Response Advisors Forum: What’s the Attribution Solution?1 Jun, 2016 By: Thomas Haire Response
As consumers exert control over the sales and marketing funnel, marketers are seeking out new and better ways to attribute new customers to the proper medium in a multichannel campaign. Members of the Response Advisory Board take on some of the key questions.
Consumers are driving marketing more than ever, responding to marketers’ messages wherever, whenever, and however they prefer. This has created a major buzzword for marketers looking to quantify return on investment (ROI) on multichannel campaigns: attribution.
“It is critical to honor the duality of attribution,” says Doug Garnett, founder and CEO of Portland, Ore.-based Atomic Direct. “Attribution can be used to dramatically increase the performance of your media spending. It can also lead you into wasteful media spending.”
The challenges of attribution differ based on the type of campaign: is it for a hard good, including a drive-to-retail element? Is it a service with a lead-generation drive to the web? And, as campaigns have grown into multichannel behemoths, different attribution models continue to sprout.
Early in the e-commerce era, the battle of attribution was between “first-click” (the first source of marketing that drove response that became a sale) and “last-click” (allocating the sale to the most recently tracked visit) models. Now, “evenly weighted” (giving credit to all drivers across the board that resulted in a sale) or “any click” (sale attributed randomly to a referring source) models, along with other custom attribution models, are gaining more relevance.
“One of the bigger discussions we’ve had is around ‘building or buying’ your attribution model,” says Fern Lee, CEO of New York-based THOR Associates. “While building custom models may work for larger agencies or clients that have the capability internally, we prefer to partner with one of the larger attribution companies for several reasons: they can apply their knowledge to determine the most appropriate model and strengthen it for your brand right out of the gate; they have the research and other infrastructure to stay ahead of the curve when it comes to refining and enhancing the models; and they can typically more easily aggregate data coming from several different tactics and platforms for analysis.”
While Lee clearly believes in the vendor model for attribution, a major expansion in the number of those vendors — many offering a “breakthrough solution” — has created more questions than answers, for many.
“The fact remains that the rules of consumer engagement continue to change,” says Peter Koeppel, founder and president of Koeppel Direct in Dallas. “As an industry, we need to remain nimble and open. No one has all the answers and anyone who claims they do should be regarded with skepticism.”
Garnett adds, “Don’t make the mistake of turning your decisions over to attribution the way some businesses try to turn anything hard over to big data. Our job is to consider the gathered data and research. But, more importantly, our job is to make decisions using instincts and judgment. Attribution, and all of its promise, changes none of that.”
Read on to find out what Garnett, Koeppel, Lee and other members of the Response Advisory Board think about these challenge presented by measuring and monitoring attribution in this second quarterly edition of the 2016 Advisors’ Forum. (This extended online version features the full and complete answers from each board member who responded.)
When you hear the word “attribution” in today’s marketing climate, what does it mean to you?
Abed Abusaleh, Havas Edge: It means the ability to assign responses to the proper source, regardless of the channel it gets generated from.
Doug Garnett, Atomic Direct: Attribution is exciting and has always been a critical part of our DRTV campaigns. At the same time, attribution is dangerous business.
Marketing is complex, and we need to know everything we can about the consumer’s path to purchase. Attribution leapt to prominence because big data has become widely available and marketers hope data reveals all realities about how consumers buy.
Yet this year’s panels at Response Expo revealed that data’s value (and clarity) is more limited than hoped. Yes, we can learn where someone finally bought the product. But we can only track that entire journey if they are the one in a million whose journey takes place entirely online and with a single device. Even with extensive data, we are still making guesses at what happens with most purchases.
Let me add this caution: attribution isn’t risk free. Errors in attribution can cause us to make poor marketing decisions. How’s that? Attribution’s twin weaknesses are that: (a) it gives us numbers; but (b) its numbers can only ever reflect a tiny proportion of what actually happens in the market. The problem is that numbers easily dominate decisions, regardless of whether the numbers are meaningful. And that error leads to market failure.
Go ahead and embrace attribution, but also remember that what isn’t revealed by attribution can be far more critical than what is revealed.
Peter Koeppel, Koeppel Direct: Despite the ever-changing methods and demands of acquisition marketing, the core question that defines attribution has remained the same: what am I earning from each dollar invested? In today’s multichannel world, we have to look at all available data — including actual lead and sales data, Nielsen, time stamping, geo-tracking, and several other data points. Whereas before a media buyer could look at the data and optimize accordingly, today we employ data scientists who take a much more sophisticated approach.
Fern Lee, THOR Associates: The consumer’s journey to purchase or to take action looks like a flight plan given all of the channels, tactics, and messages consumed — as well as the devices utilized — in the process. Based on this thought, “attribution” takes into account all of these influences and the engagement path that prompts a consumer to take an action — and gives credit to each channel.
Bill McAlister, Top Dog Direct: The No. 1 problem with attribution is that the modern campaign requires a multichannel approach and that, in itself, makes it harder to quantify the cumulative effects of a campaign as many different online efforts expose our customer to our message. What results in the final message that ends in a sale? How do you value the other exposures that contributed to the sale? In our current campaign for Spray Perfect, we are running a robust social media campaign, PR campaign, TV, and traditional search, display, and e-mail campaigns. All reach millions of young, fashion-conscious female consumers. We are hitting our customer from all sides, and that is culminating in tremendous sales. If we decided to pull one block from our Jenga marketing mountain, would it come crashing down? We are not going to gamble on that, because our retail marketing partners are too important.
Rob Medved, Cannella Response Television: It means the same thing it always has: give credit where due. The reality is, attribution is only as good as the model and rules that manage the data, and today we have more data and more marketing outlets than ever.
Richard Stacey, Northern Response Intl. Ltd.: When I hear the word “attribution,” I think of something that sounds important but for which many people have different definitions. As a product marketer, unless you’re sitting with retailer point-of-sale (POS) data from Walmart’s retail link system and hooked into Amazon’s Seller Central database, any conversation about attribution can be challenging. You can’t manage what you don’t measure. You want to avoid driving traffic to all sorts of places you are not properly measuring — or that you’re measuring but your customer doesn’t actually buy from.
So we start by looking at where people are actually buying today. Close to 50 percent of all product-related searches on Google now go through Amazon. This idea of starting “www.myproduct.com” and somehow driving traffic to it is an old model now.
What’s the top challenge marketers face in attributing sales (or leads, as it may be) to specific marketing messages?
Abusaleh: In 2002, we saw about 15 percent of all responses generated via web. That number is closer to 50 percent to 55 percent today. The biggest challenge is identifying the origin of web traffic, especially when it’s driven to a corporate site that acts as the catchall for activity driven by multiple channels.
Garnett: The biggest challenge in attributing action to messages is that we are using observed behaviors. And it’s a classic problem of market research, because we don’t know what we didn’t observe — and what’s missing may be far more important.
For example, when I was a boy in the 1960s, paleontologists were certain that dinosaurs were lizards that lived alone and had scaly skin. Now, 50 years later, we know many were ancestors of birds — some had feathers, some lived in groups, and some cared for their young.
What changed wasn’t the quality of the scientists — they were all excellent. What changed is that we observed different data. And in attribution, as with dinosaurs, the discovery of one single bit of new data could radically change what we think is going on.
This means attribution should never rely on data alone. It is only one piece of the marketing puzzle. Observed data can help understand the impact of messages and mediums. But you’ll need traditional research (focus groups, individual interviews, surveys, retail data, social media conversation tracking, etc.) to build out the real picture.
There’s a second big challenge that is overlooked — and it’s one connected with the statistical analysis of data. Namely: statistics identify big things but underrepresent critical smaller things.
In DRTV, this leads to a specific challenge. We know that low-cost airings are often the most cost effective. Yet, when using statistical analysis in your attribution model, the impact of these airings is always under-represented. Therefore, statistical evaluation of DRTV media will tend to reward underperforming larger airings and lead your attribution model to recommend wasteful spending.
Koeppel: A significant challenge is definitively identifying the connection between a lead that originates from offline media and tracking that prospect to a sales event, which is most commonly occurring online. Various attribution models are evolving, and what exists today is more directional than absolute. However, just as the smart TV is improving our ability to track offline TV exposure to online transactions, perhaps new payment methods, such as Apple Pay and Samsung Pay, will bring greater tracking to offline retail campaigns.
Lee: There are two primary challenges when attributing in a multichannel campaign. First, there’s the cost of entry to implement a cross-channel attribution model. There are several companies that have tackled the in-channel attribution challenge, but few that are able to implement a cross-channel model, and it is an enormous expense to many advertisers without the needed KPI results.
Second, ensuring the data is actionable — it is imperative that if a brand has made the investment to implement a cross-channel attribution model, there is a data analyst who is able to read and work with the marketing/media/creative teams to take learnings and apply them to future initiatives for maximized results and return.
Medved: The challenge is attributing revenue to each piece of media at the proportionately correct percentages. An order from a DRTV spot to an 800 number is clear. A click-thru online order from an online ad is clear. However, the online order that comes from organic search or direct-to-web becomes unclear. Google AdWords, couponing, custom landing pages, pay-per-click (PPC) tied to retail geotracking, and many other methodologies have helped marketers refine their model.
Stacey: The No. 1 challenge for marketers is to look across all the channels and get the proper measurement tools in place. That means media staff attending Walmart University and training on Amazon Seller Central and Facebook measurement tools. Without understanding these platforms — and measuring them and tying them back to your media spend — it’s hard to know how effective your advertising actually is. For our company, we find that’s where most of the sales are these days — not through the 800 number or a specific product website.
How do attribution challenges change based on the type of product or service being marketed?
Garnett: The more a campaign lives online, the more likely attribution is to accurately reflect a larger portion of your spending. But that is also limiting.
If we choose to adopt media based on how easy it is to measure, then our campaigns will always have limited impact. Typically, the most effective media to drive a big result isn’t the easiest to measure. Just so, a drive-to-retail campaign will thrive best with TV as the engine — even though TV can’t deliver those reams of consumer action data that make us feel better. But TV is the single best driver of retail action available.
Koeppel: The challenges of attribution for different products and services ultimately come down to consumer behavior and the ability of the marketer to track it. For example, a drive-to-web campaign will typically drive a reactive, near-term response that can be identified and attributed to a specific marketing event. One pitfall, however, is that the consumer path is not completely played out yet, so marketers shouldn’t react to this too hastily. At the other end of the spectrum, response from a drive-to-retail campaign can be significantly less reactive, and tying consumer actions to specific marketing events in a delayed-response environment can be agonizingly difficult. Lessons can be taken from both scenarios, and a balanced approach between attributing reactive and long-term benefits from marketing must be reached to overcome these challenges.
Lee: Drive-to-retail campaigns differ in the sense that you have to account for media and messaging that drove store traffic, in addition to the trackable online sales, and give credit to the tactics that drove each. When generating an online response or usage, the engagement path is trackable and the brand is able to give credit to each of the tactics and messages that influenced it.
Medved: The response is intrinsic to the product and price point. In general, the more complex the sale or the higher the price point, the more complex the attribution. The farther a marketer gets from a direct response, the more challenging that attribution becomes. Retail pull-through can look at couponing or store lift in online or offline ZIP-code-targeted budget pushes. Google AdWords can also estimate store visits based on mobile location data tied to store locations with your product.
Stacey: My business is mostly about selling hard goods, but selling services would follow similar principles. You need to be clear on your campaign objectives and clearly understand all the new platforms that are out there now. I was sitting with a group of DRTV industry people not long ago, and only a few were active on social media. Some had a social media listing, but only a handful knew how to make money from such platforms. So the first step for some is to go back to school and really learn this stuff. Simply having a Facebook page does not make someone an expert in social media marketing.
Which attribution model — first-click, last-click, evenly-weighted, or any-click — works best for performance-based campaigns driving leads/sales from both online and offline media?
Abusaleh: The first-click attribution method is the one used most effectively in order to accurately assess the effectiveness of particular media sources. This ultimately leads to the most optimized multichannel media buy. The evenly-weighted approach is the approach that should be used to evaluate the effectiveness of a particular channel — TV, search, radio, print, etc.
Garnett: You should not trust your attribution models until they have been challenged and evaluated against your data, your consumers, and in your business situation. Every situation deserves its own analysis to determine the best model. Vendors who trot out the “rules” without considering your situation will hurt your profits — just as those DRTV producers who tell you to produce “using the formula” are doing a disservice to your profits.
Koeppel: Every attribution method has its strengths and limitations. Multi-touch attribution allows a marketer to understand how individuals interact with various marketing channels, on a micro level, and determine which are lead generators, nurturers, and converters. Marketing mix modeling provides insight into the high level contribution of specific marketing channels, as well as macro-level factors that may not be identified in multi-touch attribution models. Ultimately, the best approach is to leverage the benefits of more than one method and utilize the insights derived from each, within the spectrum of their respective strengths.
Lee: Custom models are the best, as those models are developed based on the specific ask the brand is making, including key performance indicators (KPIs). While a custom model may take significant time and money to develop and refine, the long-term benefits and learnings will outweigh these challenges. A more complex model will also be able to grow as the media and marketing mix is adjusted based on what you’ve learned. Incorporating factors — such as latency between interaction and conversion, custom weights based on the type of message/call-to-action (CTA) and device usage — will build the strongest model.
Medved: First-, last- or weighted-click-based attribution was developed for digital marketing without regard to offline marketing communications. The debate over which method is better is a matter of opinion, but if translated into terms of TV, the last click is the most analogous to a TV toll-free number. No DR marketer has the ability to measure the sales impact of a viewer’s first exposure to a TV spot. They associate the sale with the most recent airing, which includes drag orders. None of these methods, however, account for multiple marketing touchpoints across media channels — which are the issues everyone is trying to resolve.
Stacey: The best model depends on your campaign objectives, what you’re trying to sell, and to whom and where — measure the wrong things and you’ll get the wrong outcome. For example, if you’re selling a product you need to be including Walmart POS data and Amazon data. If you’re selling a service, you might be measuring more alternative social platforms. So the best model is unique to each campaign objective. Platforms like Facebook are especially effective because they allow you to test and then scale with targeted marketing. However, staff needs to be trained to do this properly. Marketing on Facebook is a little more complicated than posting your friend’s birthday party pictures.
The attribution challenge seems to have driven a major expansion in vendors offering “breakthrough” data to help marketers and agencies best measure campaign attribution. What types of data are most crucial in this new era?
Abusaleh: Rentrak provides minute-by-minute viewer levels, using approximately 15 million homes as the sample size. This is a necessary component of any effective attribution model used to attribute web orders driven by television.
Garnett: This is the area where I offer the biggest and loudest caution. In my experience, these “black boxes” are the single most dangerous area of attribution. I’ve worked with a few — and each included vast assumptions that turned out wrong answers at a very high rate. Even in a panel I moderated at Response Expo 2015, both Vistaprint and Wix talked about the difficulty they’d had trying to find an attribution vendor model that worked correctly.
In the end, I have heard a lot of promises of magic solutions. But attribution is hard and complex — and it needs to be. If it ever seems too easy, it probably means we stopped thinking deeply about this very subtle set of issues.
Koeppel: The ability to track a single individual across multiple devices, marketing channels, and funnel thresholds is one of the most valuable capabilities provided by these vendors. This allows marketers to go beyond attributing aggregate value and truly understand the traits of their respondents. Information such as demographic insights, behavior traits, media consumption, interaction with the sales funnel, long-term interaction with your product or service — the list goes on. This allows one to not only have a greater understanding of the individuals to whom they successfully market, but also identify other correlating factors, whether macro or micro, that may not have otherwise been available for consideration.
Lee: Extremely critical factors include: how each message or tactic influences another along the consumer’s engagement path to the actual sale or action; device usage (how, when, and why they’re used in the conversion process); ease of analyzing the data and applying it to future buys; and analyzing the long-term value of media and marketing. This information should also fuel customer relationship management (CRM) and drive how those communications are being served based on knowing the path the consumer took for the initial engagement with the brand.
Medved: The various methodologies of attribution can be useful for different marketing needs. Media mix modeling (MMM) is a probabilistic attribution methodology widely used by marketers to determine how to best allocate a marketing budget. It helps them determine what investment mix will deliver the greatest bang for their budget. It doesn’t do a great job at identifying the granular decisions that a direct marketer is accustomed to making. MMM is a great planning tool but not ideal for day-to-day media decisions. Many service providers have built their own probabilistic methodologies in an attempt to optimize singular media decisions, and no one has developed a bulletproof model. These are good tools, but they should be used as directional information and in concert with deterministic methods.
Stacey: There seem to be a lot of attribution models and approaches, with each one claiming to be better than the other. However, I find it’s often not so much about accuracy as it is about what to measure and what not to measure. I’ve seen some sophisticated models that measure exactly where a telephone order came from — but as fewer people order by phone, who cares? Maybe we should be looking at how many people clicked the barcode SKU through the cash register at Walmart Store 757 last night. We can actually measure that in our office. The technology is there, but few marketers are using it yet to its full advantage.