The Drum columnist Samuel Scott recently fell down a rabbit hole and into the world of econometrics with effectiveness expert Les Binet. Here he explains what marketers need to know as digital attribution degrades.
Les Binet, group head of effectiveness at Adam&EveDDB
What is better – information that is cheap and wrong or information that is expensive and accurate? Soon, marketers may have to decide.
source: The Drum | Digital Attribution Is Dead! Les Binet Tells Us Why Marketers Need Econometrics In 2023
Some months ago at my day job as head of marketing at the IT mapping software company Faddom, we saw a decline in organic search engine traffic. I could not figure out why. There was no evidence of a Google penalty, it wasn’t an SEO issue.
So I decided to look into some statistical correlations. On a hunch, I made a list of many potential variables that might have affected the traffic – no matter how far-fetched – and plotted their changes against the changes in organic traffic and the numbers of Google searches and clicks for our brand name over the same several months.
One theory was that a decrease in our Google Ads spend might be a secret search engine ranking factor. Another was that spending less on Google Ads resulted in fewer people seeing our brand name, becoming interested in us and then searching for the company. But we found that there was an 11% correlation between Google Ads spend and people searching for our brand name and clicking to our website. There was a 6% correlation between Google Ads spend and total organic website traffic. So those were clearly not the issues.
Instead, there was a 96% correlation between the number of cold sales emails that our business development team would send out and the number of Google searches for our brand name and resulting website clicks. Put simply, we inferred that people would receive an email, wonder who Faddom is, search Google for the name and visit our website.
Now here’s the issue. Google Analytics logged those visits as organic search engine traffic because they did indeed come from the search engine’s unpaid results. But in such specific cases, the cold emails – not SEO – should get the credit. The decline in organic traffic had almost surely come from the team temporarily sending fewer emails.
What can readers learn from my day job headaches?
First, marketers should remember that so-called ’outbound’ and ’interruptive’ marcom can be extremely effective – no matter what nonsense HubSpot has told the industry for the past 15 years to sell its own ’inbound’ marketing software. Second, I remembered the story as yet another example of how analytics dashboards, digital attribution and the entire online world can be misleading at best or completely wrong at worst.
In fact, it will likely be one of the biggest problems that marketers will face in 2023 as we see the planned death of third-party cookies, 43% of people now using adblockers that also stop scripts such as Google Analytics from running and the iOS 14 update that stopped ad tracking on Apple devices.
Econometrics – also known as marketing mix modeling (MMM) – might be a solution. After all, the attribution-based online marketing world that many have known for the past two decades is rapidly disappearing.
The problems with digital attribution
Les Binet, group head of effectiveness at Adam&EveDDB, told me that attribution modeling began around 1900 with direct response ads in print media that had coupons. Different printing presses in the same city could run ads with different coupons to see which ones worked better.
Starting in the early 2000s, the nascent world of online advertising quickly became addicted to attribution with Cocaine Bear levels of enthusiasm. For example, at AdExchanger’s industry preview this month, Lending Tree senior vice-president of growth marketing Joshua Palau said that “all media should be performance media.”
An entire new generation of ’digital’ marketers now thinks only about what is stupidly called ’performance advertising’ because it is deceptively simple. You put an ad on Twitter. You see how many people click to visit your website and buy. You attribute those purchases to Twitter. Easy.
But it’s actually not that easy. Even before the ongoing death of ad tracking today, attribution modeling on its own has always been deeply flawed.
“If you say this ad generates a million in sales, the true answer could be anything between [zero] to a million,” Binet, whose original training was as an econometrician, told me. “It looks very scientific, it looks very precise, and it’s extremely unreliable.”
As an example, here is a slide that Binet gave me on last-touch attribution.
Here are some of the reasons why attribution is so unreliable.
The Fallacy of Immediacy
Marketers often assume that an effective ad will convince someone to buy or become a lead immediately. But many ad-driven purchases occur long after the advertisement appeared – and especially long after the ability to track the sale with digital attribution has disappeared.
Just remember one of Binet’s classic charts from his famed work with Peter Field on The Long and Short of It.
At Faddom, we recently spent a portion of our ad spend on Reddit. After two months, we received fewer customer leads than expected. A common assumption would be that the campaign results were poor. But what if hundreds or thousands of people saw the ads and made a mental note to check us out in a year because our industry has long sales cycles?
With only digital attribution, it is impossible to know. And it’d also be impossible to know the original sources of those hypothetical purchases in 2024.
The Fallacy of Last-Touch Attribution
Binet likened this to a store measuring the number of people who enter through each door and how much they buy. He gave an example where the west door supposedly resulted in 25% of sales.
“It’s clear that it’s not just the door that generates the sales,” he said. “If you shut the west door, [digital attribution modeling] would say that you immediately lose a quarter of your business. But that’s not true. What would happen is that they’d walk around to the south door, the north door or the east door. If you’ve got a healthy business and people really want to come in, they’ll find a way.“
The Fallacy of First-Touch Attribution
Many companies assign sales and leads to the supposed first touch out of a desire to have a simple way to show revenues, expenses and overall returns from each activity. But it also happens to be completely inaccurate.
“The idea that one channel can be given ‘credit’ for a given lead or sale is nearly always nonsense,” Binet said. “Each sale is usually the combined result of multiple channels working together, often over a period of months or years. Instead, the question marketers need to ask is: What is the incremental effect of each channel? If I dial spend on this activity up or down, how much will my sales rise or fall?”
He added: “First-touch attribution is just as bogus as last-touch attribution. For a start, digital data rarely goes back far enough to identify the first exposure. Digital data trails usually last days or weeks, but advertising effects can last for years. Think about all those TV ads you remember from childhood. A sale is rarely caused by the first exposure, or the last one, but the combination of all previous exposures.”
The Fallacy of Multi-Touch Attribution
This often shows only the method of a purchase, not what convinced a person to buy. If I have already decided to buy a product, then I might, say, search Google or Facebook for the item.
But it would not necessarily mean that Google or Facebook influenced my decision to purchase it in the first place – even though digital attribution would make it seem so. To use some popular modern buzzwords, attribution modeling often identifies the channels where demand fulfillment occurred but not where demand creation happened.
Binet likened this to someone deciding to buy something and then traveling along subway lines and streets to get to the store. Attribution often focuses on the subway lines and streets.
“The attribution modeling bogusly attributes the sales to the few factors along the customer journey at the last bit of sale, and it ignores the many factors that led up to it,” Binet said.
The Fallacy of Technology
’Digital’ is not an advertising strategy or tactic. It is a type of technology based on binary code. And digital attribution is biased in favor of channels that use digital attribution technology.
Say a person sees your company’s booth at a conference but does not approach. Or they watch a YouTube video but do not click. Six months later, they remember you and decide to find you and make a purchase. Digital attribution would show nothing because there would be no trail.
“What we’ve always known is that the direct attribution method is flawed,” Binet said. “It doesn’t take much thought to realize that it’s wrong.”
One of the biggest problems in advertising today is an overreliance on short-term direct response and an under-reliance on long-term brand advertising – particularly in the B2B world.
“Attribution modeling overestimates the ROI from direct response communications and underestimates the ROI from brand communications,” Binet said. “If you just follow the attribution data, you end up just doing short-term stuff. You never build a brand, you don’t grow the customer base, you don’t grow the base level of demand. And it’s a recipe for disaster in business.”
But another big problem is that much of the attribution on which online direct response is based is wrong.
Digital attribution is inaccurate
In late 2022, Chris Walker, chief executive at the Boston marketing agency Refine Labs, posted on LinkedIn a comparison of what customers said versus what software-based attribution stated.
“Attribution software inaccurately overweights search (paid and organic) and direct traffic,” he wrote. “Not because those channels drove the results, but because that’s the path buyers take when they’re ready to buy.”
Everyone today is obsessed with being so-called ’data-driven.’ But a lot of data is simply bad information. Good research, common sense and gut instinct are often more accurate.
Eric Stockton, senior vice-president of demand generation at cloud desktop provider Evolve IP, perfectly summarized the situation late last year on LinkedIn: “In an ironic twist of marketing fate, the channels that aren’t the easiest to measure are often the best contributors to pipeline and revenue … correlation of buyer behavior [is better than] direct attribution by channel.”
In January 2023, Paul Arpikari, chief commercial officer at econometrics platform Sellforte, posted on LinkedIn that branded search and “performance” channels in general are overestimated in sales attributions. Online video is underestimated due to not having clicks.
Binet said that one person at a conference told him this: “Finally, we’re getting these digital nerds to understand why what they’ve been doing has been wrong all these years.”
So, if econometrics may be a good replacement for digital attribution, what exactly is it?
How to use econometrics in marketing
Back to Binet’s history. With the birth of radio, cinema and TV broadcast media, using methods of attribution similar to those in print ads was impossible.
So advertisers started to use econometric modeling, which is doing advanced statistical analysis of sets of mass aggregate data. Basically, it is creating a machine based on a bunch of historical numbers or industry benchmarks that will then see the relationships and calculate long-term correlations to show the effects of changing one or more variables.
In the story at the beginning of this column, I told how I used a very basic form of statistical correlation to discover the effect of changing the number of cold sales emails on organic search engine traffic. Marketers who use proper econometrics can go further and see the results of raising or lowering prices, changing ad budgets on one channel or another, opening stores on days with different weather and more.
Essentially, econometrics models take every single thing that might affect sales – from advertising campaigns to the weather to pricing to overall economic conditions – and narrows down the factors from hundreds to the dozen that are the most important.
Then, the model creates an equation that describes the relationship between those factors and sales. Marketers can then run the equation to see that if they do X, sales will change to Y. Governments use econometrics to measure things such as the effects of tax rate changes on GDP. Marketers can use it to determine the best marketing mix – because marketing, of course, covers a lot more than advertising and communications.
“Attribution – quickly and cheaply – will give you an answer that is precise and wrong. Econometrics – slowly, laboriously and expensively – will give you an answer that is right,” Binet said. “Attribution modeling will tell you that things that are unprofitable actually made money. It can be incredibly misleading. And it will tell you that things that really are profitable aren’t.”
The argument against econometrics
No theory or model is without critics – especially in marketing. For econometrics, one prominent skeptic is Byron Sharp, director of the Ehrenberg-Bass Institute for Marketing Science in Australia.
“The people proposing econometrics as a solution [to the problems of attribution modeling] all seem to be people who sell econometrics,” he told me. “There is a marketing law: ‘wherever there is demand there will be supply.’ The supply doesn’t have to work, it just has to convince the buyer that it works … and marketers are not known for their mathematical knowledge, so [they] are easily fooled.”
He added: “Using econometrics as a solution to the problems of attribution modeling is like cutting off a finger to distract yourself from having a sore foot.”
Sharp gave me a 2018 paper that he co-wrote for the International Journal of Market Research with John Dawes, Rachel Kennedy and Kesten Green. I have made it available for download in PDF format here.
From the abstract: “The contribution of regression analysis (econometrics) to advertising and media decision-making is questioned and found wanting. Econometrics cannot be expected to estimate valid and reliable forecasting models unless it is based on extensive experimental data on important variables, across varied conditions.”
But if there will ever be a widespread push to adopt econometrics modeling in marketing, Sharp will likely not be the only opponent.
One basic rule in business is to maximize revenue and minimize expenses. As a result, it is extremely difficult to get companies to pay more for something that they have always gotten for cheap. For 20 years, marketers have had access to free attribution-based analytics data through platforms such as Google Analytics. Now, try telling your boss that you want to increase your analytics spend from zero to five figures or more next year to build an econometrics model.
Second, most econometrics models need at least a couple of years of historical data on hundreds of variables. Large, established brands certainly have that. But small businesses and new high-tech startups, for example, do not.
Third, it can take months to build such a model. Try telling your boss that they will not see the results of the massive increase in analytics spend for half a year or more.
The solution might be to ask your companies if they want information that is cheap and wrong or expensive and accurate – especially when they want to see the true results of long-term brand advertising campaigns. But I am still somewhat skeptical. The perfect metaphor for the popularity of the last-touch attribution fallacy is when many companies always throw parties to celebrate sales teams closing deals but barely acknowledge the work of marketing departments.