Pick an analytics approach that fits your business situation

Start by taking stock of your industry and the situation of your business. How much data do you have? How far does it go back, and how granular is it in terms of time periods and instruments? Do you collect and store this data yourself, or does someone else further down the value chain own the customer relationship? Is your business model driven by strategic long shots and brand strength, or by tactical communication and millions of small-ticket transactions? What is your primary marketing objective – attention, consideration, or sales stimulation? Refer to the insert below for guidance on how to pick the analytical approach that best fits your needs.

• Marketing and media mix modelling: Often referred to as econometric marketing mix modelling, or MMM for short, this technique uses detailed historical data and time series regression analysis to quantify the impact of marketing activities on sales or other dependent variables, such as customer acquisition or lead generation. MMM allows you to adjust for external factors – such as weather or seasonality – to make sure the marketing function is only credited with what it really contributes to business success.
• Digital attribution:⁹ User-level journey analytics and attribution modelling (often referred to as digital attribution, or DA for short) is a technique that determines the influence of a single touch point on marketing success. This could be, for example, a rich media ad that a user is exposed to prior to the completion of an online transaction. Over the course of a purchase decision journey, a consumer makes contact with a brand at multiple touch points. By analysing many such journeys, DA allows you to do justice to all touch points on the journey and pinpoint the most relevant ones.
• Consumer survey: There are various approaches. We often use consumer decision journey research, or CDJ for short. This technique uses questionnaires to take a snapshot of touch point importance as stated by respondents and determine the impact of marketing investments at specific stages of a consumer's decision journey. The emerging fact base helps you focus your investments on the instruments that have the biggest impact on consumers at crucial stages of their decision-making process. CDJ will give you an understanding of the role of non-owned marketing, such as editorials, news accounts, or word of mouth. CDJ also allows for a segment-level assessment of touch point effectiveness, using typing tools in the research.

Use econometric modelling if you have sufficient historical data

Econometric modelling is a well-established, proven approach. Its origins date back to the era of George Gallup, the pioneer of polling, who applied statistical analysis to forecast the outcome of the 1936 US presidential election. In the 1950s, large manufacturers of branded consumer goods – spearheaded by Procter & Gamble – started applying econometric models to marketing investments, driven by the aspiration to make the most of their sizable advertising budgets.¹¹ This technique employs various types of regression modelling to quantify the statistical correlation of marketing activities with sales or other marketing performance indicators (Exhibit 6.1). It uses historical data – ideally on a weekly or monthly basis – in areas such as sales, advertising investment, competitive advertising intensity, and macroeconomics. The longer the period and the higher the number of data points, the more reliable an econometric marketing mix model will be. Most companies use two to three years of weekly data for marketing mix modelling. This is sufficient for very reliable calculations, but leading consumer goods players sometimes use even longer periods. Note, however, that very long periods may bring fundamental changes in market conditions (overall economy, new players in the market, new channels, new tactics) that can be hard to account for.

Econometric modelling is based on the assumption that the future will resemble the past, at least with respect to the purported causal relationships between investment and success. It treats past data series as experiments. Since the intensity of marketing activities changes over time, regression analysis can help estimate the impact of a given activity on the dependent variable, such as sales. Specific benefits include:

• Quantified contribution of individual instruments (e.g., TV advertising, direct marketing, digital display advertising) to sales (or another performance metric). Simply speaking, this tells you which investments in your mix work and which ones don't.
• Calculation of “response curves”, i.e., the diminishing return relationship between marketing investment and sales for each instrument. This lets you calculate the marginal ROI of an extra dollar invested in a specific instrument.
• Calculation of real ROI – both average and marginal – for the budget as a whole and for each instrument in your mix. This enables you to treat marketing activities as investments rather than as cost positions.

The output of a state-of-the-art MMM tool helps advertisers separate external factors – such as the overall market growth or demographic trends – from the levers they can address themselves – such as promotional strategy or advertising investment. MMM also provides revenue impact measurements across all marketing levers in order to assess their relative business contribution. Ultimately, MMM provides marketing managers with the means to investigate the likely consequences of their actions before they act, enabling them to make fact-based decisions, instead of relying on intuition.

In theory, econometric modelling is simple enough. If the ROI for a given instrument is higher than one, increase your investment. If it is lower than one, decrease or discontinue your investment. In practice, there are a number of pitfalls you need to steer clear of. A lot of first-time users of econometric modelling focus on short-term sales or new customer acquisition as the dependent variables. This will make instruments that primarily drive brand equity and customer loyalty look bad. So, make sure to integrate indicators of long-term effects as well, such as brand health, net promoter scores, and customer retention, especially if long-term customer relations are a success factor in your industry (Exhibit 6.2). But don't let the complexity get out of hand. Start with the two or three KPIs that matter most.

Of course, getting the input data right is equally important to the reliability of the marketing ROI calculation. Include all relevant marketing instruments in your model – paid, owned, and earned. Econometric models built by media agencies are often limited to paid media – such as TV and print advertising – as influencing factors, simply because these are the only instruments for which they have reliable data. But to reflect the fact that consumers' purchase decisions are influenced by many other factors, econometric marketing mix models also need to incorporate investments in owned media – such as a company's website – as well as the viral effects of social media. In a tragic case of negative online buzz, a company suffered losses exceeding EUR 30 million that could have been avoided by investing less than EUR 1 million in social media marketing. Compare our brochure Turning buzz into gold¹² for details.

Finally, a robust econometric marketing mix model should also consider and account for business drivers outside marketing communication, such as new product introductions, pricing, promotions, changes in distribution, interest rate fluctuations, major events, seasonality, holidays, and weather. Sounds complex? It isn't. In fact, it's easy to get started. Most of the data you need to build a basic model is probably already in your system, buried in your media planner's hard drive, or readily available from your media agency. Enlist the services of experts who understand both the essentials of statistics and the principles of marketing and you're good to go.

Apply attribution modelling to the digital instruments in your mix

Digital marketing provides companies with an almost infinite number of new instruments to reach consumers in highly targeted ways. At the same time, it is also a source of massive data on what works and what doesn't work, down to the level of individual users. Digital attribution modelling takes advantage of this wealth of data to help you make the most of your investment in new media. It can also be used to automate decision making in areas such as digital ad buying, using adaptive bid management and ad serving tools. A few advanced players are already managing their entire marketing budget with the help of digital attribution modelling. For others, it is an important building block in holistic marketing ROI optimization. Today, only about 20 percent of marketers use multichannel attribution models that cover all touch points. Some 40 percent use some sort of basic model, such as the attribution of an event to the last click preceding the event, or to the click that started the customer journey. The remaining 40 percent have no attribution model in place at all.¹³ Clearly, there is a lot of room for improvement. As a rule of thumb, any company spending 20 percent or more of their marketing budget on digital instruments should look into some form of attribution modelling.

Attribution modelling uses data logged along the digital customer journey – such as ad views, searches on Google, or a click on a link in an e-mail – to determine the contribution of a specific touch point to marketing success (Exhibit 6.3, illustrative). A touch point can be any exposure to marketing communication – be it paid or unpaid – and success can be anything from a sale of a product or a subscription to a service to signing up for a newsletter or registering as a user. Once you understand which touch points have the biggest impact on the things you are trying to achieve in the marketplace, you can increase ROI by shifting investments to those high-impact touch points.

Off-the-shelf online tracking tools – such as Google Analytics or Adobe Analytics – provide a good starting point for attribution modelling. Utilizing a user's last click prior to a desired action – such as a purchase – as a proxy, you can rank competing instruments such as referrals, paid search, and organic search in terms of their relative contribution to your business objectives. Reallocating funds from low-performing instruments to high-performing ones will increase ROI immediately, if only on a small scale. Once you have mastered optimization within a given group of touch points – such as search – you can apply the same principles across different types of instruments, e.g., search engine marketing versus display advertising versus social media marketing. This works best if you compare activities that serve the same objective, such as sales generation. Last click attribution is less insightful if your aim is to compare the impact of different types of campaigns, e.g., awareness creation versus sales stimulation. To keep complexity at bay, respect the 80/20 rule and start by applying attribution modelling to the biggest positions in your digital marketing budget. You can always get more comprehensive and more sophisticated over time (Exhibit 6.4).

Scientific as it sounds, attribution modelling is not actually an exact science – at least not in practice. There are a number of pitfalls you need to watch out for. These include faulty data, misleading dependent variables, and cross-attribution issues. A recorded ad impression, for example, doesn't necessarily mean that the ad in question has actually been seen by the user. The user's attention may have been focused elsewhere on the screen, which is why an ad view is typically defined as an ad that has been displayed for at least one second in the viewed area of a website. Tests show that users often see less than 50 percent of all ads a company pays for. As far as dependent variables are concerned, a lot of specialized tools optimize attribution based on metrics that may not be in synch with your overall attribution logic. For example, bid management tools typically optimize for the price and position of an ad. But if your overall objective is converting considerers to buyers, the tool will skew your marketing mix towards instruments that don't actually do what you want them to do. Finally, it is anything but trivial to connect and weigh the relative contribution of multiple touch points and devices to a given decision journey. For example, a single user may see a YouTube ad on a tablet, then Google the featured product on a laptop, and proceed to make the purchase on a smartphone. Integrating these events into a consistent dataset and ascribing a meaningful contribution to each of them takes experience and expertise.¹⁴

A final word of caution: by definition, digital attribution modelling only recognizes digital touch points. But in reality, classical advertising, other offline activities, and a wide range of external factors contribute to a user's decision-making process. This is why attribution modelling systematically overstates the impact of digital touch points, even when the real driver of a given sale may be an offline instrument – or the fact that your competitor's website was down when the user was trying to buy from them. Unless you spend your entire marketing budget online, make sure to conduct the required sanity checks and integrate attribution modelling with a more comprehensive approach that covers all instruments, such as Reach-Cost-Quality or econometric modelling.

Consider survey-based approaches – such as consumer decision journey modelling – if you don't have sufficient historical data

Consumer decision journey modelling – or CDJ for short – is based on the assumption that marketing investments are most effective when they reach consumers at the moments that most influence their decisions. CDJ uses consumer surveys to determine what these moments are, and the surveys are designed to recognize the fact that marketing is evolving from proclamation (marketers talking to consumers) towards conversation (consumers talking to each other and providing feedback to companies). Specifically, word of mouth and the recommendations consumers make to their friends feature prominently in state-of-the-art CDJ approaches. This is why CDJ attaches special attention to the post-purchase phase – often referred to as the loyalty loop – in recognition of the need to provide an aftersales experience that inspires loyalty and repeat purchases. This helps marketers ensure that their investments are not overly skewed towards instruments that may trigger short-term sales at the expense of lasting bonds between consumers and brands.

Like the brand purchase funnel discussed in Chapter 3, CDJ is a way of modelling a consumer's decision journey. CDJ models typically work with four primary phases that represent potential battlegrounds where marketers can win or lose: initial consideration; active evaluation, or the process of researching potential purchases; closure, when consumers make a purchase; and post purchase, when consumers experience the product they have bought or the service they have signed up for.¹⁵ More recently, the CDJ model has been updated to account for changes in consumer behaviour, chiefly to account for the fact brands today can not only react to customers as they make purchasing decisions, but also actively shape those decision journeys. Shaping the decision making experience itself with content, interaction, and advice – rather than sending messages to consumers at specific stages of their journeys – is becoming a source of competitive advantage. A recent survey conducted in the US¹⁶ revealed that top performers understood the entire customer journey much better than their peers (20 percent versus 6 percent), and had much better processes for capturing insights about customers and feeding them back into their marketing programmes to improve performance (30 percent versus 11 percent).¹⁷

Once the contribution of the various milestones along a consumer's decision journey has been determined with the help of standardized surveys, marketers can start optimizing their investments accordingly – for example, by shifting funds from touch points that generate attention to those that help provide customers with a more satisfying aftersales experience. In the past, most marketers consciously chose to focus on either end of the journey – building awareness or generating loyalty among current customers. CDJ enables you to be much more specific about the touch points you use to influence consumers as they move through initial consideration to active evaluation to closure. As a car insurance company, for example, you might want to reduce your investment in advertising and increase your investment in touch points such as claims management to account for the fact that the post-purchase loyalty loop triggers 78 percent of all purchases in this industry (Exhibit 6.5).

The principal advantage of survey-based approaches is that they work without historical data. So if you don't have sufficient data on past investments and sales, or have reason to doubt its reliability and consistency, CDJ is a great alternative to assess the relative contribution of touch points to marketing success. The limitation of CDJ is twofold: any survey-based approach is just a snapshot derived from an observation at a single moment in time and it depends on the reliability of consumers' self-knowledge as far as the stated importance of specific milestones and experiences along their decision journey is concerned. For the latest thinking on new sources and techniques to compensate for these limitations, compare our upcoming publication on the future of insights.¹⁸

Build a hierarchy of analytical approaches to avoid conflicting results

Many providers of advanced analytics will tell you that their approach – be it econometric, attribution-driven, or survey-based – can help you manage your entire marketing budget. In our experience, that is rarely true. There is no single approach that works irrespective of which industry you are in, what your marketing mix is like, and how much data you have. At the other end of the spectrum, there is an almost infinite number of highly specialized solutions that will let you optimize a handful of touch points – or a single one – to a tee, but neglect the bigger question of roughly how much you should be spending on that particular instrument to begin with. In either case, you will create considerable complexity, produce inconsistent results, and miss out on part of the opportunity. What is even more important, there is high risk of frustrating your team and confusing your fellow executives with a myriad of metrics and conflicting recommendations.

To avoid this, we encourage you to establish a clear hierarchy of analytical approaches and pick your battles wisely. For example, the head of marketing for a global retailer, thrilled by the lure of big data and the prospect of scientific precision, spent months on attribution modelling to optimize the tiny fraction of the budget that was actually dedicated to digital response marketing, such as online vouchers. At the same time, the millions that went into old-school promotions and local leaflets were woefully neglected. As the CMO, you owe it to yourself and to your organization to approach mix optimization systematically, and to focus your attention on the biggest opportunities. In our experience, econometric modelling works best to determine the appropriate high-level mix between offline and online instruments. If you don't have sufficient data, you may want to use Reach-Cost-Quality (see last chapter) or survey-based models instead. Once you have established this overall frame of reference, you should try out more sophisticated techniques to optimize key instruments within the bigger buckets, such as attribution modelling, to determine the most effective touch points for search engine marketing.

And even if you have all the data in the world, don't feel like you have to go for the most sophisticated models on the market right away. If your organization isn't ready for data-driven decision making, you may end up wasting money and frustrating your team. Compared with manual mix decisions, even simple models can bring efficiency improvements of up to 20 percent. Don't buy a Ferrari if a Fiat can do the job.