Article

The Seven Deadly Ad Account Sins

Advertising Analytics Data Testing

October 3, 2023

This past month has been jam-packed with conferences, strategy sessions, pitch meetings & audits. Across all of those, one of the most common questions I’ve heard are these: “Why aren’t our [Meta/Google/Microsoft/LinkedIn] accounts performing like they used to? What are we doing wrong? How can we fix it?”

This article is dedicated to (trying to) answer that question. Before we dive in, I want to caveat everything that follows with this: while I see hundreds of ad accounts a year across dozens of industries that are spending anywhere from $1,000 to $1,000,000+, absolutes do not exist in the digital space. You may be doing things I advise against, or things that break every conceivable rule, and getting fantastic results. If so, amazing. Keep doing it until it doesn’t work, then come back here and give these ideas a try. I’m never going to tell someone that something which is working for them is wrong. At the end of the day, we’re all working toward the same goal: to help the organizations we own, work with or work for perform better.

With that out of the way, let’s get to it. The Seven Deadly Ad Account Sins:

Sin #1: Overmanagement

There was a time in the not-too-distant past when daily (or even hourly) changes to ad accounts was considered the gold-standard of management. Much has changed since that time. Automated bidding has improved by orders of magnitude, while adoption has skyrocketed. Traditional levers in ad accounts (match types, interests, etc.) have been removed.

In light of these changes, I believe that the best managers/agencies should shift to an “eyes on, hands off” approach to day-to-day management. Here’s what that looks like, and why it works:

Daily Eyes On Accounts = we monitor every account daily for material deviations from forecast, as well as anomalies in performance data.
Automated Alerts & Diagnostics = this is a free plug for Optmyzr and Adalysis, both of which have built wonderful alert systems for digital ad platforms that if configured correctly, will dramatically reduce the time required to identify an issue, so more time can be spent addressing it.
Batched Changes = all non-critical changes to an account are saved and batched, then uploaded weekly. This includes new copy/creative, keywords, etc. If it isn’t materially impacting ad account performance, there’s not a compelling reason to risk re-setting learning.
Critical Interventions = despite all of our best intentions and efforts, sometimes things go haywire. And when they do, you must be able to intervene immediately and address the issue. This occurs if there are things that are materially impacting performance, or we believe could cause a significant issue for the business, or if there are external circumstances that merit such a change (such as a product going out of stock, or a system failure, etc.)

This works for one simple reason: automated bidding has the highest likelihood of succeeding when we (the marketers) aren’t constantly shifting the parameters. Pulling levers in accounts all the time – as tempting as it may be – prolongs (or resets) learning phases, which results in more inefficient spend. I’m not saying never make changes (you should!) – but batch them. To give a physics analogy, imagine you’re throwing rocks in a pond. You could throw small rocks every few minutes. What will end up happening is the waves from each rock will reinforce each other, and you’ll end up with a pond that’s in a constant state of flux. Or, you could throw one big rock into the pond, it’ll make a splash, and rapidly return to an equilibrium.

For those people still trying to adjust bids manually, or turn keywords on-and-off multiple times per day (and yes, I see accounts regularly where this is the case): you’re trying to out-trade a machine that has access to exponentially more data than you do, doesn’t need to eat, sleep or socialize, and (here’s the key) doesn’t want to fight you. Work with the machines, not against them.

Sin #2: Solving For the Wrong Thing

The simple reality is that automated bidding has gotten incredibly good at doing *exactly* what you ask of it. That’s a wonderful thing. The issue is that marketers (myself included, at times) have tried to get too cute, and directed it toward an intermediate step vs. the thing I actually wanted.

Classic examples of this include optimizing Meta Ads for Add To Cart (vs. purchase) or Google Ads for “Form Starts” vs. “MQL” or “SQL” (and yes, I just reviewed an account that had “form starts” as a primary optimization action).

On the surface, this could be viewed as smart – after all, wouldn’t getting more people to add to cart be a good thing? All things being equal, yes – assuming that the people with the highest probability of adding to cart ALSO have a sufficiently high probability of completing the transaction. The same is true for forms, calls, chats, etc.

But here’s the rub: automated bidding is exceedingly good at getting you *exactly* what you asked for. If you optimize for ATC, you’ll get a LOT of ATCs. The rate at which those convert into purchases will likely tank, but you’ll get your ATCs. If you optimize for form submissions, you’ll get form submissions. The quality is a crapshoot, but you’ll get form submissions.

I have a theory on why this happens. This turns on two interrelated concepts:

Order-of-Fill – In each of the above examples, there is a series of progressively more restrictive events: you can’t purchase without first adding to cart; you can’t have an MQL without first starting a form, then submitting the form, then having that form be reviewed for quality.

All automated bidding works on a probabilistic model. To simplify this dramatically, this involves calculating the expected value of an impression based on the user, the creative, and the expected action rate (click rate x optimization action conversion rate), then using that expected value (plus any other requirements you have, like a tCPA, bid cap, or cost cap) to determine whether or not to set a bid, and if so, what bid to place.

To illustrate where this goes wrong, consider a simple purchase of a jacket. Let’s assume, for the sake of simplicity, that every day 500 people will view jackets; of those 500, 200 will add them to the cart, and of those 200, 100 people will buy a jacket. Now, assume we have three advertisers:

Advertiser 1: Optimizing For Jacket Views

Advertiser 2: Optimizing For ATC

Advertiser 3: Optimizing for Purchase

Each ad platform has access to a wealth of data about each user; based on historical data and behavioral patterns, Google/Meta/Microsoft can pretty reliably predict who is going to do what (view, add to cart, purchase). So, when Advertiser #1 tells Google/Meta it wants Jacket Views (while #2 and #3 optimize on ATC + purchases, respectively), the expected value of auctions for JACKET VIEWERS (the 300 people who will view, but not ATC) shifts. The result? Advertiser #1 will get an inordinate share of the people who just want to view, but not to buy. They’ll likely get some of the 100 in there because data isn’t perfect and these are probabilistic models; low-probability things happen all the time.

The same thing will happen to Advertiser #2 – they’ll get LOTS of ATCs, but many of them will be the people that tend to add stuff to a cart, but are quite unlikely to complete a purchase.

Finally, there’s Advertiser #3. In this case, the platform is going to bid much more aggressively for the people with a high probability of purchasing – so high, in fact, that they’ll likely win out over the other two advertisers for many of the people in the “100” who will buy. They won’t win them all, and some of the 100 will likely be quite random. But over the long-term, Advertiser #3 will get a disproportionate share of the jacket purchasers, while #1 and #2 will be left wondering what happened.

Sensitivity To Initial Conditions – If you’ve ever heard of chaos theory and wondered what it’s good for, here’s a concrete example. All optimization-driven targeting (i.e. Broad Match or Broad or Lookalike) is incredibly sensitive to initial conditions. To illustrate what this looks like, check out the below graphic:

Despite the fact that each of those three segments (All customers, $200+ spenders, and trial kit buyers) have overlaps, the outputs are dramatically different for each group. The same thing happens for other event-driven targeting. To use a metaphor, if you’re optimizing for Adds-To-Cart or Form Fills, you end up (effectively) fishing in a very, very different pond from the one where your ideal customer is swimming around.

This is an example of the compounding, negative impacts of optimizing for the wrong thing: not only does it hurt your short-term performance, over the mid-to-long term it actually creates a self-reinforcing negative spiral, where automated bidding continually expands your targeting further and further into the pool of people who have a high probability of doing the exact thing you asked, but NOT the subsequent things you actually want.

If you want to succeed in today’s era of digital advertising, you need to do two things: (1) set the right target – which should be the thing as close to your end business objective as possible and (2) allocate the right cost parameter. I’m a huge proponent of tCPA / tROAS / Cost Caps / Bid Caps (if you have the data) for this reason. Having the right objective alone is not enough; you need the right cost target so that hitting that target is done at-or-below budget.

As an aside: if you don’t know what your end business objective is, or what your target cost should be, or if you don’t have a forecast or strategy in place, that’s not an ad account problem, that’s a strategy problem. You should solve that first.

Sin #3: Bad Data

As digital advertising becomes more and more automated, the relative importance of good, clean, highly relevant data increases. Data becomes the optimization lever.

From everything I’ve seen, there are two major reasons behind bad data:

Poor Capture Setup: While I think GA4 is absolutely necessary and the right long-term move, there’s no denying that the rollout and migration has been nothing short of an unmitigated disaster. This is one example of a broader issue: data capture setups are rarely, if ever, audited and updated. The result is that duplicate data, bad data, inaccurate data and sometimes irrelevant data are all passed to ad platforms. Those platforms assume everything they’re given is accurate and complete. And, when you combine this with the order-of-fill and sensitivity to initial conditions from above, the result isn’t pretty.

Unwillingness To Share: The second major blocker is that most brands have been loathe (to put it mildly) to share more of their business data with Google, Meta, Microsoft, etc. – whether that’s through enhanced conversions (i.e. appending more data to forms/purchases to make matching conversions to users easier), or offline conversions, or leveraging smart business data to create value rules, or something else. For some, there’s a fear that Google or Meta will use this data to compete with them or take their business; for others, there’s a general (and, at this point, well-earned) distrust of every platform; for many, the import of good data has never been properly communicated. Whatever the reason, the solution is simple: start feeding better data back to the engines, so they can help make your business better.

The good news is that there are solutions to both of these challenges. These can – and should – be addressed immediately.

Sin #4: Meaningless Segmentation

Many digital marketers grew up in the era of hyper-segmentation. And, for a time, segmenting everything was 100% the right way to structure most digital marketing accounts. That era is over. The importance of segmentation is not. But it looks different today than it did in 2010 or 2015.

My bedrock principle for segmentation today is this: Segmentation is not free. If you’re going to segment, the marginal gain of doing so must exceed the cost of fracturing your data set.

Translating this principle into practice can be tricky, so here are five questions we use to determine if segmentation is warranted:

What’s the Impact of Creative/Offer/Angle? – this is the #1 reason to segment: there’s a meaningful gain to be had by using a different creative, offer or angle. There’s often no point in segmenting a bunch of different keywords or audience segments if you’re just going to copy-and-paste the same exact ad for each one. If you’re going to go to the trouble of segmenting, ensure that you have tailored messaging for each segment at the very least.

Is Audience Performance Likely To Be Materially Different? – the second major reason is if your audience performance is likely to be materially different. We often segment out existing customers from prospects, because I want to focus our acquisition efforts on net-new customers. The need for this is slowly going away with features like “new customer acquisition” bidding in Google Ads, as is the efficacy of some exclusions (especially for remarketing, where we see 20% to 40% leakage), but with good data capture + passback, I think this is still viable.
Does This Traffic Need A Different Destination? – the second major reason to segment is if you want to send traffic to different destinations. For Google Ads, this is required (you can only have 1 destination URL per Ad Group); for Meta, it isn’t, but it should be. Sending traffic from the same ad set to different landing pages is likely to muddle the mixture + result in a bunch of craziness. The same holds true for queries with vastly different underlying intents – someone searching for “replacement windows vs. reglazing” likely needs to be sent to a page providing specific information, not a lander for replacement windows.

Is There A Compelling Business Reason – Your ad account(s) needs to work for your business. If there’s a compelling business reason – such as different geographic offerings, divisions, budgets, etc., then you might have to segment. I don’t love doing it, but at the end of the day, your ad account needs to work for your organization. The cost (time + money + effort + gray hairs) of changing business ops to suit a fickle-in-the-best-of-times ad platform is rarely, if ever, worth it.

Is There A Material Change In Value? – This one is trickier – but if there’s a dramatically different expected value from a set of queries, segmentation can make sense. A classic example here is searches with and without the word “attorney” or “lawyer” – [Truck Accident Injury] and [Truck Accident Injury Lawyer] have dramatically different expected values. In a perfect world, some of this would be picked up by automated bidding; but we don’t live in one of those, and you might want to set bids for those differently. I think this rationale is slowly going away as data imports become easier + better, but for now, it’s still a consideration.

Here’s the bottom line: if you’re going to segment, it should be for a good reason. There is a very real cost to segmenting – you are diluting the data available to the platform to optimize, which means you need *more* optimization events (and more budget) to exit the learning phase. Your default should be consolidation with segmentation as a lever to improve your ad account’s performance.

Sin #5: Failure To Fund

This might be the most frustrating issue I see when doing audits: brands that over-diversify or are limited by budget.

I’ve written about the costs of over-diversifying before (see here), but it bears repeating: until you’re able to consistently spend $50,000+ per month on a single platform (Google Ads, Meta Ads, Amazon Ads), diversification tends to be a net-negative. There are exceptions, and sometimes a balance of two core platforms (Meta + Google, for instance), can yield an amalgamative result (that is, the result of the combination is greater than the sum of each channel individually).

If you’re a smaller (<$10M/yr) brand and you’re trying to advertise on Google, Yelp, Microsoft, Meta, LinkedIn and TikTok, you’re not going to make it. Not only do you likely not have the resources to invest in each channel, you likely don’t have the human capital to leverage what you’ve learned on each platform to accelerate your acquisition flywheel. The end result is that you spend more, and the only thing you get more of is confusion. Instead, fund and scale one channel to at least $50k/month, then add another.

The second issue – and the more insidious one – is being limited by budget. There’s nothing more frustrating than a client wondering why they aren’t hitting growth targets, while their properly-configured ad accounts are limited by budget.

My preferred setup is for an account to never hit its daily budget; if an account is achieving its target outcome at or below target cost, then spending should be unconstrained. If that’s $1,000 one day and $5,000 the next, fine. As long as the client has the capacity to deliver (whether that’s goods, services, call center capacity, whatever), and the conversion quality is at or above target, the budget should not limit delivery.

The reason for this is simple: demand is not constant. Consider the below graph, which uses anonymized data from a client account in July:

In this case, the entire area above the yellow line (the budget) and the red line (the demand) is a missed opportunity. These are leads + sales that this account could have captured at or below target cost but did not due to budget constraints. For a client in this industry, this could represent hundreds of thousands of missed contribution dollars. This is a silent growth killer.

Instead, use your efficiency target (Target CPA, Target ROAS, Cost Cap, Bid Cap) as your steering wheel, and your budget as a guardrail. Assuming the platform is being fed good data and the efficiency targets make sense for your organization, this setup gives the highest probability to raise the floor of your account (by eliminating bad/non-productive/blow-up spend) while also raising the ceiling (since you’ll have the resources to capitalize on volume spikes).

Sin #6: Lack of Experimentation & Evolution

If there’s one theme from this week’s edition that should be clear, it’s that things change (often quickly) in this space.

For an ad account, every day is a battle against gravity. The equilibrium state of an ad account (Google, Meta, whatever) is failure. As advertisers, our job is to fight that gravity – whether that’s through better data, more compelling creative, capitalizing on latent/hidden opportunities, a better product, a superior offer, something else, or all of the above.

The tools that we have to do that are experimentation and evolution.

Experimentation is one of the most under-utilized features in every ad account I review – you should be running Experiments and/or A/B tests (and those tests should be bigger swings, as I wrote about in 10% or 10x constantly. You should be running more (and more diverse) creative tests on Meta or YouTube or Pinterest. You should be testing different bidding strategies and account setups. You should be testing new audiences + keywords + topics + targeting. Test shifting your optimization event closer to people and/or profit – for instance, going from optimizing on MQLs to SQLs. What works today isn’t likely to work tomorrow. The purpose of experimentation is to discover what’s going to work in the future.

Remember: the equilibrium state of an ad account is failure. Every day you’re not trying to find new ways to succeed – to capture incremental alpha – is a day you’re getting closer to failure.

But experimentation means little without evolution. Once you find what works, you must evolve your account to include it. Every month, I see accounts that haven’t been updated in years (one in a decade!). Those legacy structures can produce acceptable results for far longer than we’d like to admit, but eventually, everything succumbs to gravity.

Experiment, then Evolve.

Sin #7: Neglecting Your Creative

Finally, there’s creative. I’ve touched on it in various sections of this issue, but I can’t overstate the importance (and difficulty) of building a sustainable creative engine.

Structural + account-based challenges can be solved (at least for a time); creative is an ongoing, evergreen challenge so complex and massive it deserves its own issue. But, since this is already quite long, I’ll hit on the primary creative issues I see in accounts:

Lack of Creative Diversity: most brands find a creative style they like or that “works” for them (and yes, “UGC” is a style), and they typecast themselves into it. That’s a mistake. There are dozens of different creative styles you can (and should!) try, then try again. I wrote about some of them here.
Neglecting (or ChatGPT-ing) Your Copy: In a video-dominated world, I’ve never been more bullish on the power of words to rescue brands from the abyss of mediocrity. The more brands turn to ChatGPT (and other LLMs) to write their ads and posts and websites, the more incisive, punchy, witty, devastatingly brilliant copy will shine. In a world of sameness, differences become exponentially more powerful. A great copywriter might be the best investment a brand can make in a post-ChatGPT world.
Failing to Customize Your Ads: One of the least-used (and most powerful features) in Google Ads is Ad Customizers. Less than 5% of Google Ads accounts I’ve reviewed use them. And their impact can be MASSIVE – both in terms of performance and in freeing up your team to do other things. The more relevant your ad, the higher your chances of success. Ad customizers help make more relevant ads.

These certainly aren’t all of the creative issues I’ve seen, but they are certainly the most prevalent. Your creative is the connection point between your brand and your audience. It can either be the spark that starts a beautiful relationship, or the hole you have to dig out of just to take your shot.

Thanks for sticking through this with me – I know there’s a lot to digest. If you’re wondering how to apply this (or just don’t have the resources to do it), this is my shameless plug that I have a few audit spots available – just respond to this email and we’ll get it set up.

Cheers,

Sam

Related Insights