Online Attribution is Dying. Clicks are Dying. What Should You Do About It?
A thoughtful response to SparkToro's viral article
My friend sent me this SparkToro article last week, which was blowing up on Hacker News, a popular news site for startup founders. His immediate reaction was the following:
“I’d be curious as a founder how this affects my type of business. This also means a potentially seismic shift in Google’s business model. Maybe we see a distribution of Google’s ad / search business into other businesses? Where is the money going? Is it IRL events? Does this mean knowledge influencers (like those on Twitter) are about to become a whole lot richer?”
Initially, I was a bit puzzled by the response. My friend is a startup founder who has raised a few million dollars, has a small but fast growing team, and has done virtually no marketing yet. While the article itself was extremely informative, well-written, and makes some really compelling points (props to the SparkToro team), very few of the things mentioned in the article applied to my friend’s startup. The fact that the article produced such a visceral reaction in him and produced so many unanswered questions convinced me that it is missing two critical things: context and clear steps on what to do next.
Before diving too deep, it is crucial that I define what online attribution is. As defined by Google’s AI generated answer:
“Online attribution is a marketing process that identifies and values the actions of users that lead to a desired outcome, or conversion. It involves measuring data points related to a user's online activity and connecting them to a conversion event. The goal is to understand which combination of events, in what order, influence people to engage in a desired behavior.”
For example, if a user clicked on 3 ads prior to purchasing your product (e.g. a banner ad on the web, an ad on Instagram, and a Google Search ad), how do you determine which ad was ultimately responsible for the conversion? Do you assign equal credit to all three events, do you say the last ad was the most impactful (commonly called the last click model), or some other approach?
An alternative approach to measuring the efficacy of marketing efforts, lift analysis, takes a different approach: a baseline level of performance is established, an experiment is run, and then performance is measured again to determine how much of a lift can be observed (essentially - looking at before vs. after). At a high-level, attribution attempts to track every user touch point precisely and attempt to arrive at a level of mathematical precision, while lift analysis is much more about estimating the impact of various changes.
While I would suggest you consider reading the full article (it’s relatively short), the key points it makes are as follows:
-Online ad tracking has massively diminished in effectiveness due to some recent developments.
-As a result, attribution does not work anymore.
-Instead of traditional attribution models, you should use lift-based models to measure your marketing’s effectiveness.
-You should also do plenty of research on who your audience is and where they spend their time online to identify the best channels to invest your marketing dollars.
These points all have merit, but absent context, can lead readers to some bad conclusions. As far as SparkToro’s suggestions on what to do about the points above, they did write a follow-up article which suggests creating a robust dashboards to track everything related to marketing. In their own words:
“I’ve put real numbers from Google Analytics, Google Search Console, Twitter Analytics, LinkedIn Analytics, our Mailchimp email list, Crowdcast webinar attendee analytics, and others to build a reporting system for how the top, middle, and bottom of our funnel relate.”
The issue is that these types of dashboards can be tricky to create, require a lot of time and effort to update consistently, and it is very easy to drown in too much data and get analysis paralysis. So what should you do instead? I’m going to give you a three step framework which is far more actionable for most entrepreneurs, business owners, and marketing VPs, but first, I need to establish a key point.
Attribution modeling is still superior to lift analysis
While I found SparkToro’s article informative and data-driven, one issue I had with it is the alarmist headline and the black and white thinking. The phrase, “attribution fundamentally doesn’t work anymore” can easily lead one to throw up their hands and choose to give up on any type of attribution entirely. Furthermore, stating that one should “optimize lift-based measurement and not attribution” makes it sound as though lift based measurement is, in 2024, a far more reliable option than traditional attribution methods. I think this is wildly off-base.
The major challenge with lift-based models is the extreme difficulty of keeping all variables constant and running controlled growth experiments. The article assumes its target audience are part of businesses where things are relatively static outside of the single growth experiments that are consistently run and measured. In reality, almost every business has constant change, many moving parts, and a ton of noise that makes running precise experiments excessively tricky.
You might run an experiment where you spend $50,000 doing direct media buys with popular sites such as the NYTimes, CNN, ESPN, etc. to place banner ads on their website. You then see a 10% boost in sales month over month, but how sure can you be that these two are related? It could be some mix of seasonality, statistical variance, word of mouth, one of your social media posts going viral, a change the engineering team made to the website you were not aware of, a conference your CEO attended that brought in a new big customer, etc. Sure, you can run these experiments for multiple months, get more data, and come closer to a clear answer, but by that time, you have spent a huge amount of budget and potentially lost a lot of money in the process.
The second issue I have is that there are still a sizable number of marketers who are creative types that are allergic to math and numbers and who will grab any possible opportunity to muddy the waters and avoid accountability for results. I started my career in the programmatic advertising industry. While the firm I worked for, MightyHive, did its best to operate ethically and transparently, the industry itself has an excessive number of shady actors. I wouldn’t be surprised if the gas station boner pill industry was more trustworthy.
I’ve seen countless examples of marketers claiming their $500 ad campaign was the reason their company saw a 15% boost in sales one month, while claiming the next month that the 10% drop in sales was “due to external factors outside of our control”. I’ve seen revenue numbers for the quarter miss wildly, with the response being, “it’s still a successful quarter, due to the increased click-through-rate and 8% boost in our social media following.” I’ve seen marketers claim a new website change massively improved results, to later claim it is actually unclear, to later claim more data is needed, to eventually claim the change made no real difference. This is not to say that lift analysis is not useful in some scenarios - it is just that old school attribution, when it is possible, is a superior option. Here is my three step framework on how to implement attribution as effectively as possible.
First: Determine if you need to worry about attribution yet
When you are attempting to convert traffic for your online advertising campaigns, there are two different scenarios: demand capture or demand generation. Capturing demand involves converting traffic that knows they have a problem and is actively looking for a solution. Essentially, as laid out in this article demand capture is targeting someone who is at or beyond the “Solution Aware” stage.
For example, let’s say I own a pest control business in New York City and someone searches on Google for ‘pest control service nyc’, clicks on my ad, and signs up for my service. This is a perfect example of hitting the right user with the right ad at the right time and thus it is very easy to push the sale over the line. The vast majority of these conversions are going to be 1-click or 2-click conversions that do not have too many user touch points, and thus you can have a solid degree of confidence in the conversion numbers reported by each ad platform.
The only issue: there are always going to be hard limits on the amount of pre-built demand you can capture. At a certain point, you need to start moving up the ladder to convert those who are “Problem Aware” and potentially even those who are “Unaware”. What might this look like?
On Google, it might involve pursuing an SEO strategy where you create content specifically targeted at people who have a pest control problem but are not yet sure they need to pay for a solution. Some topics might include, "Can I Get Rid of Pests on My Own?”, “Is Pest Control Safe for Kids and Pets?”, or “Is Preventative Pest Control Necessary for My Home and Business?” These people may not yet be sold that they want to pay for a pest control service, but by educating them and helping to handle some of their objections, you can progressively nudge them closer and closer to being willing to open their wallets.
Another example might be with Instagram Ads. A user has recently moved into a new NYC apartment and is scrolling on Instagram when they see a bold image overlayed with the text, “70% of NYC residents experience pest issues in a 5-year period. Take action today to ensure you’re not one of them.” This bold statistic, (which I made up), could easily stop a user in their tracks and make them think, “Wow, I didn’t even realize this was an issue. I better take action to avoid a big problem.” A perfect example of generating demand where none previously existed.
The crux of attribution is determining, “How are my demand generation efforts contributing to my demand capture efforts?” Without doing this analysis, it is easy to overvalue the final conversion event and not recognize all the mini steps in between that got your product on the prospect’s radar or warmed them up.
A key thing to keep in mind though: capturing demand is much cheaper and more cost-effective than generating demand, and you should max it out first. While it does differ from client to client, if you are spending under $10k / month on ads and you are not a local business, odds are very good you can coast entirely off captured demand for an extended period of time; possibly indefinitely. If that is the case, attribution should not really be a concern of yours, since the vast majority of your conversions are going to be 1-click or 2-click conversions, most of which occur on a single platform, and which are easily attributable.
If you are spending $10k - $30k/month on ads, you probably need to start thinking about attribution options. If you are spending $30k+/month on ads, you definitely need to implement a good approach to attribution. Personally, I have never seen an advertiser spending $100k+/month that did not need a robust attribution system.
Second: Set up tracking so you can accurately attribute as much as possible
The article from SparkToro is correct that at some point, if your business gets large enough and you start to expand your marketing efforts, you will run into very real limits with attribution. With that said, there are still many events that are within your control to properly attribute and you should absolutely not adopt a mentality of learned helplessness and give up on attribution entirely.
One thing needs stating though: for many businesses, the number of touch points and the complexity of attribution is vastly overstated. When I started working at MightyHive in 2014, it was the heyday of attribution. Not only were very few of the issues stated by SparkToro concerns yet, but multiple technologies and ad platforms had sprung up to provide as clear a picture as possible of user journeys.
Google released DoubleClick Campaign Manager (DCM), a tool that enabled you to pull in and get a centralized view of both impression and click data. Many platforms, including Facebook and LinkedIn, enabled you to implement trackers to pull both impressions and clicks into DCM, making it possible to see how every ad touch contributed to the ultimate conversion (almost no platforms today allow impression trackers). There were articles at the time stating it took an average of 25 or more touch points to drive a user to make a sale, and this was treated as gospel within the industry.
I was assigned to be the head of MightyHive’s efforts to expand DCM usage, and a big part of that was helping to set up and analyze the attribution reports for our portfolio of of over 100 clients. What I found over and over was that the vast majority of our client’s conversions (I’d estimate about 80% - 85%) resulted from a single channel, usually 1-2 ad touches, and thus were easily attributable.
This shocked me at the time, as there were frequent claims at the time that ‘last-click attribution was causing search ads to steal the credit for display ad’s results’ and that ‘user journeys were complex and involved multiple devices’. While I would certainly say that attribution is far trickier today than it was in the past, I would still estimate that for most companies, at least 60% of conversions involve a single channel and are not exceedingly complicated to attribute (for many companies, the % may be much higher). Want to be able to properly track even more of your traffic? Here are a few tips (this list isn’t exhaustive, but it should give you a good place to start):
Use call tracking software: Tools like Call Tracking Metrics or Call Rail allow for real-time number swapping so you can see the source of calls to your business. Even if someone clicks on an ad then calls your businesses’ number directly (with no button click) the tool’s technology will still enable you to identify which paid source they came from.
Use QR codes or unique landing pages: Running offline marketing efforts that are difficult to track? A QR code enables you to append custom tracking codes to any visit, enabling you to see its source. Custom landing pages are also good options (e.g. if your site is www.example.com you could send users to www.example.com/1 for your 1st campaign and www.example.com/2 for your second campaign).
Set up server-side tracking: Many of the issues with tracking created by Apples iOS 14 update can be minimized by setting up server-side tracking. This article does a good job of explaining why.
Ask users how they found out about you: One simple, but under-appreciated way to aid your attribution efforts is simply to ask users how they discovered you. This could be a question on your website form, it could be something the sales team asks before closing the deal, or at any other point in the process. While I would not take the results at face value, they should help steer you in the right direction.
Shorten the user journey: The sooner you are able to get users in your ecosystem, the better you are going to be able to attribute the final purchase. Imagine you have a $500/month B2B product. If the call-to-action on your landing page is, “Purchase now!” It is going to be tough to convince anyone on the 1st visit and it may take 20+ ads to get them to even consider a purchase. On the flip side, if you instead say, “Give us your email to get a free 15 page guide on _____”, you are sure to get a much higher conversion rate from the get go. Once they are in your ecosystem, it becomes much easier to track what pushed them to purchase. It is far simpler to say, “The 4th email in our drip campaign resulted in a click that led them to hop on a sales call where we closed the deal” then it is to say, “They visited our site on 5 different devices over a 30 day span and we can determine which ad touch was most impactful.”
Third: Use attribution estimates or lift analysis to fill in the gaps where needed
Some percent of your conversions will undoubtedly involve a long user journey, many ad touch points, and lot of ‘dark traffic’ that is difficult to attribute. In these scenarios, there are two things I would suggest: attribution estimates and lift analysis.
To make effective attribution estimates, it is imperative that you have solid tracking and accurate data for all steps in the funnel, and a dashboard similar to what SparkToro has created (below) is your best bet to do this.
It is worth stating that this dashboard is a good jumping off point but should not be copied verbatim. The key metrics that matter for each business are going to be somewhat unique and your dashboard should be one you are confident you can update consistently (I would suggest 1x/week minimum).
With that said, once you do have your custom dashboard in place, you can use it to start making some estimates on attribution. For example, let’s say that you have identified that from your organic social media strategy (e.g. posting content 3x/week on Facebook Ads and LinkedIn Ads) you drive 10 click-through conversions per month that are easy to attribute, but also some number of view-through conversions that are difficult to attribute.
One thing you could do is start to ask in your online demo form ‘how did you hear about us?’, see how many people reference your social media, and then attempt to estimate the connection between your monthly social media post views and the total number of social media view-through conversions. Over 12 months of data, let’s say you discover that for every additional 500 post impressions, you will see 1 incremental person state that they found you through social media. This then enables you to start making some intelligent guesses about how your demand generation efforts affect conversion numbers down the line. This is an approach that can also be used for other hard to track marketing efforts, such as how many sales came from conferences that company leadership attended. To be clear, this is not a perfect approach (e.g. user reported data can be off base), but it is a decent starter for making educated guesses.
Finally, there are going to be some events that are very tricky to attribute accurately (or even to estimate) and in those scenarios, lift analysis is going to come in handy. One of the best use cases for lift analysis is when your marketing experiments are broken out geographically. For example, let’s say you have 3 markets: A, B, and C. Market A, you run $200,000 in billboard advertising, market B, you run $100,000 in billboard advertising, and market C, you do not run billboard advertising at all. Looking at year-over-year sales data by market, you should start to be able to arrive at some reasonably firm conclusions about how that specific experiment affected the sales in each market.
Conclusion
In conclusion, online attribution may have gotten harder over the years, but it is certainly not dead. Many marketers are not even spending enough to need to worry about attribution, and can confidently rely on platform specific tracking. Those that hit the point where attribution becomes important should attempt to track as much as possible and then use attribution estimates and lift analysis to fill in the gaps where needed. While these solutions are not perfect and will not allow for perfect mathematical precision, they are certainly preferable to throwing out the baby with the bathwater and relying 100% on lift analysis.