The more things change, the more they stay the same.
For decades, marketers have been trying to achieve the “holy grail” or marketing perfection: reaching the right person, with the right message, at the right time.
And throughout that time, generation after generation of marketers have come away with the same realization.
What is simple in theory is maddeningly difficult in practice and near-impossible at scale.
But then things changed (or so we thought) as the advertising ecosystem began a slow, unyielding march online.
After all, digital advertising is superior to its predecessors.
It is built on mountains of user data, dominates every aspect of our life, and provides every advertiser with the level of granular control previous generations could only dream about.
At the core of this digital advertising revolution is “behavioral advertising” – the technique by which advertising platforms use thousands (or more) data points to micro-target users, who can then be served hyper-targeted, hyper-relevant messages at the “moments that matter,” to borrow a phrase from Google.
The holy grail is finally within reach.
Behavioral advertising – in just the last few years – has been credited with:
- Shifting the outcome of a presidential election.
- Building dozens of “unicorns.”
- Helping thousands more companies survive a once-in-a-generation pandemic.
It has been touted as an advertising revolution – the thing that will forever change how advertising and marketing are viewed.
But does it really work?
Or is it all just a fiction?
In his forthcoming book, “Subprime Attention Crisis,” Tim Hwang (the former policy director at Google) argues that the entire digital ad business is built on a series of frauds, which if exposed could bring the digital economy to its knees.
Wired recently reviewed Hwang’s book – and it is this review on which I’ll base this article.
As I understand it, Hwang’s argument goes something like this (also, this is an exciting opportunity for me to put that logic course from college to good use):
- Digital advertising doesn’t deliver the value it claims (more specifically: digital advertising “doesn’t work” and is “worthless”), due to a horrible combination of bad data and waste, fraud, and abuse.
- Nevertheless, brands continue to invest hundreds of billions into digital advertising each year, so ad prices continue to increase despite no corresponding increase in the overall value of the good being sold (the digital ads).
- The profits generated by digital advertising fuel the growth of the digital economy, from content/news sites to cutting-edge research into artificial intelligence (AI), machine learning, and green energy.
- The digital economy plays a central role in the health of the overall economy.
- If brands realize (1), they’ll cease online advertising.
- If (5), then by (2), digital ad prices will fall.
- If (6), then the profits referenced (3) will no longer exist.
- Thus, by (3) and (4), a cessation of digital advertising could have wide-ranging and catastrophic impacts to the broader global economy.
Hwang’s argument seems reasonable and logical.
But (as with everything) things are not exactly as they appear.
Let’s take each premise and examine it – then return to the overall argument.
Does Digital Advertising Deliver the Value It Claims?
One of the central tenants of Hwang’s argument is the notion that programmatic display ads today are akin to the subprime mortgages of 2000s, which played a central role in the 2008-09 Great Recession.
While there is ample evidence to suggest that digital advertising isn’t as effective as platforms like Google, Facebook, Twitter, Microsoft, and Amazon claim and that attribution (the process of apportioning “credit” for positive outcomes to contributing channels) is still murky at best, it simply does not follow that these ads are “worthless.”
But perhaps this is due to a fundamental misunderstanding about the value of an ad, to begin with.
Hwang claims that the value of an ad is a user’s “putative attention” (which he claims is as worthless as a toxic security of 2007) – I’d argue that definition is incomplete.
When you buy a digital ad, you’re buying the probability that the ad contributes to a positive outcome that has some value (X) to your organization.
In short: you’re buying expected economic value (“EEV”).
Given that each organization is unique, the positive outcomes desired are unique, as are their economic values, risk tolerances, and the associated probability density functions.
Are All Ads the Same?
Before we go further, let’s take a brief digression to expound on one critical point: all digital ads are not programmatic display ads.
Total spend on programmatic ads is about $106 billion in 2019, according to Statista.
But this is less than one-third of total digital ad spend amounting to $333.25 billion per eMarketer.
What else is there?
There are also:
- Search ads.
- Social ads.
- Video ads.
- Private marketplace deals.
- Shopping ads (Amazon, Walmart, Kroger, etc.).
Each one comes with its own challenges, targeting and bidding processes, set of control levers, and level of transparency/accountability.
If that’s too technical, the simplified version is this: ads (digital, traditional, etc.) are worth different amounts to different organizations at different times.
So, What’s the Value of an Ad?
And herein lies a critical difference between ad units and subprime mortgages – the value of the underlying asset in each case is fundamentally different.
On one hand, a subprime mortgage is a loan, with a set value (the principal), an expected return (interest rate), and a third-party assessment of risk (a rating).
Wired, in their articles, notes as much – and suggests that a more apt comparison would be comparing MBS to the stocks of Google and Facebook.
Unfortunately, this comparison also falls short – if for no other reason than stock prices rarely (if ever) reflect corporate fundamentals in the way the MBS prices reflect the value of their underlying assets (the mortgages that comprise the MBS).
Where Hwang is correct is that the entire digital ad ecosystem is supported by a vast, opaque technological infrastructure, which enables thousands of real-time auctions to determine what ads are shown to which users on what pages, all in near-real-time.
This does (in some ways) resemble how algorithmic trading has come to dominate segments of the financial sector, but it is important to note that just because something is algorithmic, automated, or opaque does necessarily mean that it’s bad.
All it means is that additional scrutiny is warranted to ensure the automation performs as expected.
This aspect of the digital ecosystem deserves further scrutiny – especially as platforms aggressively move to hinder transparency into these auctions and their outcomes (including what queries search ads are shown for, what websites ads Facebook Audience Network ads are served on, etc.). Many of us working in and around the digital ad ecosystem have called on platforms to be more, not less, transparent – calls that are finally being echoed in both the House and the Senate, as well as by regulators around the globe.
But What About the Waste, Fraud & Abuse (WFA)?
The second similarity between subprime mortgages + digital ads Hwang points out is the prevalence of waste, fraud, and abuse – by platforms, middle-men, and other third-parties (including ad agencies).
Examples cited include click fraud, charges for non-viewable impressions, obscene mark-ups (exceeding 50%) by agencies; ads purchased in bulk by agencies via a negotiated fee, then re-sold to clients at an exorbitant mark-up and more.
To those unfamiliar with the world of advertising, these things sound awful, with eerie parallels to “The Big Short” coming to mind.
But while each of these is (in my opinion) a suboptimal business practice, none of it is new.
Click fraud is the issue de jour – and Hwang is correct that it is a serious problem that must be addressed if our industry is to move forward.
As of this writing, there are few acceptable, global standards for identifying and resolving click fraud (and even when new regulations or protocols are rolled out, bad actors like click farms seem to uncover new ways to circumvent them).
Platforms, including Google and Facebook, have made efforts to address the issue on their end, including reimbursing advertisers for fraudulent clicks – but they’ve tended to come up short, especially for smaller advertisers.
Click verification technology has improved markedly over the past few years. However, far too few agencies offer it by default and far too few brands know to demand it.
There is no doubt that click fraud/digital ad fraud is a problem, but not as large as Hwang might believe.
Click fraud impacts are estimated at ~$6.5B to ~$19B – or ~2% to ~6% of the total global digital advertising ecosystem, according to eMarketer.
What about non-viewed impressions?
Well, this too is a new version of an old problem. Consider that TV ads are bought + sold on a CPM, using estimated viewership (i.e., Nielsen ratings).
These rankings use calculated panel data to approximate how many people are watching a given show at a given time. But that’s all they are. Estimates. Extrapolated from sample data taken at a given point in time.
Even if we grant the estimation’s general accuracy, how many of us continue to sit on the couch during commercial breaks, eyes glued to the screen?
Is this not a comparable situation to ads being paid for, but never seen?
Similarly, newspapers price ads by readership, but these too are estimates – and even if your customer opens the paper containing your ad, what is the probability that the individual actually looks at Page A8 and reads your specific ad?
The same holds true of billboards and radio ads and airport takeovers and stadium sponsorships and all the rest. To put it simply: non-viewed impressions are a reality of every form of advertising. The only difference is that digital ads have the potential to be better – they just haven’t lived up to it yet.
Likewise, many traditional ad units come with various forms of “shadow” pricing for agencies or media buyers – from commissions paid to the agency to a gross/net spread to various kickbacks (surcommissions, over-riders, backpools, volume deals, discounts, etc.).
This practice is, and has been, known for years.
And as with the above, I find it obscene.
It’s important to note, however, that this issue tends to be concentrated with holding companies and large agencies, who see it as an alternative revenue source at a time when their other fees are being squeezed.
Agency Holding Companies have engaged in various forms of media arbitrage for decades.
This isn’t new.
The ad units have changed, the nature of the arbitrage has changed (from “upfronts” back in the TV days to the arbitrage of programmatic ad space we see today) but the fundamentals are the same.
I’ve argued for years that these parts of the advertising business are in dire need of reform.
But they aren’t new.
Brands have caught on to many of these tactics – including CMOs of prominent brands like Proctor & Gamble – who have publicly shared what they’ve learned.
While these shady practices are being exposed and (hopefully) phased out, it is important to remember that these deals account for only a small fraction of all programmatic ad buys, which themselves account for a small fraction of total media spend.
The more things change, the more they stay the same.
What About the Bad Data?
Hwang is once again correct that behaviorally-targeted ads aren’t as precise as many vendors + platforms would have advertisers believe.
That isn’t surprising to anyone in the digital advertising world, or anyone who’s received a poorly-targeted ad.
This is especially true given the mountains of data being fed into platform bidding algorithms, both from:
- Platform-enabled trackers (cookies, pixels, tags, app data and the like that capture online activity).
- Third-parties and other advertisers (data pass back from CRM systems + augmentation with other offline data sets, etc.).
This data is of varying quality, accuracy, and relevance – and mixing it all together can have disastrous consequences to the overall accuracy of the data.
As an analogy, consider what would happen if you were to mix perfume with odorous excrement – the result is would still be quite unpleasant.
The same thing happens with great data is augmented with poor data.
And when that poor data enters other ecosystems, it can pollute those as well.
Making things more complex is the fact that this process is often automated, with many machine learning algorithms defaulting trusting the validity of the data being inputted and assuming that the view is complete (i.e., that all relevant data is included within the set).
Both of those things are almost never true – so most of the data that flows through these algorithms is somewhat flawed.
As a practical consequence of this, many algorithms are quite good at flagging when a user’s purchase intent rises, but dreadful at pinpointing when it falls.
This leads to awkward situations where a user makes a relevant purchase (i.e., a kettlebell or a microwave) on a website but is continually followed by ads for more of the same item for days, weeks, or months after the purchase.
While this is a specific example, it’s illustrative of the broader issue:
Ad dollars being wasted (directly or indirectly) on users who simply are no longer relevant to the brand in question and are not going to buy.
The cynic in me believes this is intentional. After all, having larger “in-market” audiences expands auction participation, and in so doing, would seem to make platforms more money (more bidders = higher bids = more money for ad platforms).
The technologist in me believes that (i) fusing massive data sets in near-real time, each of which is generated by sources of varying quality and (ii) then relying on those data sets to understand user intent and behavior is an insanely difficult problem to solve in a vacuum – and virtually impossible at scale and without perfect information.
The reality is probably somewhere in-between, with some platforms trying their damnedest to crack the puzzle, and others content to sit idly by and collect checks from clients/advertisers.
This is a long-winded way of saying: yes, some of the data used to target programmatic ads is varying degrees of rotten – but that’s not necessarily a problem.
All data deteriorates – there’s no escaping this – the question is how effective are platforms at removing rotten data before it pollutes more of the data set.
To me, the more relevant question is: does that really cause as big a problem as Hwang suggests?
After all, in Hwang’s argument, the flawed data is akin to bad coordinates being passed to a Tesla rocket – just a small error is sufficient to result in a big boom.
This seems fundamentally misguided.
To illustrate why, consider an alternative analogy: betting on a game of roulette.
For those of you unfamiliar with the game, placing a $1 straight up bet (a bet on the ball dropping on a given number – all of which pay out 35:1) in a U.S. game of roulette (38 spaces – 18 red, 18 black, 2 green) comes with an expected value of -$0.053 (35*(1/38)+(-1*(37/38)).
In the short term, you could win a bunch or lose a bunch in a row, but over the long term, you’ll likely lose about a nickel each time you bet a dollar on a specific number.
Ironically, the rate of hitting on a straight-up bet is about the same as an average online conversion rate (~2.6%), making this analogy all the more appropriate.
So, where does data come in?
Data – even data of mediocre quality – can help remove spaces from the roulette wheel by filtering out non-winners.
Even removing a single space cuts the expected value loss in half.
Remove just a quarter of the board (9 spaces), and you’ve shifted from losing $0.053 per play to gaining $0.25 on every spin.
Bad data is just less reliable in removing bad spaces (and maybe removes some winners, too).
Great data is more efficient and effective at removing bad spaces from an audience.
Will digital advertising ever be able to perfectly narrow our roulette wheel down to just the single space where the ball lands?
Regardless of what any marketer tells you, there’s always a bit of luck involved in a positive outcome.
But as long as your data is sufficiently good to remove even a small fraction of non-winners, you’ll shift the expected value of your advertising to your favor.
What About the House of Cards Created by Digital Advertising Profits?
The next part of Hwang’s argument (and this part is doing a lot of work) is that when brands realize what’s happening with digital advertising, they’ll pull their spends, jump-starting a chain reaction that will obliterate the bottom lines of not just the ad tech sector, but of the broader digital economy.
Newspapers will collapse without the lifeblood that is ad revenue.
Start-ups and creators will fail without the ability to monetize their content via banner ads.
Pandemonium will reign supreme.
In my view, this is both laughably and demonstrably false.
For another, most online platforms, content creators, and start-ups have multiple ways to monetize an audience – one of which is ads.
- Subscription services (newspapers have embraced this recently).
- Freemium models.
- Direct sponsorships/private deals (i.e., influencer agreements, private marketplace deals, direct deals).
- Affiliate links.
- Direct ecommerce.
And if you find all that unpersuasive, there’s always history: whenever a platform rises to prominence (either in the real world or the virtual world), businesses have been willing to pay a premium for the opportunity to reach those individuals.
Today, those platforms are online – and more than ever, people are turning to the internet for everything from research and product discovery to shopping, schools, employment, and more.
According to recent stats:
- The average U.S. adult spends about 451 minutes (nearly a full work day) with digital media.
- The average smartphone user touches his/her phone more than 2,600 times each day.
- 87% of purchases start with an internet search.
The centrality of the internet to our lives is increasing with each passing day – and (as they always have), brands continue to vie for a chance to capture just a small fraction of our attention while we’re there.
But What If Hwang Is Correct?
Even if Hwang is correct, and even if the entire $106 billion programmatic industry goes to zero overnight, the loss won’t cause a great recession or anything close to it.
The stocks of ad tech giants like Google and Facebook will surely feel the pinch in the short-term, but long-term, they’ll do what they’ve always done: find ways to leverage their data + audiences to generate revenue for their shareholders.
While $106 billion seems like a big number (and it’s objectively a lot of money), it’s worth mentioning that banks booked 5-10x that number per year in subprime mortgages from 2005-2007 (and plenty before then), with the size of the MBS-fueled market rising to nearly $10 trillion – about 40% of the entire global consumer credit market – between MBS, CDOs, Synthetic CDOs, CDO-Squared, and related securities by 2007.
The programmatic industry is a drop in the proverbial bucket relative to the consumer credit market – and one with far less centrality to the global financial system.
So, Where Does This Leave Us?
Returning to his initial argument, it’s pretty clear that Hwang is wrong about (1), (2), and (3) – and when those premises fail, the conclusion no longer logically follows.
Good news, crisis averted!
But there’s more to the story than just a guy with a flawed thesis.
While I think Hwang’s assertion that the digital advertising market is akin to the subprime housing market is fundamentally misguided + deeply wrong, there are nuggets of truth embedded within his argument that deserve to be examined, debated, and discussed.
Not all digital advertising data is accurate, reliable, or useful.
Many platforms are excessively opaque and automated, hindering an advertiser’s ability to control their spend and where their ads are served.
Further, programmatic advertising (and digital advertising more generally) is not immune to the flaws of traditional advertising – and even has some flaws of entirely its own (like click fraud).
It is no secret to any of us who work in digital advertising that transparency is needed – both from platforms and advertisers.
To many of the readers of Search Engine Journal, I’m preaching to the choir on this point – but it bears repeating.
I sincerely hope that Mr. Hwang’s book brings more attention to these matters – and spurs more brands to scrutinize their digital media plans, purchasing, and efficacy reporting.
But does any of this entail that digital ad technology is the next internet bubble?
Digital advertising is not perfect.
On the whole, it is improving (mostly) over time – largely due to major ad tech platforms building better, more accurate, and more complete data sets (which comes with its own set of issues, which Kirk Williams discussed here).
There have been a lot of bumps in the road to improving advertising, and there will be plenty more.
My only hope is that, as you run over them, don’t mistake a pothole for the Grand Canyon, as Mr. Hwang has done in his recent book.