The Kaizen Framework: How to turn marketing data into growth

Transcript

Edward:
We're back with another episode of the Marketing Intelligence Show, and today we're talking about the Kaizen Framework and how to turn marketing data into growth. And I'm joined by a very special guest, Zeke Camusio, CEO, at Data Speaks. Welcome to the show, Zeke. And I guess a good starting point for this discussion is the concept of Kaizen itself. What is Kaizen?

Zeke Camusio:
It's great to be here. And Kaizen is a Japanese word that means continuous improvement. Personally, I've always used that to analyze performance in every aspect of my life and figure out how to improve. But later in life, I came across the word and thought it was perfect. And later, it just became the framework through which we built our entire business.

Edward:
I think it was in Toyota that it became popular in terms of applying the Kaizen concept to business, management, production, and efficiency. And very interesting to think about that concept from the perspective of marketing data, which is what we're going to get into further today. And you have a long history of working with marketing, growth, and data, so you've seen a lot over the years. How did you land on applying the Kaizen framework to marketing? And more specifically, how do we, as marketers, work with data?

Zeke Camusio:
Yeah, of course. I've been doing this kind of work running analytics for over a hundred different brands for the last 20 years. And something that became evident in the early days is that many people were looking at data as if it was looking through the rearview mirror to see, how did I do last month, last week? But they weren't proactive in terms of what is this data telling us. What action do we need to take based on our data? And the ones doing that were measuring how effective those decisions were. I found an opportunity there to give them a framework, something, a tool that they can use, to not only look at data to see the past performance but also to inform future decisions, make those decisions, take action, and then be able to measure how effective that action was.

Edward:
I like it. And we spoke about Kaizen being associated with Toyota so well that we have a car analogy here as well. I don't know if that was a coincidence or planned. But yeah, we do use data to typically look at what has happened in the past. As you said, the rearview mirror, but how can you use data to look forward through the windshield to see where you're going, and what does that tell us about the direction and the decisions you need to take as a marketing team?

Let's dig into this framework on how to actually turn that raw marketing data into business growth, which is what we are all about here at Supermetrics. And that process you've developed is built on four points:
Collecting all your marketing data in one place.
Identifying high-priority insights.
Formulating a hypothesis, and then the fourth is testing it, learning from the experiment, and repeating.
Which closes the loop as you then have more data to collect. Let's take these one at a time. Let's start with the first point. How do marketing teams go about collecting marketing data in one place?

Zeke Camusio:
It'll be useful if I give an overview of the framework and then go into that main point. I always say that data is only as good as the action you take based on it. You have raw data and need to take action from that, but you need to transform that data into insight. And the reason for that is, let's say you have 10 different data sources times, 10 metrics, times 10 dimensions, so that's 10,000 data points. Well, a thousand data points that you have daily. It's impossible to keep track of all that. And that's the first step. You need to take all that raw data, and you need to have in front of you the key things that you need to focus on today so you can take action. Once you take action, it's really important to set up that as an experiment. You formulate a hypothesis. You set up an experiment and then see whether the hypothesis is correct and how accurate it is. That's an overview of the framework itself.

Now, to answer your question about the first step. Yes, the average brand has 12 different data sources, and it's impossible to log in to 12 different dashboards every day and try to figure out... even metrics are named differently from one data source to the next one. It's not possible to have a data-driven culture if your data is fragmented across multiple silos. The first step, what you guys do really well, is getting that data out into a data warehouse, a Google sheet, or somewhere where you can normalize it so all your data from different sources is in the same format. You can then access it more easily and build BI systems and machine learning AI systems on top of that and use different data sources combined to get deeper insights.

Edward:
Absolutely. And is data collection the responsibility of the marketing team, or is it the responsibility of a data team if you have one?

Zeke Camusio:
That's a really good question. I think it's a collaboration of data teams supporting not only marketing teams but product teams, leadership, understanding what each team needs, and being able to provide that in a way that is easy and accessible. The main roadblock that I found doing this kind of work is that most people are not data people. If you expect a CEO to write a SQL query, the data might be there, but it's not going to happen. You really need to understand what are the things that each team needs to have in front of them every day to be able to do their job the best way possible.

Edward:
Absolutely. And another follow-up point here before we move on to the second point of the framework. At Supermetrics, we have a saying which is, crap in, crap out, for want of a better expression. How do you make sure that the data coming in is high quality? Because if it's not, then obviously, that is going to impact and mess up the rest of the entire process.

Zeke Camusio:
Of course. No, I totally agree with that. There are a few layers to that. The first layer is making sure that you track everything you want, so you need a good tracking plan. You can't expect, for example, to have Google Analytics out of the box and do everything you want it to do. You need to customize user ID, custom dimensions, custom events, and content grouping, so you need to do a lot of work to track every event with every property. That's a question of understanding as a business — what's important to you, and how do we track that? That makes sure that you start the process the right way by collecting as much data as possible and increasing the quality of that data. Then there's a cleanup process as the data gets into your data warehouse or wherever you collect it. Things don't always track the way they should be tracking.

Hopefully, you caught some of those things early on, but often you don't, so you have to clean up the data. And then, you need to apply a semantic layer on top of the data to help business users understand deeper insights. For example, you can categorize your products by pants or sweaters, low-cost vs. premium products. You can apply many of those layers, and many of those texts in the semantic layer. Then when the data's brought into a BI tool, machine learning AI, that data is available. There's a lot of truth to what you said. The models will only be as good as the data you put into them.

Edward:
Absolutely. And we could probably spend the entire episode talking about data quality itself and how to maintain those standards, as you said. But for time's sake, let's move on. Some super valuable points there, but let's get into the second point. Once you've centralized your marketing data, wherever that happens to be, as you said, how do you then actually identify high-priority insights?

Zeke Camusio:
That's a great question. As a data strategist and business strategist, you always give the metaphor of a pyramid. You have at the very top; you have the most important metrics for your business. Usually, that's profit and or revenue. And then what our system does is it reverse engineers how you get to that revenue level or that profit level. For example, profit is all your revenue minus all your expenses. Then if you take the revenue branch, that is, the orders you get times the average order value. Then to get more orders, how many of those are going to come from Amazon? How many from Shopify? For Shopify sales to increase, how much traffic do you need to get from Facebook Ads or Google Ads? And then for each of those, how many clicks you need, I'll click, cost per click. We could think about it in hierarchies where you have the ultimate business objective at the very top, and then you have metrics at the bottom. At the lowest layer, you have how many likes you get on Facebook.

Naturally, our system prioritizes the metrics at the top more highly than the ones at the bottom. That's because there's a chain of events that leads to the ultimate business objective. Then there's another layer to that, which is the delta for each metric. You may have a metric that's halfway through the pyramid, but if you're looking at a monthly delta month-over-month change, you could see, for example, that your average order value may be increased 20%, or your traffic from email decreased by 50%. And that's another component that is built into our algorithm. It's not only how high or low it's on the pyramid but also how much it changed.

Those two factors get combined into the algorithm to weigh every metric that comes in. And at the end of the day, it's really a filtering process. You have an overwhelming amount of data. You need to know the top five to 10 things that I should be paying attention to today. That's a really important process. I think that sometimes what prevents organizations from being more data-driven is that they have too much data and are unable to figure out the key data points and what they should be paying attention to today.

Edward:

I think that's really important that more data does not mean a better situation, but really it's about finding out what is relevant from all the data you have. And just to ask a question on a different example to the one you spoke about there. Let's say you're a B2B company with a longer sales cycle. It's a more complex buying journey with many online and offline touchpoints. How do you derive those insights from your data?

Zeke Camusio:
We rely heavily on multiple regression, a statistical technique that allows you to see the correlation between an input and an output. Input is anything you can do, such as sending more emails, increasing advertising spend, et cetera. And an outcome is what do you expect to get out of that? Like more sales, more leads. Ultimately, we use multiple regression because most modern digital brands aren't just doing one thing at a time. They're doing multiple ad campaigns and marketing initiatives. You need to be able to understand the weight that each one of them has had on the outcome that you got. And once you make the regression and understand the correlation between you doing that and the outcome you're getting and the impact, what's the ratio?

For example, if you increase your spending on Facebook by $1, how will that affect your sales to your leads? Will it increase leads by... for example, a ratio could be, you're paying $250 for a lead. At the end of the day, the outcome is the same. You need to understand whether it's ecommerce or B2B. You need to assign a value to that lead. And you could do that by saying, what's the average deal size for every lead? How many leads do I need to get to get a deal? And then, you can divide and understand how much every lead is worth to you. And one of the things our system will do, for example, is be able to weigh and score every lead because they're not all created equal. Based on various factors that we put into a machine learning model, it will tell you the value of some leads. Some could be $400, and some could be 50, depending on a variety of factors.

Edward:
And so from here, then you got the data in, you got the insights, how do you as a marketer then take insight and turn that into a hypothesis?

Zeke Camusio:
What I was saying about the impact is really the beginning of the process. Let's take, for example, the ecommerce case that I just mentioned, where for every dollar that you... the prediction is that for every dollar you increase ad spend on Facebook, you increase sales on your website by 250. That's based on historical data. It looks at the trends for every marketing activity you've done, plus any external factors such as promotions or economic conditions, it looks at all that. It tells you that that's what you can expect moving forward based on historical data. That's the hypothesis, it's not only just a correlation, but it actually measures. If you increase this by this much, then you can expect this to increase or decrease by this much. That's a hypothesis, and it's important to understand that just because something behaved a certain way historically doesn't necessarily mean that that will hold true in the future.

There are a lot of reasons for that. For example, it could be diminishing returns in an advertising campaign. Economic factors, more brands bidding for your keywords. There are many different things. That's why the hypothesis isn't true until you actually put it to the test. We found that being able to turn hypotheses into experiments easily is by far the greatest asset a brand can have. Where you could say, for example, in Data Speaks, you get that prediction, you could say it is an experiment, and then it will track data, new data coming in, and you'll compare it against the hypothesis. Maybe it's not 250. It's 230.

What we do is on a weekly basis, we keep adjusting that prediction so it gets more and more accurate. Normally the first time a system makes a prediction, it's between 80 and 85% accurate. We found that for three or four weeks, usually, you're at 95% accuracy. You're never going to get to a hundred percent, but being 95% confident that if you do this, this will happen. If you do this, this will happen; it gives you so much power as somebody making decisions for your brand.

Edward:
Absolutely. And finally, then, talk us through the final step, which is what you just started alluding to in terms of experimentation. How do you experiment and learn from those results, which generates more data, closing the loop and taking us back to step one?

Zeke Camusio:
There are two pieces to the experiment. There's the input and the output. In this case, we said the input is we're going to increase spending for Facebook Ads by this much, and then this is what will happen to our sales. There needs to be a rationale behind it. You can't just say, I'm going to double my conversion rate, and if I ask you how you have no idea. There has to be something you're going to do to make that happen.

That should be part of the hypothesis that if we do this, then this will happen. And you have to be very, very specific. I can't say if I improve the user experience on a page. You have to say exactly what you're going to do and why you believe that will happen. At that point, that's the experiment. You set up those two metrics, the activity that will lead to the outcome, and decide with which frequency you want to check that. Depending on the data you have, it could be every other day. It could be weekly. It could be twice a month. I like doing it weekly. That usually gives you enough time to collect the data you need to be able to adjust your forecast.

Edward:
Absolutely. From here, could you give a real-life example of the end-to-end Kaizen framework in action?

Zeke Camusio:
A couple of months ago, we started working with a brand that was growing significantly in the previous few years and suddenly started seeing their sales decline, and they couldn't figure out why. I connected Data Speaks to all their data, had the first meeting with them, and showed them what was happening. And what was happening is they were investing less in Facebook Ads because their ROAS was getting lower. Now, it was still profitable. In this case, the breakeven point, they needed their ROAS to be at 1.5 to break even. It had decreased from 2.5 to 2.2, but it was still profitable. What happened is, at first, they were just targeting their... they were doing retargeting to their customers, then they started doing lookalikes. And then as they started getting... those audiences are very targeted, but they're very small. As they increased their reach, they got a lower bang for their buck but were still profitable.

They decided to start shrinking that based on their recommendation from the advertising agency to keep ROAS as high as possible. But in reality, let's say you're making... for every dollar you invest at the core, you're making a $1.50, here you're making, let's say, two here, a $1.50, a $1.25. When you start making 95 cents, that's not profitable anymore. But if you're making $1.20, that's a profitable segment you should go after. We formulated the hypothesis that if they invested at different budget levels, they were going to get different returns. And we took into account the curve of diminishing returns. And not only did we tell them that they needed to invest in that channel, but we broke that down by every dimension available. And we showed them what calls to action they needed to use, what types of creative, and whether they needed to be a video or a carousel. And we essentially broke it down for them. What campaigns, what audiences, and then it was a matter of tracking that performance, understanding how accurate the prediction was, and making adjustments moving forward.

Edward:
And so from here, any final tips for our listeners on getting more systematic when working with marketing data?

Zeke Camusio:
I'll say this. You have enough data as it is. You don't need more data. The main thing you need is to start working on creating a data-driven culture. As the leader of your organization, you need to start by setting a time of the week when you discuss performance. And everybody in your team and all your teams need to understand what success looks like for the channels they manage. There's no better gift you can give someone, for example, managing your TikTok Ads, than understanding what success looks like. What they can report that can make you happy. Because that way, you define the targets and you define the narrative. You don't leave it up to each team to come back to you each week with this is what I think is important. The priorities and the focus for each channel need to come from the top down.

Edward:
Absolutely. That's super good advice. And there were many really good takeaways within this episode in the framework. But the piece you said on data is only as good as the action you take on it's a great, great point. A lot to take away from that. Our CMO always says that it doesn't matter until the rubber hits the road, which is when you actually start doing stuff. You can talk and get all the insights you want, but you need to take action to ensure everything will happen. Super, super good. Zeke, thank you so much for joining us today on the podcast.

Zeke Camusio:
Thank you, Edward, for having us. And thank you, Supermetrics, for making this possible because if we couldn't have data in real-time to apply to all these models and all these advanced AI and machine learning and data science, what we do couldn't happen, so I appreciate you guys.

Edward:
Our pleasure.