|
This edition of TMAI Premium is free for Standard Edition subscribers.
In Premiumlandia, we've had an exciting few weeks. We dived deep into brilliant brand measurement - tackling optimal metrics, formulas, test - control measurement, and advanced topics like headroom and minimum detectable lift. The difference between mediocre and magnificent impact via data? Leaders! Over two editions, I'd shared the 12 leadership values that'll transform your influence and impact. You can upgrade to TMAI Premium here - it comes with a money-back guarantee. All Premium revenue is donated to charity. | | TMAI #327: Attribution Modeling? Stop.
| Today I’d like to share a perspective that’s, perhaps, controversial. It'll thoroughly change your execution strategy - and impact via data.
As is likely true for your team, we never have enough Analytics resources (people, budgets) to do all the work that needs to get done, either with our team, or with our Agency partners.
In that context, it is likely not surprising that I have to be constantly vigilant about what we are working on. One poorly chosen strategic priority can jeopardize the year’s ROI From Analytics.
Recently, I made the recommendation that we should stop our multi channel attribution projects, as they are a massive waste of costly resources when something far more important is being ignored that needs attention.
There is a high probability that you are making the same mistake.
Let’s take a step back, and then 17 steps forward. | What’s Attribution? Say we work at Expedia. We are running a large digital campaign focused on selling our delightful inventory of tropical island resorts. Perfect for this year’s winter travelers. (On wishlist: Remote Resort, Fiji.) The campaign is a wonderful success, high-fives all around. It delivered 100,000 conversions as reported by our tool of choice, Google Analytics. | Now we all know that when Google Analytics shows 100,000 conversions from Facebook, we are not getting the complete story.
Last click reporting sucks. Because, it lies.
For the campaign we spent our marketing dollars on Facebook, Spotify, and Google. Surely, the other two added some value.
This is asking your analyst to do attribution modeling. | Take the 100,000 conversions and distribute credit across Facebook, Google, Spotify. If you have Google Analytics, just use the Data-Driven Attribution (DDA) model that leverages machine learning to help identify how credit should be distributed across Facebook, Google, Spotify. It is excellent for most use cases, and really good for all others. [Bonus Read: Digital Attribution's Ladder of Awesomeness: Nine Critical Steps.] Let’s assume DDA indicates attribution of 30 - 30 - 40 or 10 - 20 - 70 or some combination. Great. Expedia won’t dump all the marketing budget on Facebook next quarter. Super great data-identified decision. | What’s Incrementality?
In this quest for attribution modeling, you are forgetting a significantly more important question: How many of the 100,000 conversions we are reporting - however attributed - are actually incremental?
Put another way: How many of these 100,000 conversions would NOT have happened anyway? Plenty of people are already looking to book winter getaways this time of the year. At least some of them did not need the enticement of Expedia ads to end up on expedia.com. Attribution modeling does not answer this incrementality question. |
|
At the end of all that attribution work, you can’t answer a CFO’s most important question: What if I gave Marketing no budget? Industry studies, and my experience across clients, indicate that Marketing’s incrementality will be in the range of 0% to 25%. Closer to 0% if all your marketing reaches your existing customers or buying two brandiest brand keywords is your entire search strategy or your paid advertising only reaches new customers who’ve already made up their mind to buy from you because of customer reviews or your “free” strategies like email or your TV ads are being delivered to people already frustrated by your competitors and are on their way to switching. Closer to 25% if your Marketers work super-duper hard to execute strategies that meet two outcomes: Persuade those who would not have bought from you. Reach those who have never heard of you. 25% is rare. In our hypothetical Expedia example, let’s assume Expedia’s Marketers possess God-like targeting/persuasion skills, hence are delivering 30% incrementality for this Tropical Islands campaign. Here are the results, viewed through that lens… |
|
30,000 Conversions.
Not 100,000 Conversions.
And, 30% incrementality is not even a reasonable assumption!
[New language for you: Claimed Conversions: 100,000. Actual Conversions: 30,000. Start using that in your reporting/dashboards.] | The Problem with Attribution without Incrementality?
Bad marketing.
Let’s say your attribution model identified that on a base of 100,000 claimed conversions, Spotify needs a 50% increase in the budget expended.
You don’t know if everyone you got via Spotify was non-incremental. I.E. Spotify is uniquely blessed to be rich in an audience who is going to buy your product any way.
Replace Spotify with Facebook or Google, in our example.
All that work you are putting into attribution – without understanding incrementality – gloriously marching you down the path of marketing that is no better than using bad-old-last-click or some other guess-based assessment of success.
30,000 < 100,000.
By a lot.
Every moment you spend on attribution modeling, without understanding incrementality, is a moment you are spending with the behavior of those 70,000 non-incremental outcomes polluting your analysis and contributing to poor decisions.
Consider how much worse this whole exercise is if you are paying an external vendor or your Agency to build you custom attribution models or hyper-customized systems that take years to build, only accounts for x% of use cases/data and uses modeling that sounds impressive only to non-analytical leaders.
There are very few ways to flush good money after bad, apart from doing the above.
The marginal value from these custom alternatives is so low, they are simply a colossal waste of time, talent, budget – while stealing the focus from what’s essential: Understanding Incrementality. | Incrementality, then Attribution. NOTHING I’ve stated above is to imply that attribution is not important. There is a vocal minority that dunks on attribution because it does not meet their personal agendas. I am not a part of this group. Attribution modeling is of value. At the minimum, do attribution modeling for your digital Owned, Earned, and Paid marketing efforts (as you can out of the box from Google Analytics, with DDA, for free). To get to Over Achiever status, do attribution modeling across your online and offline paid media efforts. (Repeating bonus read link above with how-to accomplish that.) Just do attribution modeling after accounting for the incrementality of your marketing efforts. It is not uncommon that your Direct Response / Performance Marketing is delivering 5% incrementality (actual conversions), then you are wasting time, talent, budget doing attribution modeling across 100% of your claimed conversions. It matters little how sophisticated your attribution models are. |
|
If you can only afford time, talent, budget to solve for one answer: Solve for incrementality. It is far more essential, it will have major impact on your financial results. If you can afford time, talent, budget to solve for the optimal scenario: Solve for incrementality, then off that base, solve for attribution. It is not easy, but it will transform marketing’s value to your organization. Every time a recession comes around, Marketing won't be the first budget cut. | Deep Dives & Gotchas. Like all things worth having in life, what I’m recommending is hard. [It can also be painful, at least at the start because Marketing teams typically have no incentive to identify incrementality. They are happy to get credit for Claimed conversions which are magnitudes higher than Actual conversions.] Google Analytics does not have any concept of incrementality. Hence, this measurement is all on you. There are three types of incrementality measurement. Channel Silo, x-Stack, and Portfolio. [TMAI Premium members, see TMAI #233. If you can’t find it, email me for a copy.] The good news is that Channel Silo incrementality measurement is offered out of the box by top channels. Look for Conversion Lift Studies, on Facebook, Google, others. x-Stack incrementality typically requires Geo Experiments / Matched Market Tests (MMTs). Some degree of sophistication means that most people can undertake this with a measure of Design of Experiments background and familiarity with Causal Impact. Portfolio Incrementality is complex and applies to large companies with loads of spend. Advanced statistical models or MMMs leveraging machine learning are the way to go. [Premium members, please see #233 and #260.] [Bonus: If you discover, as almost everyone does at the start, that your performance is not so good, see TMAI #307: Got Low Incrementality? Six Causal Factors.] IMPORTANT: Out of the box, Portfolio incrementality measurement comes bundled in with attribution across channels (and online plus offline success). Depending on how you structure your MMTs, you can get clean attribution of credit for x-Stack incrementality as well. Channel Silo incrementality by definition excludes attribution (as you are only looking at a single channel). So… By solving for incrementality… You solve for attribution as well… And… Marketing that actually matters. Then, during downturns, Marketing is not the first budget to be cut. Happy birthday! | Bottom line.
Multi channel attribution was religion for me for a long time. It was not just a path to suck less, but a path to rock way more.
The lenses I used to see the world changed. With it came a new view. And, a different way to see the landscape of how to rock way more.
I’ve always reserved the right to change my mind when presented with new information.
My professional life has a different purpose and direction now. Way more exciting, because of just how transformative the impact through data is.
More than anything, I hope I’ve changed the lenses you use to see the world.
Good luck.
Avinash. | PS: On a lighter note... : ) | Committed to investing in your professional growth in 2023? Upgrade to TMAI Premium here - it is published 50x / year. | |
|
|
|
|