Marketing is about people
Marketing is defined as all customer-facing functions of the enterprise. Decisions about brands are made by people; acquisition, retention, and advocacy are all human choices. Despite speculation that autonomous AI agents will be the customers of the future, the current reality is that people drive growth.
However, modern marketing is increasingly mediated by technology. Analysts and marketers interact with data, processes, and colleagues through screens, often reducing customers to digital avatars. This trend risks losing the human element in marketing decision-making.
Marketing analytics has followed suit, becoming technical and quantitative. This has enabled precision and accountability, allowing marketing to better communicate with finance, justifying and explaining marketing’s contributions. This is a good thing.
Marketing analytics is the de facto quality department of the go-to-market function. The analytics team—or the analytical resources in the marketing team—are ultimately responsible for identifying data problems, revealing invalid inferences, and uncovering inefficient programs.
However, quality-focused analytics team must beware of disembodiment. Our data represent people, and people are very complex. Our representations of them and their actions are crude indicators of their latent properties. Losing empathy is a road to poor quality, and poor quality leads to long term brand decline and de-growth.
What the Toyota Production System teaches us
In the 1950 and 60s, Detroit dominated the U.S. auto market. Their cars were big and beautiful. The factories of the big three automakers were high-tech—management science was born in Detroit, drafting after the mass production successes of WWII. However, they soon learned that data, technology, and automation do not necessarily mean quality. In fact, layers of abstraction can hide defects. By the mid-1960s, Detroit was shipping lemons, and buyers were quietly getting fed up.
Enter the Japanese importers, led by Toyota. Taking advantage of labor arbitrage, Toyota delivered less expensive cars. They also filled an unmet niche need for fuel efficiency. However, their value proposition didn’t stop there: Buyers soon discovered that their cars were also better built, had fewer defects, lasted longer, and consequently, kept their value over time.
Toyota did not accomplish this via technology. Toyota’s approach emphasized quality through human-centric principles:
- Andon cord: Stop processes when defects are found
- Hansei: Learn from mistakes
- Just-in-Time: Request or provide only when needed
- Kaizen: Empower the workforce
- Genchi Genbutsu: Direct observation of problems
- Nemawashi: Open information sharing
- Genba: Visibility on the factory floor
Quality improvement in marketing analytics requires a similar focus on people and process transparency.
Five Principles for Quality in Marketing Analytics
These five principles have emerged over the past decade at the Marketbridge team. They are best practices, proven through countless hours of focusing on delivering the best quality. They share many of the same values of the Toyota Production System:
Truth Seeking (Hansei)
The word “analysis” comes to use from Greek: Ana (untie) and Leuin (knot). Said differently, working to find all of the holes and loops in something complex, and getting to truth. For marketing, this means analysts should challenge strategies, not support them. There are many ways that this value can be practiced in real life. Some common use cases include:
- True up CPA: Demand capture channels’ last touch CPAs are usually much too low. Truth seekers can decompose lower funnel channels’ effects into accretive, distribution tax, and duplicative, and then report on CPAs taking each of these effects into account.
- Seek Outliers: The most common explanation for an unexpected result is a data issue. For example, if a tactic shows an extremely low CPA—say, two standard deviations below the mean—in almost all cases there will be some error in data collection. Of course, sometimes these are real results, but a good rule of thumb is that 9 out of 10 two sigma outliers aren’t real (unsurprisingly.)
- Eliminate Double Counting: It is rare that the marketing contribution attained from summing reports from individual channel owners adds up to less than total revenue, but very common that it is more. This is understandable; managers want to claim as much credit as possible. It is up to truth-seeking analysts to ensure that totals foot and sum. This will cause hurt feelings temporarily, but this is the cost of truth seeking, and in the long run, trust will be established.
Humanization / Embodiment
Customers are humans; we use data to represent them. A high-quality team should flip their default view, and consider what each customer receives, not the aggregate marketing mix going out the door. Humanization of prospects and customers is a consistent challenge for analysts who are almost always a few levels of abstraction away from people. However, technology also provides us with tools to humanize—if we choose and use them carefully.
- Center Data on People: A Longitudinal Human Record (LHR) aggregates all known and probabilistic data about prospects and customers, enabling empathy at scale. It also powers multi-touch attribution, MMM, and UMM, along with Customer 360 dashboards and reports.
- Go Small: The typical data science workbook flow is to chain data transformation steps together to get to an end result—typically a data frame or a model. However, this approach can miss detail. The best “data detectives” examine individual customer records and follow their journyieys. For example, following a journey from pre-sale to sale to customer—with all of the touchpoints in between—can unearth insight or quality issues that would otherwise be missed. When these findings are scaled up, they can yield big results.
- Build Audiences from Real People and Events: Too much abstraction can quickly yield muddled pictures. While advertising technology is amazingly sophisticated, real signal can be hard to find. Fortunately, addresses are still real, and so are purchasing events. Building audiences from addresses and events is grounding, and yields better results than models built upon models.
Transparency (Nemawashi, Genba)
Reproducibility is a term we borrow from scientific research. When a researcher reports a finding, it is important that other researchers can come back to the work, and at a minimum, follow the steps from raw data to results. Generally, transparency drives quality because more people can see “inside the box” and find problems.
Technology is often seen as impenetrable, but this does not have to be the case. Choiceful adoption of transparent tools and methods can make data simpler, not more complex.
- Use Version Control Tools like Git and Github: Code (R, Python, SQL) and documentation (markdown, readmes) are not just stored and available, but are change logged. In other words, the code base—a manifestation of an organization’s IP—can we seen evolving over time, and nothing will be lost in anyone’s desktop folder.
- Use Workbooks and Notebooks: Workbooks (for example, Databricks) and Notebooks (for example, RMarkdown and Quarto) are visual mash-ups of code, narrative, and results outputs. They facilitate inspection and intuitive understanding.
- Adopt a Medallion Architecture: A medallion architecture is a loose framework that acknowledges three states for analytical data. First, unstructured “dirty” data sits at the Bronze level. Second, taxonomized and QA’s data that is used commonly sits at the Silver level. Finally, use case-ready data sits in scrubbed data frames. supports both formal and innovative use cases, with ad hoc analysis (Bronze), centralized clean and governed data (Silver), and use case-ready data frames (Gold). The key is that all data end up in the data lake—even one offs (in the Bronze layer.)
Ownership (Kaizen, Andon)
There is a temptation to think that technology, data, and automation require less human ownership—indeed, this is the primary idea behind technology driving productivity. Toyota realized that people owning technology drove quality, and that hasn’t changed. In a marketing team, this means that everyone should understand data, have access to it, and be able to perform analysis.
- Own (at least partially) Martech: Marketers and marketing analytics are deeply dependent on Martech. Concretely, most data problems originate in source systems, and, by corollary, are easier to fix there than to band aid later. The people who use the data generated by these systems should at least have a large seat at the table for Martech implementation and customization—and should ideally own the systems altogether.
- Advanced COE: Analytics centers of excellence (COEs) should avoid the guild mentality, and continually focus on only the most advanced use cases—yielding simpler analysis to the marketers themselves. In other words, they should be an innovation lab, not a walled off set of protected jobs. Marketers owning their own data and analytics is a good thing.
- Celebrate QA: Data mistakes are scary. There is a human tendency to want to hide problems, particularly when they were your fault. World class organizations know that admitting failure and then correcting it should actually be celebrated. This is the basis of Toyota’s Andon Cord. The best marketing analytics teams have targets of “bugs found”—and give shout outs to those who find them (even if they created them.)
Probabilistic Communication
Because we are working with people, we can’t know everything about them. This is one of the things that makes marketing fun. When technology really started taking off in the early 2000s, some thought we would start knowing everything about audiences and customers. If anything, the opposite has happened. There is so much noise in the system that we know less, not more. However, leaders want precision—whether in terms of ROAS (return on advertising spend) or CPA (cost per acquisition.) It is up to analytics leaders to not given them exact answers—but rather to train them that everything in marketing is about probability and confidence, and to help them make probabilistic decisions.
- Always Include Error Bars: Error bars show stakeholders that the mean value, while the most likely result, is not precise. They can then make decisions that account for the chance that a result is higher or lower than reported.
- Forecasts that Get More Uncertain: Weather forecasts end at 10 days, but generally don’t include a range around a chance of rain or temperature. Marketing forecasts should not make this mistake; confidence definitionally falls further into the future, and confidence bands should follow suit.
- Beware Extrapolation: Diminishing response curves—which are typical in marketing—are sometimes fit with a small range of stimulus (x) data. The curves look sharp all the way out on the right, but in truth we have no information to predict those points. So, if we are recommending spending a lot more, we have to be honest about what is likely to happen. Flipped on its head, marketers should seek to add variation to spend levels to “soft test” on the outer parts of curves.
Quality Pays
Most readers will nod their heads to the points made above, but might wonder if they are worth doing. Companies, after all, care about profit—and focusing on quality might sound like an expensive nice-to-have. Well, Ford adopted its “Quality is Job One” tagline after they got their clock cleaned by Japan, and if you went back in time and asked their executives, they would have told you that not focusing on quality probably cost them hundreds of billions of dollars over several decades.
But what about marketing? Marketbridge has been collecting data on quality and precision among its clients since 1997. Specifically, we have been interested in its impact on four key metrics: customer lifetime value (CLV); net CPA; long-run growth; and the percentage of go-to-market dollars that are “working” vs. “non-working.” In all four cases, quality-focused organizations have achieved superior results.
- CLV is particularly affected; on an industry-by-industry basis, CLV in quality-focused marketing organizations averages about 10% higher—mainly result of a scientifically-determined, less sugary acquisition diet
- High-quality digital targeting yields around 20% better 6-month retention than cookie-based, spray-and-pray approaches
- Organizations that adopt high-quality approaches have better working / non-working dollar ratios than those that do not, a result of better efficiency and less waste that can be translated into higher media spend
- Overall, high-quality marketing organizations grow at healthier long-run rates than those that are more reactive
A focus on quality isn’t sexy or flashy—it is a cultural shift, and a long-term commitment. However, it might just be the highest ROI marketing mix decision an organization can make.
Wondering how to ensure your organization is focusing on quality in marketing and measurement? Reach out!
