmm. And then the packaging itself is blue.

Peter Grafe: Yeah, it is blue.

Vishesh Duggar: How did you come up with the name Blue Alpha?

Peter Grafe: We initially wanted a different name which costs too much money because the domain was like 40k. And then alpha is the constant in the regression. And our like, data science models are ultimately a lot of like, like simulations, Monte Carlo simulations and also regressions. And then alpha also is money in finance. Right.

Vishesh Duggar: Yeah.

Peter Grafe: And then blue is just a, uh, very trustworthy and calm color. And then. Which is bluealpha. AI.

Vishesh Duggar: That's interesting.

Peter Grafe: And I like blue.

Vishesh Duggar: Blue is a nice color. I like blue too. And what about Creatine? How did you come up with that as a gummy?

Peter Grafe: So the. Well, creatine. Because first of all, it gives you strength. And then second of, there's been studies. Right. That also increases your cognitive behavior. And when companies start working with us, we help them to optimize their entire marketing mix, make better decisions and ultimately create more revenue. Right. So it was like a good supplement to actually like associate the brand with. Yeah. To push Creatine. Yeah.

Vishesh Duggar: Interesting story. Very interesting story. A lot of people come up with these merch, but I think if a merch has a story and a lot

Peter Grafe: of people have like merch like cups or T shirts. Right. I didn't. Or post. Yeah, exactly. Right. Or like linear or something like this. And I didn't want. It's very generic. So I wanted something else. And we actually saw competitors doing the same stuff that we started to do, like literally crazy, you know, like, competitors are way bigger than us. Started copying us, you know, like not even like trying to like use something else. It was literally just like copy paste. Yeah.

Vishesh Duggar: Right. We have reached what I call never get a cold email round, which is questions that will help probably listeners get to know you better.

Peter Grafe: Yeah.

Vishesh Duggar: What's been the biggest lesson so far

Peter Grafe: in the journey of building the startup as a founder? Hire the right people. People. Hiring the right people is, I think, everything. It's not everything, but it's like a big chunk of it. Because you cannot do it all alone.

Vishesh Duggar: Yeah.

Peter Grafe: And I think having an incredibly good team around you first of all, makes the entire journey so much nicer because you actually like working with people that you like admire to a certain degree as well. Right. And that you like build great stuff with, but also makes it so much faster. This is just me coming from a more like data science, um, and statistical background. Storytelling matters a lot, you know, matters like more than you actually think, especially if you want to raise money.

Vishesh Duggar: Right?

Peter Grafe: Yeah. And no, I think those are the two biggest learnings from, from my end, you know, of like storytelling and just making the right hiring decisions and working with the right people.

Vishesh Duggar: I think both of them are really important. So you did mention, and maybe one

Peter Grafe: last thing, that money doesn't solve all of the problems.

Vishesh Duggar: All the problems.

Peter Grafe: Well, because uh, the thing is you still need to hire the right people. Right. So the things, even if you raised M10 million, if you don't hire the right people will not work out. But it's a bit of like, if you raise 10 million, you also have better signal that you're like a legit company, you know what I mean? If you raise, raise money from the right VCs, this is also a strong signal. Syria attract, um, better talent.

Vishesh Duggar: Yeah, money, money at least attracts the right talent. And you did mention about a bad hire earlier in the sales. Multiple. Multiple.

Peter Grafe: But all like I wouldn't classify them as bad. Like they're not bad people, they're like good people. And the person that's responsible for these decisions is me or like is me and Stefan. Um, because we made the decision to hire these people. So it was like more a misjudgment on our end. They weren't the person that we needed in this particular situation in order to get to do the job that we wanted.

Vishesh Duggar: And so post that experience, have you changed certain things about what you look

Peter Grafe: for and how you judge on the hiring process? I think we changed things quite drastically. So we used to do normal process of first interview, second interview, take home assignments, system design, and then like get to meet the team. And problem there was just took way too long and we lost actually like talent of like having like a six week process. And time is everything, especially at the startup. Right. Like speed is our biggest advantage. And now what we do is we do one call max to only the founders to like check the vibe and then we invite people for work trials. And these work trials are paid, you know, and are compensated and then we try to get them as quickly as possible. Like, usually within a week, for example, into the office and have them work with us for, like, three days. And then during these, like, two to three days, the, uh, the plan is, okay, you come in, you get like, we onboard you for two hours of like, this is the problem scoping session of, like, this is what we need to do. Then you go, the person goes out and scopes, presents to us, and then starts to execute. And during this process, we basically check critical thinking. How do you structure your thoughts? How can you explain it? Right? Can you actually understand the problem? And then execution part of, like, how quickly do you ship, right? And then also once you execute it, can you explain what you have done? Right? And this is like, what you used to try to do with, like, take, uh, home assignments and, like, system designs and all that kind of stuff. And at the same time, you're already meeting the entire team and you're like, like, you need to work with us together during that, like, work travel part. So that has been an incredible improvement. And actually, like, helps a lot with, like, understanding how people think, um, and so on. And, and then we also hire more younger people now than what we used to. I think we, we, yeah, we just hire, uh, because. And, and, and the reason is actually somewhat simple. We've built up a lot of experience that we can actually, like, lead and guide better. And then the question is, like, what are you looking for? Are you looking for someone who has a lot of energy, a lot of creativity, and particularly with, like, new situations, like AI for example? No one has worked. Like, yes, there are some people that have worked a long time in the air, but there's, like, so much new stuff coming out. So the thing is, the type of person that you actually need is just someone that stays really on top of things, things, is aware, super hungry, and, like, soaks up this entire knowledge, right? So, and when I say younger, I mean, I'm, um, we're talking like, 20s, you know what I mean?

Vishesh Duggar: Like, yeah, way, way past that.

Peter Grafe: I mean, to be honest, I thought about it. I was like, if I'm, like, in like four years, you know, or five years, sold the company, everything's fine. But maybe looking for a job, maybe not looking for a new job. I was like, oh, I'm gonna be like, you know, like, maybe I'm, I, I, I might not even qualify anymore, you know, like, but it could be, you know, it could be. We got. I was featured in the Business Insider article about, like, hiring and a lot of hate. It was so funny. It was like oh, you don't pay the people. I was like, we literally said we pay them. And I was like, oh, Elon, Elon run company. Or like Elon infused company. And of course you exploit them for, for work and you can only hire people that, that don't have a job anymore. I was like, guys, let's all take a breeder here. You know what I mean? Like if someone is actively looking, you know, if someone's actively looking, they will take a work trial. You know what I mean? Like what does this mean? Using up the pto? You're like, you're looking to switch jobs, you know what I mean? But it was quite funny because it really upset a lot of people and I also understand it. But it's like in our situation, speed is everything. Right. And that the general uh, hiring process for a corporate is different than from a startup, a hundred percent. But also the Persona that you attract for, uh, for a corporate and the type of skills that you need are completely different from what you need for a startup. So the thing is, of course we need to adjust our type of hiring to actually like serve our needs.

Vishesh Duggar: Yeah, of course, you know, it doesn't have to work for everyone. Uh, it just has to work for the people you want to hire. Best hire you've made at Blue Alpha. What did you see that others missed?

Peter Grafe: Everyone is a great hire that we currently have. They're all incredible. No, it's true. I think everyone that currently works with us is exceptional in the area that they work and the topic and the responsibility, like the topic that they cover. So. And I wouldn't. And uh, I think it's unfair to say like one is better than the other because they're all exceptional in like their job.

Vishesh Duggar: Yeah, I think it's a bad question as well. Who asks?

Peter Grafe: I mean especially like ultimately you potentially throw someone on the bus, you know, um, yeah.

Vishesh Duggar: Worst money you've spent on, on the business so far.

Peter Grafe: The worst money that we, you know, we're actually, we're a German run company so we have very efficient capital allocation. So there's actually, I mean ultimately it's, it's, it's a, uh, hire that you spend too much money on that you unfortunately had to part ways, you know, but I think there we don't splurge on things that are unnecessary. Yeah.

Vishesh Duggar: And so being a German founder, are you a Jira shop or a linear shop or a GitHub issues shop? Shop.

Peter Grafe: We're a Linear shop.

Vishesh Duggar: Linear shop.

Peter Grafe: We're linear. Excalidraw. Um, Shop. Yes, that's where we are.

Vishesh Duggar: And one tool.

Peter Grafe: Sorry, one tool we can live without. Yeah, Claude, that's.

Vishesh Duggar: I think I've been. I need to cross this answer off now.

Peter Grafe: Yeah, exactly, exactly. I think, I think, I think. Yeah. I mean, so our stack is, you know, obviously, Claude. Everything. Notion. Right. Then we use linear and Excalidraw. Excalidraw is like, to, like. It's like almost a, um, board, you know, where you can, like. Yeah. Whiteboarding session. A whiteboard as well. We also, obviously, we love to draw on whiteboards. Um, but obviously it's not a stack that's like, uh, a.

Vishesh Duggar: Like, are you guys all fully on site or.

Peter Grafe: Yeah, yeah, this, this, this. Yeah, we actually. This was also one of the. We're only hiring on site now, so this was also one of the learnings. We used to do remote. And we still have people that work remote, but we have worked with them personally before, for example, at Tesla. So there's like, this trust that has been built and we would also, uh. Especially at this current stage, everyone needs to be on site because decisions are just being made so fast that you need to be present in the future. Me and Stefan are both, like, very remote friendly, you know, and we're also, like, very hybrid friendly. Like, the thing is, not everyone works five days a week in the office. It's just that if you have to be in, you can be in, you know, and you cannot be in if you sit somewhere else in the world or like, even in a different state.

Vishesh Duggar: Yeah, A, uh, book that changed how you think about data marketing or building.

Peter Grafe: Oh, I forgot the name of it. One second.

Vishesh Duggar: Yeah, we'll chop that bit out.

Peter Grafe: Uh,

Vishesh Duggar: um, or maybe leave it.

Peter Grafe: Building. Hard. One second. The hard thing about, like, hard things. Exactly. From.

Vishesh Duggar: Yeah. Uh, the hard thing about hard things.

Peter Grafe: The hard thing about hard things from Ben Horowitz. Exactly. There you go.

Vishesh Duggar: It's probably the easiest book to remember.

Peter Grafe: Yeah. But to be honest, I have my AI agent now, and every book that I want them, um, to read, I just feed to Klaus. And then Klaus, like, just perfectly summarizes me to every question that I have. I outsource the book. Reading.

Vishesh Duggar: Interesting. I kind of like the stories. And I think at least I learn better when I have, have, you know, contextual stories around the lesson. But, yeah, uh, AI has made things easy to learn, for sure. That was it. Where can people, uh, find you, um, and know more about you? Do you want to share anything around hiring or fundraising or any other topic for Blue Alpha that you want? No.

Peter Grafe: Well, you can find us on bluealpha. AI. My LinkedIn is Peter Grafer Graphi, how you would say it in, uh, English. We're also an X, but we're not that public on X. And we're hiring for a go to market engineer to actually perfect the system, um, even further. And yeah, thank you so much for having me on. This was, uh, this was great.

Vishesh Duggar: No, thanks for joining me today. And yeah, it was fun to chat. Likewise. Enjoy it. If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or whatever your favorite podcast app is. See you in the next episode.