Article

Marginal Cost + Frontiers

Advertising Analytics Consumer Behavior Ecommerce Economic Analysis google ads

by Sam Tomlinson

October 30, 2023

This issue was inspired by (seemingly) every brand’s favorite near-year-end question: how much should we budget to spend on marketing next year, if our goal is to [whatever]? 

Fundamentally, this is three questions jam-packed into one: 

1. At what point does investing an incremental dollar in marketing no longer yield a positive return? 
2. What is the optimal distribution of those dollars across channels/tactics in order to maximize return and/or minimize spend? 
3. What is the basis we are using to determine #1 + #2? 

Let’s start by unpacking #3, proceed to #1 and #2, then explain the tie-in between this (very loaded) question and marginal impact.

What is the basis for determining return on investment?

For all the talk in the marketing universe about “ROI” – there’s virtually no consensus on what it means. That’s a huge problem. When you’re not clear how success is defined, you’re not going to be clear if/when it’s achieved. From a big picture perspective, there are a few ways you can calculate return; each one comes with tradeoffs: 

1. Revenue/Lead Maximization – the most common approach is to compare advertising spend to total revenue. This method boils down to: spend until the ROAS of the next dollar is less than 1 or the cost of the next lead exceeds the mean expected value of that type of lead. 
2. Contribution Margin Maximization – the least common approach (from my data set) takes a decidedly different tact: spend until the contribution dollars from the next sale is less than or equal to $0.00. Of course, this assumes that (1) we all have the same definition of contribution margin and (2) that your marketing team even knows what contribution margin is. If those are true, this method maximizes total contribution dollars generated by your marketing efforts. 
3. Lifetime Value Maximization – for the last ~3 years, LTV-based acquisition was the “in” thing. And in theory, it makes sense: of course you’d pay more to acquire a customer who was going to spend 2x or 3x their initial purchase over the next 3/6/9/12 months. Of course you’d pay more to acquire a lead that had a 2x higher expected value than your standard leads. The problem here is threefold: (1) a lifetime is a long time; (2) inflation is a very real, very painful thing; and (3) the risk profile of LTV-based acquisition is radically different from anything else here. 

If you think of these three approaches on a curve, this is an approximation of what they’d look like:

As you can see, depending on how you (or your client) is thinking about return drastically changes the ad spend required to achieve that goal. Once you have clarity on this, answering #1 above is a relatively straightforward question. #2 is still a bit tricky – but that’s why you should build a tiered, cohort-based forecasting model.

Armed with that understanding, the next challenge is modeling the marginal impact of your advertising. Whenever I discuss marginal costs, I think it’s helpful to articulate three core principles, all of which are intimately intertwined:

1. The Next Is More Expensive Than The Last – every digital ad platform is designed such that, all things being equal, the next click (or impression, or whatever) is more expensive than the last. Obviously, our job as advertisers is to ensure that all things are not equal – whether that’s through a better product, better data, better creative, better targeting, or some combination thereof. Any way you slice it, advertisers are in a never-ending battle against gravity – ad platforms want us to spend as much as possible at the lowest-acceptable return; we want to maximize our return on every dollar spent, while meeting or exceeding our overall volume (sales, leads, etc.) goals. 
2. Averages Obscure Insights – Every major ad platform reports on averages: CPC, CTR, CVR, CPA, etc. While these can be useful for some tasks (like a top-level analysis of performance, or period-over-period comparison), they obscure what’s actually going on in an ad account. Take the following, hypothetical example:

If you were to review just the above CPAs, you’d likely (and correctly) conclude that each of these ad accounts had performed comparably over the time period. But when you see the underlying distribution, you see that Ad Account #3 was staggeringly more efficient for the first 5 transactions – then the blow-up 6th transaction happened. Had that not occurred, Ad Account #3 would have posted a CPA of ~$3.62 – less than half of the other two. 

For anyone investing in digital marketing channels, the above is likely both a familiar and vexing reality – for no other reason than platforms (like Google, Meta + Amazon) make discovering the marginal cost of a given sale near-impossible. 

This cuts both ways – just as blended ad platform metrics are bad, so too are blended business metrics. Combining new + returning customer revenue, or all SKU revenue, or all leads together obscures the insights you need to make an informed decision.

3. Blended Always Trails Marginal – the reason to focus on marginal metrics is because they are as close as you’ll get to current; by the time a blended metric turns negative, the marginal metrics have been negative for some time. For platforms, this is a benefit that allows them to recoup some of the surplus value you’ve gained on your previous spend; for advertisers, this is the silent profit killer. Marginal is where the magic happens – the closer you can get to true marginal performance, the better you’ll be able to allocate marketing investments, and the closer-to-optimal you’ll be able to perform. 
4. Marginal Is Never Perfect – you’re never going to pinpoint your marginal impact precisely; the best you’ll get is the marginal return at different levels of spend – for instance, $100/day vs. $150/day vs. $200/day. The difference in performance at those levels is the marginal impact of those investments. If you’re thinking about your advertising investments in these terms, you’re already ahead of 90% of advertisers. 
5. The Tension Between Efficiency & Growth – finally, there’s a real, under-discussed tension between growth + efficiency. This is far too complex a topic to cover fully here, but I’ll do a newsletter on the topic later this year. Here’s the jist of this – and why it’s relevant: whenever you’re trying to maximize your return while minimizing your spend, there’s a temptation to overindex your investment toward “capture” channels/tactics vs. growth channels/tactics. Examples of this are everywhere, but the common ones include branded search, remarketing, email and SMS. None of these are inherently bad things to do (and you should do them!); the issue is that they aren’t expanding your customer base. These channels/tactics are designed to either (a) capture demand created elsewhere and/or (b) extract more value from your existing customers. From a blended metric performance perspective, these tactics tend to look incredible – sky-high ROAS, lower acquisition costs, higher contribution dollars per sale. But when you split out new vs. returning customer performance, they are less attractive. And when you include the costs associated with increasing branded search and driving your original traffic (which then gets rolled into remarketing), it gets worse. This is where marginal impact shines: it forces you (the advertiser) to hone in on the point where advertising no longer makes sense from a growth perspective. 

Cool Theory – How Do We Apply It?

One of the most common applications of this is within Google Ads, Amazon Ads + Meta Ads accounts. All three have an (all-too-common) “suggestion” box that asks you to modify your budgets and/or Performance Targets in order to drive more sales/conversions/leads/whatever. 

Google also has their “performance planner” (screenshot below), where you can see the performance of your ad group or keywords at varying levels of spend: 

This provides all of the data a smart marketer needs to calculate the marginal impact of incremental spend. I’ve stitched two of these together in the below; for spend levels where two different conversion totals were given, I used the lower of the two. 

From there, it’s pretty simple subtraction (subtracting the conversions from the Spend+1 level from the Spend level gives you the incremental conversions driven by the increased spend; the same holds true across the board for impressions, clicks and cost). Once you have the incremental numbers, you can calculate incremental rates (for instance iClicks / iImpressions = iCTR). This reveals what Google (or Meta, or Amazon) don’t want to make readily apparent: that the marginal performance of your advertising as your budget increases tanks. 

Look at the incremental cost per conversion at the $590.94/day spend level Google “recommended”: using the above method, I calculated that $599.52 is the incremental cost for every conversion over 29. Yikes! 

But – just as stated above – look at how the “blended” Cost/Conv. obscures the insane increase: it’s just $381.52 at this level of spend – which, for this client, is bordering on the “acceptable” threshold. An advertiser who isn’t familiar with marginal costs is quite likely to look at that and think, “Let’s do it” – after all, the Cost/Conv. is right at the upper bound of acceptable. But, by knowing and calculating the marginal cost at each spend level, you can see that spending above $250 is going to dramatically decrease your efficiency, to the point where you’re giving back some of the surplus value created at earlier spend levels. 

Conversely, there’s a weird inflection point at about a $140 tCPA where the incremental cost/conversion actually decreases with a higher tCPA. That’s a pretty good spot to bid (spoiler: definitely did!) – likely created by suboptimal targeting and/or allocation from competitors. By doing this exercise, this client saved upwards of $9k in a single month, while hitting their desired lead target. And while this specific example focuses on lead gen on Google, the same exercise can be done with eCommerce on Meta or Amazon. 

Armed with the data on your marginal impact, you can set more accurate targets in your ad platforms, allocate investments to more optimal channels, and avoid falling into the traps that platforms set to recapture the surplus value you’ve already created. 

Segregate New v. Returning

Just as ad platforms blend metrics, so too to organizations – we lump all revenue and leads together. We group all ad spend into a single line item. And the reality is that doing so is every bit as toxic as what Google + Meta do. Instead, think in terms of New vs. Returning. In eCommerce, this is typically noted as “aMER” vs. “rMER” (acquisition Marketing Efficiency Ratio and retention Marketing Efficiency Ratio). You can do the same thing in lead gen. These calculations are relatively straightforward: 

aMER = [new customer revenue] / [new customer ad spend]

rMER = [returning customer revenue] / [returning customer ad spend]

If you plot either of these figures on a graph, you’re likely to get something that looks like a slight negative line (remember principle #1 from above: the next click is always more expensive than the last, all things being equal): 

This tells you two things: (1) every dollar you invest in advertising tends to yield a progressively smaller return (though there are exceptions!) and (2) the rate at which returns decrease is NOT always linear. Plotting these marginal curves is what enables smart advertisers to invest smarter – both by setting better targets, as well as by avoiding “traps” that ad platforms set to capture more of your hard-earned money.

Channel-Specific Allocations:

Finally, the 2nd question from the outset of this issue (what channels/tactics should we invest in?) can be answered – at least in part – using the same principles from the last two examples (ad platform budgets + targets and new vs. returning customers). Each channel will have different marginal cost / return curves. When you plot each one, you’ll often find that marketing dollars should be shifted between your platforms. 

Take the below example: 

In this case, if your total advertising budget is $50k per month, you’d likely want to allocate ~$30k to Google + ~$20k to Meta, as doing so maximizes the number of conversions (leads, sales, whatever) within that budget. But, if you have a $100k advertising budget, the better distribution is $40k to Google + $60k to Meta. This is obviously a simplified example, assuming that all conversions have the same expected value and are more or less homogenous. 

In the real world, that’s rarely, if ever, the case – but there are well-defined, widely-accepted methods to adjust your marginal metrics to account for those platform/channel/tactics specific differences: use lead qualification rates by channel, or contribution dollars (which accounts for different COGS for different SKUs). 

Once you start doing this – and start thinking in marginal terms vs. blended terms, you’ll unlock additional levels of performance that you didn’t think were possible. 

This is what enables you, the savvy marketer, to answer the question that we started with at the outset of this newsletter: how much should you invest in marketing? (the answer, of course, is, “Well, it depends.”)

And I hope this issue helps you explain what it depends on, and how you can go about answering that question in a thoughtful, data-informed way. 

Until next time,

Sam

Article

Insane Principles That Lead To Better Outcomes

Article

The Road To Ruin Is Paved With ROI