Building a Shared Language: How Incrementality Brings Marketing and Finance Together

Foreign. Welcome to Modern Marketing and Measurement, the Advertising Week podcast that takes you inside the fast changing world of data driven marketing. Each episode brings you sharp insights from industry leaders as we decode analytics, attribution and performance through a future focused lens. From emerging technologies and shifting privacy rules to the next wave of measurement strategies, this is your guide to staying ahead of the curve. Today on modern marketing and measurement. I really enjoyed this conversation. We sit down with Michael Kaminski, co CEO of Recast, to learn how incrementality, experimentation and media mix modeling combined can finally bring marketing and finance onto the same page. We envision incrementality as the connective tissue these teams have been missing. How MMM and experimentation work better together than alone, and why forecasting with built in plans for learning are the new hallmarks of high performance teams. Again, great conversation. It's packed with practical, science driven insights and excellent sound bite analogies you can take to the next meeting. So grab a cup and settle in. Hello Michael. Welcome to the show. Hi Becky, thanks for having me. To get us underway, tell us about Recast. Recast is an incrementality platform that helps marketing teams measure performance and forecast into the future. One of the ways that we like to talk about it is we like to say that Recast is the connective tissue between marketing and finance teams. Coming from marketing, it would be nice if there is more connective tissue. That pendulum swings quite a bit between finance and marketing and who runs the show. So having something in between to keep them cohesive is important, right? It's a, it's a tough battle, but an important one. So we're, we're happy to be working on it. Why is it important for marketers and finance teams to share that common vocabulary around uncertainty and forecasting and risk? And how can organizations start building that shared language? Well, I think obviously it's important because it's important for like businesses to work together and have a common understanding of what the business is trying to do. Right? And I think in my experience talking to lots of marketers, lots of CMOs, right? Marketers and leadership, I often hear either sort of implicitly or explicitly about the friction between marketing teams and finance teams where the marketing team is like, we have this goal of acquiring all of these new customers or doing X, but the finance team won't give us the budget to let us go do that. And then they end up having, you know, potentially again, sniping at each other in meetings or blowing up in the boardroom or whatever. Like it can, it can yield really nasty outcomes. And so I think it's really important for organizations to work together so that finance understands, okay, what is marketing actually doing? Right. This isn't just a cost center. We're actually trying to build something here and we have goals and there's ways to think about return on investment in very solid ways. And marketing team used to help bring the finance team along so that they don't think about like, oh, marketing just wants to go buy, you know, small, spend a ton of money and work with celebrities on super bowl ads. Like, you need to come together to think about what are we actually trying to do here as a business and make that case really clear so that the business can actually work together. I think finance teams are on board with the idea of invest now to get payoff in the future, but they want to understand what are the actual mechanisms that work here that actually bring that to life. Yeah, they tend to have very different cultures and it really helps to make things less nebulous. Particular on the finance side, they have difficulty understanding that there are hard goals for marketing. What are the key questions that marketing mix modeling or MMM and incrementality testing can answer? And where do they fall short? So marketing mix modeling and incrementality testing are both ways of trying to understand incrementality. And maybe I'll, I'll explain what that word means. It's, it's sort of been growing in popularity, but not everyone has, has a shared understanding. So at least to me, the idea of incrementality just means causality. Right. When marketers use the word incrementality, they mean when we, you know, do some additional marketing activity, how much additional revenue does that drive or how much additional customers does that bring in? That's the idea of incrementality. And so the idea of causality is really baked into that. Which again is distinct from this idea of correlation. Right. A lot of other marketing measures are correlational in nature. Right. Where it's like. Or we see these two things are associated, but we aren't really sure if one caused the other. Right. Did putting up the billboard cause those additional sales or were those sales going to happen anyway? So media mix modeling and incrementality are really about trying to understand that causal relationship between marketing on one side and then some business outcome on the other. And so these are really great tools for that can help marketing teams work with the finance teams because they can start to get to the things that finance cares about. Right. Finance cares about, oh, if we spend an extra million dollars on this TV campaign, How much additional revenue are we going to get? That should make sense to a finance person. They can plug that into their spreadsheet, Right. That's going to fit really well with their way of thinking about planning and profit and loss. And so media mix modeling and incrementality testing can be used to get at those numbers, which then can be shared with the finance team in a, in a way that they can understand, they could start to grapple with again. That's distinct from something like brand awareness. Right. If a marketer says, oh, brand awareness went up by five points. The finance team says, well, where do I plug that into my spreadsheet? Right. How do I think about that? Or, you know, we got a bajillion impressions, right. The finance team says, okay, that sounds great, but like, how does that flow into, into profit and loss and cash flow? And so the idea behind incrementality testing and media mix modeling is really to try to make that connection a lot stronger. So, so that the finance team can really understand what is being driven by marketing. Yeah, it gives them like a shared vocabulary that they can both use back and forth. So everyone's on the same team still. Exactly. How can brands effectively blend then that. Mmm. And incrementality to create more reliable forecast and guide budget decisions. This is a really good question and there's, there's lots of different ways to think about it. This, this is again, this is another sort of hot topic in marketing science and marketing measurement. Right now. I think it's really important to think about incrementality testing and experimentation and media mix modeling as sort of two important tools that work better together. And so one thing to keep in mind about media mix modeling is that it's a very complex method with lots of assumptions. And when you're doing this sort of statistics and econometrics, anytime you have an assumption that sort of is like a way that the model can go wrong, right. And if you make a bad assumption, unfortunately, it can really like, it can invalidate all of the results of the media mix model. And because there's often hundreds or thousands of assumptions in these different models, any one of those assumptions, if it's wrong, can actually invalidate everything else. And so it can be a very powerful tool, but also extremely error prone. The nice thing about experimentation is that experiments generally have many fewer assumptions. Right. If you do a randomized controlled experiment like you see in healthcare, if you think about clinical trials, the beauty of them is that you don't need very fancy statistics to analyze those results, right? They're very straightforward to analyze. They don't require a lot of assumptions. And so the idea is that you can take this very powerful but very error prone method of media mix modeling and then use it alongside experimentation, which is less, much less error prone, in order to sort of confirm or invalidate the results. And so if your media mix modeling, if your media mix model says one thing, you can check and go run an experiment and say, does that say the same thing or does it say something different? Right. And if it says the same thing, that's good. That sort of indicates that those two approaches, very different approaches, have a shared understanding of reality. That's like confirming. If they say something different, then it's like, well, probably the media mixed model is wrong. There's probably some bad assumption in there and we need to go back to the drawing board and figure out what might be wrong in order to actually have faith that we can use this going forward. So that's sort of like the practical way of thinking about using them together in terms of thinking about, okay, what, how are these error prone in different ways and how can we use different types of evidence to come to a, a better understanding of true causality and a good understanding of the world? The other way, I think is again going back to this idea of working with finance. Finance teams over the years have often been burned by media mix models, right? Some big brand hires a consulting firm to build a media mix model for them. They produce some results, but you know, different people were putting their thumb on the scale. Person A like really wanted TV to look good. And so, you know, TV looks good magically in the model. And the finance team looks at these results and they're like, look, this can't possibly be right. You know, you said TV drives X, but we spent this amount of TV last year and that would imply some number that is just not possible. And one of the nice things about combining these techniques together, media mix modeling and incrementality testing, is that the incrementality testing again has many fewer assumptions. So finance teams should be able to buy into the methodology much easier than they buy into the methodology of the M and M. And so if you can convince them about incrementality testing and experiments, again, just be like, look, it's just like, you know, a clinical trial, right? You've heard of that, you understand how that works. And the results are consistent with the media mix model. That starts to build the case that, okay, these things, you know, make sense together and we can really use them to drive the Business forward. Right. So you're saying that like if good queries give good data, but then also your incrementality idea is almost like a peer review in this case. That is a great way to think about it, right? It's a way of like checking the results and making sure that they line up experimentally. So I think that we talk a lot about taking the scientific method and then really applying that seriously to marketing. And that's a great way to think about it, of being like, look, we're going to publish these results, but then other people should go check them, right? Running their own experiments and make sure that the results log. Why not? That's really great. With that in mind, then what are the practical steps to help marketing teams convert those campaign metrics into finance aligned KPIs that resonate with the CFOs and boards, how do you then translate that information to your CFOs and boards so that they are on, well on board with it? So there's a couple of tactics that I think are helpful and that I think are sort of critical to doing this well. So one is, is being able to forecast, right? So again, these like marketing metrics that sort of float out in the universe, you know, the number of impressions that we drove, right. It's hard to know what to do with that. And so I think it's, it's really the responsibility of the marketing team to figure out a way to convert those metrics of what they're doing in the marketing world and link that to revenue. Right. So make some claim about if we do X, we should drive this amount of revenue so sort of a falsifiable claim into the future about if we do X then Y will happen. This forecasting exercise is hard. This is really difficult to do. But I think it is really important for the marketing team on its own to start to get this understanding, make sure that they have like a complete model in their mind of what are we doing and how is that actually driving the business on the other side. And if you can do this, it builds that shared language with finance and starts to give finance insight into. Okay, this is, you know, how you think the world works. And now we can start to go test it. And so this idea of making forecasts gives you something that you can go test and you can go check, right? We're going to make the forecast today about what's going to happen next quarter. Does it actually come true or not? And I'm going to get to another point in, in terms of how to think about this and how not to shoot yourself in the foot. But that's a really good exercise to start, to start doing this, to build this shared understanding of, of can we make a forecast? Can we see if that actually comes true? If I say we're going to get a bajillion impressions and that's going to downstream yield this amount of revenue, if you can call your shot with the finance team, they're going to start to really trust you and start to say, okay, this is a model of the world that we can rely on for actually managing and operating the business. The second piece, and this is critically important, is you really need to think hard about uncertainty, right? A mistake that I see in all kinds of organizations and not just marketing teams is this sort of across the board is that people do this forecasting exercise in like an Excel spreadsheet and they plug in all of these assumptions and they get a number out on the other side and it's like, great, we're going to do, you know, $75.3645 million next quarter and everyone's like, amazing, great. And then they're two weeks in and someone changes one assumption in that spreadsheet and all of a sudden the number is like totally different. It's like, oh no, now we're only doing $60.44732 million next quarter and everyone panics. And the problem here is that this is an exercise in false precision, right? The number was never 75 point whatever, it was never 60 point whatever. All of those assumptions in the Excel spreadsheet actually have some baked in underlying uncertainty in them. And when you're making these forecasts into the future, you need to make sure that you're communicating what that uncertainty is and how the uncertainty in the inputs, the number of impressions you're going to get, the configuration conversion rate, the ROI on your marketing channels, whatever those numbers are, how that yields uncertainty and what you expect is going to happen in the future. I think a lot of people have this idea that like, oh, you know, the leadership team really expects certainty and if I go in and I'm uncertain about something, they're going to eat me alive. And I actually haven't found that to be true either in my own work or my consulting or with pretty much anyone that I talk to. Leadership teams, finance teams, they understand the idea of uncertainty, they understand the idea of risk. They want to understand what the risk associated with some plan is, what those trade offs are, what's the high case, what's the low Case if you as a marketing team or as a marketing leader can talk through that thoughtfully and still make a recommendation, hey, look, there's risk with this plan. Here's what the downside risk is. But I still think that we should do the campaign. I think they're really going to appreciate that and they're going to understand that. And you're not setting yourself up for failure three months or six months later when you don't actually know the forecast. Right. You can talk smartly about, hey, what's the downside case? What's the upside case? And then that is going to build a lot of credibility for you as you sort of work through doing this in a repeated fashion this quarter and then next quarter and then the following quarter. So that was sort of a long answer, but really it's these two ideas of, one, tie what you're doing to forecasts that link to business outcomes. And then two, be very clear about uncertainty and risk and how that impacts the forecast and the potential business outcomes based on what you're planning to do. Yeah, that makes sense. I mean, marketing 101 is. Marketing has to equal sales, and that's how you know it's successful. But on the other hand, there's going to be uncertainty. And I think, in my experience, anyway, if you can convey your agility and ability to pivot when something's moving, not the direction you want it to go, and pivot into a way where you can still get some successful numbers out of it, you, your management team or your CFO team is going to be behind you the whole way. Becky, that's such a great point. Like, I wish that I would have made this point. So this point, I, and I want to double down on it. This point of, like, having the escape valve, I think is also really important. Like, if you can come in to the, to the finance team and talk about the forecast and talk about the uncertainty, that's great. And then if you could say, look, if we don't see X results by Y date, we're going to make this change, we're going to do this adjustment, then that's even better. Right? That sort of reduces the risk. And again, it helps people feel like they're being brought along on the plan of what are we going to do if something doesn't work out? How do we salvage it? How do we make the best out of this? This is like a really amazing tactic, again, to make this successful and make the partnership super collaborative and super productive. I will. If I'm pitching something immediately up Front being like, these are my sandbag numbers and I almost always hit numbers, but these are my sandbag numbers. And if we're hitting the sandbag numbers, then we pivot to this. Yeah, excellent. Want to lead in marketing and the age of machines? Then plug in and join us on AI Machine Made Marketing. AI Machine Made Marketing examines how artificial intelligence is transforming how brands and marketers engage audiences, drive growth, and future proof their strategies. Each episode we bring you front row access to the people building the algorithms, the brands deploying them, and the insights no one else has. Talking about whether you're a cmo, a creative director, a growth marketer, or just someone who knows that if you're not thinking about AI, you're already behind. This show is for you. Subscribe to AI Machine Made Marketing Now. On your favorite podcast platform. Can you share a case study or example where combining MMM with incrementality meaningfully improved the forecasting accuracy or investment decisions? Oh, this is a great question. So I don't know that I have like a specific example, but we have, we see versions of this constantly. One of the things that we think about businesses doing, and this actually came from, from some of our customers is they, they talk about this idea of a learning agenda that they have for the quarter or for the half, like, what are the most important things that we want to go learn? And one of the ways that this particular brand thinks about it is they say, okay, well, what we're going to do is we're going to look at our media mix model and we're going to look at two things. So one, where is the most uncertainty in the estimates that are coming out of the media mix model? And those are sort of our list of channels that we potentially prioritize for certain experiments. And then two, when we look at our forecast for the next six months or the next year, what we want to do is we want to understand what are the drivers of that uncertainty to. Right? And that's sort of the com. That's sort of the, the multiplication of your largest channels that you are most uncertain about. And so if you multiply those numbers together, you get this sort of prioritized list of what are the drivers of that uncertainty. And then basically what you can do is you can say, okay, based on, on these different lists that we have, we can come up and potentially other like, strategy items, right? You might say, like, oh, we want to test this new channel that we've never been in before for strategic reasons. You pull all that together and you start to develop a Roadmap of what do you want to learn over the next three, six or 12 months. And then you can sort of think about what are the techniques that you have at your disposal for trying to learn. Maybe something like tv. Right? We can't do a geographic lift test because of the way that that's bought. And so what we're going to do is we're going to triple our spin for one quarter in this channel and we're going to see how much sales go up and that's going to be our way of evaluating that. For this other channel, we're going to run geographic lift tests for this other channel. We have the ability to, you know, run ghost ad individual level lift tests via the platform. And so we're going to schedule that in and we build the schedule, we build a learning roadmap and then we bring the whole company along as we get readouts on all of these different experiments and then we feed that information back into the media mix model. And so as we get more certainty about all of these different channels and their true performance, we actually can narrow the forecasted result of where we're going to end the year or whatever that number is that we really care about. And so this like cycle of learning agenda planning, different experiments and then feeding that back into the mmm I think is a really powerful way of thinking about operating these different tools in a system that again also drives profit at the end of the day. Right. It's all connected back to these financial results that the business cares about. I love this. I love this. So it's really interesting. Marketing and advertising in particular are often so dynamic and ever changing and there's new channels all the time, you know, but then also sometimes because of the pendulum, you have to go back to the old channels. Like events are really big and hot, but then they cycle out and then social media is really hot and then people get tired of that and cycle out and then something new will come along, like let's start using AI chatbots or something like that. And then those cycle out. You know, comment if you want to have this new recipe, gets tired really fast and then you go and do something else. And so you're saying that you can forecast these and also give yourself a safety net. Yeah, that's exactly right. And I think again, really astute point about marketing always changing. So one of the things that is tough about marketing science is that marketing science is very different from like a physical science like physics or chemistry, where we can learn things about the universe or the world that we live in that are pretty much eternal, right? Like the gravitational constant is eternal from our perspective at least. Different ways that chemicals combine together, these are just like fundamental rules of the universe. They are unchanging. If someone discovers it 4,000 years ago, it has the same effect today. Marketing is not like that at all, right? And everything is constantly changing. And whether it's as you, as you note, there's like new channels that open up because people engage with the world in new ways. Whether it's, you know, for any individual brand, their particular market might be changing, competitors might come in and launch new products, they might change their price, you might launch a new product, right? All of these things change the way that your brand exists in the world, changes the way that your marketing works, changes the way that customers or potential customers engage with your brand and buy your products. And so that means that something that you learned three years ago, two years ago, one year ago, right? Might not still be true anymore. And so it's not. We don't exist in a world where we can like really sort of build on the learnings over very long time periods, because learnings from Marketing 100 years ago just don't apply today. Instead, we have to think about how do we build systems of learning, right, where we can continuously learn about what works and what doesn't and continuously adapt and be responsive to that changing environment. And so it's a really interesting distinction to think about as we think about, okay, what tools can we borrow from the hard sciences and what do we have to sort of do a little bit differently because we're in this ever changing world where we don't, we don't have these like hard laws that we get to fall back on. That's a great analogy. So looking ahead, what innovations or methodological shifts do you think will most influence how marketers plan and measure impact over the next few years? This is a really good question. So I think there's been a lot of really exciting developments in causal inference and statistical modeling over the last couple of years. And this is actually independent of sort of large language models, AI. I'll actually talk about that in a second. But in terms of causal inference, there's been a bunch of new methods that are very powerful, allow you to do causal inference in like smaller sample sizes and sort of in environments where you don't necessarily have these like large scale randomized studies. So those are really powerful. Those are again, sort of what power a lot of modern incrementality testing. And then two, we have a lot of interesting New developments around the ability to fit very large complex Bayesian models. And so if you've heard of like Google's release of Meridian or some of the open source packages, a lot of these are powered by under the hood, Markov chain Monte Carlo methods for estimating very complex statistical models. And only in the last like 10 or 15 years have we had the algorithms to be able to fit such large and complex models effectively and like within, you know, from hours to days. Right. So they're not super fast, but they're a lot faster than they used to be. So those two things combined I think are really bringing to the forefront the sort of new era of incrementality and causality. And so we see a lot of these open source tools being released, we see a lot of new entrants into the market of measurement that are taking advantage of those new tools. And I think they've penetrated somewhat, especially in like the startup scene. Like a lot of like new companies, you know, are really taking to these methods and thinking hard about the right ways to use them. I think one of the big advantages of these methods is that they can be a lot faster than what was possible in the past. You can now run lots more incrementality studies than you could in the past. You can now refresh your media mix model much quicker than you could in the past. And so that adds this dimension of speed and agility that just wasn't possible before. Outside of that, we have again, AI LLMs. I think you need to be really careful when you think about the application of those tools to marketing science. LLMs, again, everyone knows they're very good at giving wrong answers confidently. And so if you're going to use those tools in the field of marketing science, you need to be very, very careful about exactly how you are using them and how you are avoiding the wrong answer confidently problem, which is a really serious problem. And so I think there are ways to think about incorporating, incorporating those sorts of tools into marketing science. But the way that I think about it at least is you need to have a really solid base of like scientific method, causal inference and then have the LLMs sitting on top of that. So the LLMs aren't like a replacement for the science and the causal inference. That sort of still needs to happen. You can build automated systems that do that, but it requires a lot of thought and a lot of rigor to make sure that that happens correctly. And the LLMs can think about, okay, how do I sit on top? How does the LLM sit on top? Of how does the LLM sit about? Think about helping you build that learning agenda based on the underlying substrate that you have a lot of confidence in. Like, that's sort of a good way to, I think, think about the application of LLMs to, to. To marketing science. So if I were to bring in Recast and have you come in and you would probably suggest these ideas and help future proof my forecasting for marketing. That's exactly right. And that's what we spend a lot of our time doing. One of the things that we talk about a lot internally and with our customers at every guest is that it's not that difficult to like generate insights from a media mix model one time. But as we just talked about, those insights go out of date. And so what's really important is actually building this system, right? The system for continuous learning. Right? The. You know, we sometimes talk about this incrementality system of how do you do these things together over time, Produce a forecast, track your pacing against the forecast, plan your experiments, feed the experimental results back in that whole process and system. That's actually the hard work, right? And that's what actually drives a lot of value. But it's not easy, right? It requires some amount of organizational change and requires there's no silver bullet, right? It's not like, oh, we're going to give you the answer and then you're going to have the perfect marketing budget and then we can all go home. We're going to have to do this over and over and over again over time. And the trick is building that system, bringing along the rest of the organization and then just trying to focus on getting a little bit better every day. Fantastic. If listeners wanted to find out more. About Recast, where should they go? So the best place to engage with me is on LinkedIn. I'm Michael Kaminsky. On LinkedIn, I post a lot about marketing science. Less about Recast specifically, but I love talking about causal inference, marketing science. If you want to check out Recast specifically, you can go to get recommended recast.com ton of information there. We've got a lot of interesting information on marketing science, causal inference. Great. Thank you so much for coming by, Becky. Thanks for having me. This was a blast. Thanks for listening to Modern marketing and Measurement. If you enjoyed this podcast, please consider subscribing and sharing with your colleagues and friends. For more podcasts like this one, be sure to check out Advertising Week's growing roster of podcasts for the advertising, marketing and technology industries, including AW360, Agency Alchemy Retail Media, Unboxed and AI Machine Made Marketing. All available now at www.advertisingweek.com. Podcast. Sam.