Note the key phrase: Customer Data Platforms (CDP). Only the hottest mar-tech category to appear in at least a decade, and we literally wrote the book on it. At least, the first substantial mainstream book on this key topic. We cover the category from data integration and identity management to exploration, activation and A.I.
It’s accessible and a quick read, full of charming illustrations and architecture drawings. If you’re interested in this fast-growing tech category that points the way toward the converged platform future, our book is a great place to start.
Happy launch-day, my friends! Here’s to a more integrated, privacy-friendly and analytical 2021! The future is bright. peace mk
Loading and learning the models took about an hour. And given a few-word prompt, GPT-2 happily takes about five minutes to churn out 20 “ideas,” without breaking for lunch.
Text generation – or robo-writing – has made startling leaps in the past few years, moving from punchline to something that may deserve a seat in the creative lounge. Believe it or not, the best robo-writers are almost the equivalent of that most annoying/wonderful phenomenon: the eager beginner, completely inexhaustible but creatively uneven.
Most of GPT-2’s “ideas” were not quite ready to be presented; some were nonsensical. Oddly, it had no clue what to do with the prompt “Jason Alexander.” And one of its “Wow no cow” completions was “Wow no cowbell can be quite like the best in the universe.”
Which is probably true and not at all helpful.
In the near future, the smartest creative teams will be those that can use A.I. writers in productive ways, as a computer assist to a creative session and a source of ideas that might spark better ones.
GPT-3’s largest setting has 175 billion parameters. Think of these as individual knobs that the model has to tune, based on human writing samples, in order to predict the next word in a sentence. Google just open-sourced an even larger text model called Switch Transformer that reportedly has more than 1 trillion parameters.
The human brain has about 3 trillion synapses. Let’s leave that right there.
GPT-3 takes GPT-2 out of the sandbox and sends it to middle school. Give the model a prompt (e.g., “Once upon a time…” or “It wasn’t me…”), and it can continue at some length, generating text that is often plausible and sometimes uncanny. Early testers ranged from awed to skeptical – and often both.
In fact, that’s what makes this new generation of robo-writers different: they are flexible models, verging into the space called artificial general intelligent (AGI). This is the kind of intelligence we have: not pre-trained in any particular discipline but capable of learning. GPT-3 seems to perform well on a range of language tasks, from translation to chatting to impressing electronic gearheads.
Ad copywriting isn’t such a leap. As a tool to build creative prompts from catch phrases ready for human filtration, so-called Transformer models make a lot of sense.
Robo-writers are already employed in shrinking newsrooms. So far, they’re mostly stringers on the high-school sports reporting, weather and stock market desks — churning out endless Mad Lib-style pieces in routine formats about games and finance that no sane journalist would want to write, even for money.
Computers thrive at tedium. It’s their métier. The for-profit OpenAI takes a brute force approach to its innovation. Two years ago, it gained notoriety for developing a machine that could beat the world’s best players of a video game called Dota 2. It did this by having software agents play the equivalent of 45,000 hours of games, learning by trial-and-error.
The GPT family of tools were also developed by pointing software agents at a massive corpus of data: in GPT-3’s case, millions of documents on the open web, including Wikipedia, and libraries of self-published books.
GPT-3 is exposed to this mass of human-written text and builds a vocabulary of 50,000 words. Its model’s weights predict the next word in a sequence given the words that came before – and develops meta-learning beyond simply memorization. It requires a prompt and can be guided by sample text that provides a “context,” per the Super Bowl examples above.
To answer the obvious question, it can be shown that GPT-3’s creations are not just plagiarism. But are they pastiche? The model learns to predict words based on its experience of what others have written, so its prose is predictable by design. But then again, isn’t most writers?
Limitations include a “small context window” – after about 800 words or so, it forgets what came before – so it’s better at short pieces. It has a short attention span, matching our own.
For this reason, people who have spent more time with the model grow less impressed. As one said: “As one reads more and more GPT-3 examples, especially long passages of text, some initial enthusiasm is bound to fade. GPT-3 over long stretches tends to lose the plot, as they say.”
But ad copywriting isn’t about sustaining an argument or writing a book. It’s about trial and error and inspiration. Increasingly, that inspiration may be coming from a robot.
This post originally appeared on the blog of the NYC Data Science Academy, where I am a student. We were assigned the Ames Housing dataset to practice Machine Learning methods such as Lasso, Ridge and tree-based models. I added the Plotly Dash out of sheer exuberance.
The Ames Housing dataset, basis of an ongoing Kaggle competition and assigned to data science bootcamp students globally, is a modern classic. It presents 81 features of houses — mostly single family suburban dwellings — that were sold in Ames, Iowa in the period 2006-2010, which encompasses the housing crisis.
The goal is to build a machine learning model to predict the selling price for the home, and in the process learn something about what makes a home worth more or less to buyers.
An additional goal I gave myself was to design an app using the Plotly Dash library, which is like R/Shiny for Python. Having been a homeowner more than once in my life, I wanted a simple app that would give me a sense of how much value I could add to my home by making improvements like boosting the quality of the kitchen, basement or exterior, or by adding a bathroom.
Conceptually, I figured I could build a model on the data and then use the coefficients of the key features, which would probably include the interesting variables (e.g., kitchen quality, finished basement). These coefficients can predict what a unit change in a feature would do to the target variable, or the price of the house.
One advantage of the Ames dataset is that it’s intuitive. The housing market isn’t particularly hard to understand, and we all have an idea of what the key features would be. Ask a stranger what impacts the price of a house, and they’ll probably say: overall size, number of rooms, quality of the kitchen or other remodeling, size of the lot. Neighborhood. Overall economy.
To set a baseline, I ran a simple OLS model on a single raw feature: overall living space. My R^2 was over 0.50 — meaning that more than half of the variance in house price could be explained by changes in size. So I didn’t want to overcomplicate the project.
The data had a lot of missing values. Many of these were not missing at random but rather likely indicated the feature was not present in the house. For example, Alley, PoolQC and Fence were mostly missing; I imputed a ‘None’ here. Other values needed special treatment.
LotFrontage was 17% missing. This feature turned out to be closely related to LotShape, with IR3 (irregular lots) having much higher LotArea values. So I imputed values based on LotShape.
There were a lot of categorical values in the data which I knew I’d have to dummify. Before going there, I looked into simplifying some of them. ‘Neighborhood’ seemed ripe for rationalization. It was clear (and intuitive) that neighborhoods varied in house price and other dimensions. There didn’t seem to be a pattern of sale price and sale volume by neighborhood (except for NorthAmes), but Neighborhood and Sale Price were obviously related.
Now I know it isn’t best practice to create a composite variable based only on the target, so I created a new feature called “QualityxSalePrice” and split it into quartiles, grouping Neighborhoods into four types.
There were many features related to house size: GrLivArea (‘Graded’ or above-ground living area), TotalBsmtSF (basement size), 1FlrSF, even TotRmsAbvGrnd (number of rooms). These were — of course — correlated, as well as correlated with the target variable. Likewise, there were also numerous features related to the basement and the garage which seemed to overlap.
To simplify features without losing information, I created some new (combined) features such as total bathrooms (full + half baths), and dropped others that were obviously duplicative (GarageCars & GarageArea).
Looking across both continuous and categorical features, there were a number that could be dropped since they contained little or no information; they were almost all in one category (e.g., PoolArea, ScreenPorch, 3SsnPorch and LowQualFinSF).
Finally, I dummified the remaining categorical features and normalized the continuous ones, including the target. The sale price (target) itself was transformed via log, to adjust a skew caused by a relatively small number of very expensive houses.
Data Exploration & Modeling
Everybody knows (or think we know) that it’s easier to sell a house at certain times of year, so I looked at sale price by month and year. These plots showed a peak in the fall (surprisingly, to me) as well as the impact of the 2008 housing crisis.
Correlation for the continuous variables showed a number aligned with the target variable, and so likely important to the final model. These included the size-related features, as well as features related to the age of the house and/or year remodeled.
Because my target variable was continuous, I focused on linear regression and tree-based models. I ran OLS, Ridge, Lasso and ElasticNet regressions, using grid search and cross-validation to determine the best parameters and to minimize overfitting.
The error functions for these models on the test data set were all similar. The R^2 for Ridge, Lasso and ElasticNet were all in the range 0.89=0.91. The significance test for the OLS coefficients indicated that a number of them were very significant, while others were not.
I also ran tree-based models Gradient Boosting (GBM) and XGB (also gradient boosting) and looked at Feature Importances. The tree-based models performed similarly to the linear regression models, and pointed to similar features and being significant. Of course, this is what we’d expect.
In the end, the most important features across all the models were: OverallQual (quality of the house), GrLivArea (above-ground size), YearRemodAdd (when it was remodeled), NeighType_4 (high-end heighborhood), and the quality of the kitchen, finished basement and exterior. If nothing else, the model fits our intuition.
Feeding the features through the model one by one, additively, it became obvious that the most oomph came from 20-25 key features, with the rest more like noise.
Plotly Dash App
Settling on a multiple linear regression as the most transparent model, I ended up with an array of 23 features with coefficients and an intercept. To use them in my app, I had to de-normalize them as well as the target.
The app was aimed at a homeowner who wanted to know the value of certain common improvements. The coefficients in my linear model were the link here: each coefficient represented the impact of a UNIT CHANGE in that feature on the TARGET (price), assuming all other variables stayed the same. So I just had to come up with sensible UNITS for the user to toggle and the impact on the TARGET (price) was as simple as multiplying the UNIT by the COEFFICIENT.
Since I was building an app and not trying to win a data science prize, I focused on the significant features that would be most interesting to a remodeler. Some things you can’t change: Neighborhood and Year Built, for example.
But some things you can: in particular, features related to Quality would be in the app. These were Basement, Kitchen and Exterior Quality. Other features could be changed with significant investment: Baths (can add a bath), Wood Deck (can add a deck or expand one), and even house size (since you can — if you’re crazy and rich — tack on an addition somewhere).
I also included a couple Y/N features since they could affect the price: Garage and Central Air.
I knew Plotly as a way to create interactive plots on Python, but the Dash framework was new to me. It was introduced to me by one of our instructors and an active online community and examples also came to my aid.
Dash is similar to R/Shiny in that it has two key components: a front-end layout definer, and a back-end interactivity piece. These can be combined into the same file or separated (as they can in Shiny).
The general pseudo-code outline of a Dash App starts with importing the libraries and components. It’s written in Python, so you’ll need numpy and pandas as well as Plotly components for the graphs.
Dash converts HTML tags into HTML when they appear after “html.”, which is straightforward. I included Div tags, H2 (headers) and P (paragraph) tags. Different sections of the Div are called “children” (as they are in HTML). I used “children” here because I wanted to have two columns — one for the inputs and the second for the graph of adjusted house prices.
The rows and columns can be done in a couple different ways, but I used simple CSS style sheet layouts (indicated by “className”).
Interactivity is enabled by the layout and the callbacks. In the layout, as in Shiny, I could specify different types of inputs such as sliders and radio buttons. There aren’t as many or as attractive a selection as you find in Shiny, but they get the job done. I used sliders and buttons, as well as input for the avg starting price.
The functional part of the app starts with “@app.callback,” followed by a list of Outputs (such as figures/charts to update) and Inputs (from the layout above). The way Dash works, these Inputs are continually monitored, as via a listener, so that any change to a slider or other input will immediately run through the function and change the Outputs.
Right after the @app.callback section, there’s one or more functions. The parameters of the function represent the Inputs from @app.callback, but they can have any name. The name doesn’t seem to affect their order, since they’re read in the same order as the callback Input list.
This function includes any calculations you need to do on the Inputs to arrive at the Outputs, and if the Output is a graph or plot — as it almost always is — there’s also a plot-rendering piece.
In my case, I wanted the output to be a normal distribution of potential new values for the sale price, given the changes in the Inputs. For example, if someone moved the “Overall Quality” slider up a notch, I needed to generate a new average saleprice, a difference (from the baseline) and a normal distribution to reflect the range.
I did this by turning each coefficient into a factor, adding up the factors and multiplying by current sale price. I then generated a random normal distribution using Numpy with the new target as the mean and a standard deviation based on the original model.
The final dashboard looked like this:
Granted, there are a caveats around the app. It’s based on the Ames housing dataset, so other areas of the country at different times would see different results. It requires estimating a ‘starting price’ that assumes the default values are true, and this estimate might be difficult to produce. But as a proof of concept, it has potential.
I think there’s definitely a data-driven way to estimate the value of some common improvements. This would help a lot of us who are thinking of selling into the housing boom and wondering how much we can reasonably spend on a kitchen remodel before it turns into red ink.
The following article first appeared on the Salesforce Marketing blog in January.
Last May, at the height of the COVID pandemic’s first wave, PepsiCo raised some bubbles by launching not one but two websites where consumers could browse and purchase a selection of the company’s more than 100 widely munched snack and beverage brands. At a time when many outlets were closed and people ordered more food online, the company quickly made it easy for consumers to buy directly.
One ecommerce site, PantryShop.com, took a lifestyle-themed approach, offering bundles of products in categories such as “Rise & Shine” (Tropicana juice, Quaker oatmeal, Life cereal) and “Workout & Recovery” (Gatorade, Muscle Milk, and Propel electrolyte-infused water). A second site, Snacks.com, offered a more straightforward lineup of Frito-Lay brands such as Lay’s, Tostitos, Cheetos, dips, nuts, and crackers.
These platforms complement PepsiCo’s retailer channels to ensure the company continues to deliver consumers their favorite products on the right platform, in the right place, at the right time.
Whenever, wherever, however has been the mantra of ecommerce digital marketing for a while, but it’s become more important in the age of COVID-19.
Whenever, wherever, however has been the mantra of ecommerce digital marketing for a while, but it’s become more important in the age of COVID-19. Most of Salesforce’s customers – particularly those in high-velocity consumer categories such as consumer packaged goods (CPG), retail, and restaurants – have had to become extraordinarily flexible over the past 10 months to adapt to ever-changing local guidelines and consumer behaviors.
What was most striking about PepsiCo’s foray into direct-to-consumer (D2C) commerce and marketing was its speed. “We went from concept to launch in 30 days,” said Mike Scafidi, PepsiCo’s Global Head of Marketing Technology and AdTech. “Within 30 days, we stood up our ecommerce capabilities and started delivering direct to our consumers.”
PepsiCo’s products are consumed more than a billion times daily in 200 countries. It boasts 23 brands with more than $1 billion in annual sales. How does such a large and complex global company pull off such impressive footwork?
The answer, Scafidi said, was preparation. “Digital is inherently disruptive,” he explained. “We’ve been training for this for a long time. We’ve been preparing to adapt to disruption for 20 years.”
Planning for change is a skill
Scafidi and I spoke from our remote locations about “Building Resilience and Adapting Your Marketing Tech in Uncertain Times” during Ad Age’s CMO Next conference, co-hosted by Ad Age’s Heidi Waldusky. Scafidi stressed that tumultuous times took PepsiCo back to basics — inspiring the company to lean on skills it had been developing for years — especially in consumer research and media measurement.
Part of the reason PepsiCo was able to launch Snacks.com and PantryShop.com so quickly, he said, was “we were leaning on what we were doing already.”
He reminded me of an analyst quote I read recently on embedding resilience into sales and marketing plans: “[T]he more an organization practices resilience, the more resilient it becomes.”
Over the past year, many organizations have had plenty of time to practice being resilient. As stores shut down and millions huddled at home, there was a surge in digital activity across all channels. Media consumption soared: people around the world watched 60% more video, for example. And they shopped. Salesforce’s latest Shopping Index shows that comparable online sales were up 55% in Q3 of last year after climbing 71% in Q2.
We’ve heard from many of our customers that they needed to launch new capabilities faster than ever before. Otherwise, they’d lose business. Curbside pickup, buy online, pickup in store, expanded digital storefronts, appointment scheduling, contact tracing – the list goes on.
Our desire to help customers adapt to rapid digitization inspired us to launch Digital 360, a suite of ecommerce digital marketing products combining marketing, commerce, and personalized experiences under a single umbrella. With it, Salesforce Trailblazers like Spalding and Sonos were able to scale their online commerce dramatically, making up some of the shortfall in brick-and-mortar sales.
When times are changing, it’s too late to build up basic skills. If you have a foundation in place, that allows you to adapt.
MIKE SCAFIDI, PEPSICO GLOBAL HEAD OF MARKETING TECHNOLOGY AND ADTECH
Unilever also faced dramatic market shifts in the recent past. Keith Weed, the company’s chief marketing and communications officer, pointed out back in 2018 that the pace of change “will never be this slow again” – not knowing just how fast that pace would get. And like PepsiCo, Unilever met the hyperfast present by relying even more on its customer research skills.
“We know that people search [online] for problems, not products,” Weed said. So the company created Cleanipedia.com, which offers detailed solutions to cleaning problems in 26 languages. Built before COVID-19, the site was ahead of its time and has attracted 28 million visitors to date.
Building a foundation that scales to meet customers wherever they are
When times are changing, it’s too late to build up basic skills. “If you have a foundation in place, that allows you to adapt,” Scafidi said.
For example, the PepsiCo team was able to rapidly restructure its internal media measurement analyses because it had already put in the work to develop an ROI Engine, which helped determine the real impact of its advertising, promotions, and email. The ROI Engine automates data inputs, processing, and algorithms to improve paid media optimization decisions. Combining the ROI Engine with a customer insight capability called Consumer DNA, “We were able to stabilize our understanding of the consumer and adapt to where they were,” Scafidi explained.
PepsiCo’s Consumer DNA project is an example of a custom-built tool that allows the company to gain a 360-degree view of the consumer to enable more effective and targeted media buying and marketing activation.
At Salesforce, we help our customers engage with their customers. To do this in 2020, we too relied on core skills, built up over years, to adapt to an environment that seemed to change by the second. The result was launches like Digital 360 and Work.com. The latter helps companies safely reopen their workplaces. We also introduced a customer data platform (CDP) called Customer 360 Audiences, which serves as a single source of truth to build a unified profile for marketers and others.
The Ancient Greek philosopher Heraclitus said, “the only constant in life is change.” As customers like PepsiCo show us, the best way to adapt is to build core skills that can help you pivot quickly in the future.
The following column originally appeared in the mighty AdExchanger on 11/10/20
Measurement is a footnote – outglamorized by targeting and the opera of browsers, pushed into the corner during debates about the future of media. But it’s arguably more important than aim and requires more discipline.
A few weeks ago, the World Federation of Advertisers (WFA) released a statement and a technical proposal for a cross-media measurement framework. It was the outcome of a yearlong global peer review and an even longer discussion among participants including Google, Facebook, Twitter, holding companies, industry groups such as the ANA and MRC and large advertisers including Unilever, P&G and PepsiCo.
Reactions ranged from enthusiastic to less so, but few people seem to have read more than the press release. After all, it’s not a product and could be yet another in a parade of grand ambitions in online-offline media measurement, dating back to Facebook’s Atlas.
But it describes a realistic scenario for the future of measurement. Sketchy in spots, the WFA’s proposal is ironically the clearest screed of its kind and is worth a closer look.
To be sure, this is a project focused on solving a particular problem: measuring the reach and frequency of campaigns that run on both linear TV and digital channels, including Facebook, YouTube and CTV. In other words, the kinds of campaigns that cost participating advertisers such as P&G a reported $900 billion a year.
And P&G’s own Marc Pritchard is on record calling the proposal “a positive step in the right direction.”
The need is clear. Advertisers today rely on a patchwork of self-reported results from large publishers, ad server log files, aggregate TV ratings data and their own homegrown models to try to triangulate how many different people saw their ads (reach), how often (frequency) and how well those ads fueled desired outcomes, such as sales lift.
The latter goal is acknowledged in the current proposal, which doesn’t try to solve it. But the WFA, building on previous work from the United Kingdom’s ISBA, Google, the MRC and others, lays out a multi-front assault on reach and frequency that covers a lot of ground.
How does it work?
The proposal combines a user panel with census data provided by participating publishers and broadcasters, as well as a neutral third-party data processor. The technical proposal spends some time talking about various “virtual IDs” and advanced modeling processes that are loosely defined – and the goal of which is to provide a way for platforms that don’t share a common ID to piece together a version of one.
Needless to say, a lot of the virtualizing and modeling and aggregating in the WFA’s workflow exists to secure user-level data. It’s a privacy-protection regime. It also engages with the much-discussed third-party cookieless future.
Panel of truth
The proposal leans heavily on a single-source panel of opted-in users. At one point, it calls this panel the “arbiter of truth,” and it’s clear most of the hard work is done here. Panelists agree to have an (unnamed) measurement provider track their media consumption online and offline. Panels are a workhorse of media measurement as provided by Nielsen and others, but they are expensive to recruit and maintain. It’s not clear who would build or fund this one.
In the past, other panels have struggled to collect certain kinds of cross-device data, particularly from mobile apps. Panels also get less reliable in regions or publishers where they have less coverage, a problem that could be addressed by joining multiple panels together.
In addition to the media consumption, demographic and attitudinal data it provides, the panel is used to “calibrate and adjust” much more detailed census data voluntarily provided by publishers (including broadcasters).
No walls here – at least in theory. Given that Google and Facebook support the WFA’s proposal, it’s implied they’re open to some form of data sharing. It’s already been reported – although is not in the proposal itself – that some participants will only share aggregated data, but it’s better than nothing. The WFA’s idea of “census data” includes publisher log files, TV operator data and information collected from set-top boxes.
This census data is married at the person-level with the panel data using a series of undefined “double-blind joins of census log data with panelist sessions.” Joined together, the different data sets can correct one another: The panel fills gaps where there is no census data, and the more detailed census data can adjust the panel’s output.
Virtual ID’s, anyone?
The census data will have to be freely provided, and so wide-ranging participation across many publishers is required for success. Another requirement is a way to tie impressions that occur on different publishers (which don’t share a common ID, remember) to individuals to calculate unduplicated reach and frequency.
The output here is a pseudonymized log file with a VID attached to each impression, overlaid with demographic data – at least TV-style age and gender cohorts – extrapolated from the panel.
Doing the math
In the final step, each individual publisher will perform some kind of aggregation into “sketches.” These sketches are likely groups of VIDs that belong to the same demographic or interest segment, by campaign. And it is worth noting here that the “sketches” can’t be reidentified to individuals and are somewhat similar to proposals in Google’s “privacy sandbox.”
At the penultimate step, each individual publisher sends their “sketches” to an unnamed independent service that will “combine and deduplicate VIDs” to provide an estimate of reach and frequency across the campaign. The WFA has a proposal for this Private Reach & Frequency Estimator posted on GitHub.
A GitHub explainer mentioning data structures and count vector algorithms is ad tech’s new sign of sincerity.
Finally, outputs are provided via APIs and dashboards, which support both reporting and media planning. End to end, it’s an ambitious proposal that has many of the right players and pieces to work. Its next steps are validation and feasibility testing led by the ISBA in the United Kingdom and the ANA in the United States.
Whatever happens, we’ve learned something from the WFA’s proposal. Even in a best-case-scenario, accurate global campaign measurement will definitely require heroic levels of cooperation.
Our TV screens are growing larger than our walls, surging eight inches on average in the last five years. And the most prevalent cultural phenomenon other than TikTok dance challenges is probably binge-watched streaming services, which spend more than $35 billion a year on new video content.
Blogging is dead and other favorite headlines
You’d be forgiven for believing that we’ve forgotten how to read. Judging by our popular culture, we’re becoming a post-literate, oral society, one whose always-dominant visual sense has overwhelmed our reasoning to the point where 72% of consumers now say they prefer all marketing to be delivered via video.
We don’t notice this trend because we’re part of it, but historian do. An iconic 1958 VW ad cited for its visual austerity contained just 165 words, including ’Lemon’. Last year, VW ran a magazine ad – remember magazines? – that had all of 13 words, including ’Volkswagen, it’s plugged in.’ Newspapers have seen subscriptions decline 70% since 2000, echoed by the 20 million hits received to the search “is blogging dead?”
Of course, it’s understandable. Sight is our dominant sense and is more primitive than the others. Almost 90% of the information going into our brain is visual, and 40% of neural fibers flow to our retinas. Images are processed much faster than text, which is a learned input that requires years of practice.
Words still exist. This sentence is proof. In fact, it may seem odd to be making such a statement in words, to a literate audience. But mass culture increasingly treats words as a kind of visual fillip, a graphic element used for iconic rather than informational content. Increasingly, American consumers are like English-only speakers who visit Tokyo, struck by the occasional familiar word among the kanji script.
Marketers: visualize this
How can a marketer adapt to the rise of the post-literate consumer?
First, make sure your brand has a strong visual identity – stronger than you think it needs. A recognizable logo and color palette aren’t enough. Assume consumers will not look at your logo – after all, they’re certainly multitasking. Your brand identity must be so strong it can be communicated simply through a consistent, insistent drumbeat of the same colors, fonts, shapes and styles.
Byron Sharp and the Ehrenberg Bass Institute influentially made a similar point in How Brands Grow. Sharp stressed the importance of “mental availability,” which is not awareness (as he often reminds his Twitter followers), but rather how familiar a brand’s (primarily visual) sensory associations are to consumers.
So be visually consistent, like your brand depends on it.
Second, simplify and streamline your cues. This is another trend that’s obvious to those who’ve looked for it. Logos, fonts, web pages, graphic design – all are retreating from clutter and complexity. It’s almost as though a decline in literacy has extended to the visual realm, or maybe we’re all just overwhelmed and told our visuals that – in the words of Taylor Swift – they need to calm down.
Cut the text. Lengthen the tweet?
Plenty of research confirms that most people prefer simplified designs that are neither complex nor particularly original. Simplification includes increasing the amount of white space, beautifying images – and cutting text.
Third – and most importantly – know when to ignore this advice. We have been talking here about mass consumer audiences. If you’re selling advanced hydroelectric plants, different principles apply. And remember that for every trend there is a counter-trend.
Some years ago, I worked at an ad agency doing social analytics for a luxury car brand. Examining the Twitter conversation about the brand, I noticed an odd phenomenon: it was bimodal. I mean that about 80% of the comments were inane, silly, and crass – what you’d expect. But 20% were very different: intelligent, thoughtful, almost nerdy. I concluded there were two different Twitters out there.
If you’re appealing to the second Twitter, the realm of academics and the informed, don’t sound dumb. You may need to increase rather than decrease your word count.
As I suspect you already know, since you’ve made it this far into a 1,000-word essay on marketing, there’s still a lot of life left in words.
Martin Kihn is senior vice-president strategy at Salesforce.
There’s a tart scene in Mad Men when Don Draper eyes the iconic VW Beetle ad with the one-word caption: “Lemon.” After a thoughtful pull on his Lucky Strike (“It’s Toasted!”), he says: “I don’t know what I hate about it the most — the ad or the car.”
These days, too many people seem to agree: they don’t like ads. They’re interruptive and meddlesome. Besieged on one side by blockers and browsers, on another by unfriendly memes, advertising seems to be having a moment.
A few years ago, NYU gadfly Scott Galloway went so far as to declare “The End of the Advertising-Industrial Complex,” scripting a future of subscription-only media and brands propelled by word of mouth alone. In a dark analogy, he called modern digital marketing a world “full of anxiety, humidity and …” so on.
Ouch. He reminds me of the observation ascribed to John O’Toole, late chair of Foote, Cone & Belding, that most critics don’t like advertising just “because it isn’t something else.”
It wasn’t always thus. The first major history of the ad business concluded advertising was “a civilizing influence” and “an educative force.” That was in 1929. Times have changed, although as recently as the 1970s a voice as acrid as media theorist Marshall McLuhan’s said, “Advertising is the greatest art form of the 20th century.”
Maybe so, but an increasing number of people seem to be asking: Do we even need it anymore?
What Does Advertising Do, Anyway?
The best and most obvious case for advertising is that it creates (or encourages) demand. As the economist John Kenneth Galbraith famously said in The Affluent Society (1958), advanced societies can’t prosper unless they convince people they need more than they actually do. In addition to information, advertising is supposed to supply desire.
A lot of academic angst has been directed at this point, but let’s avoid that for now and assume ads are supposed to bring in more at the cash register than they cost. Certainly, a lot of – let’s also assume –intelligent people believe this to be true. Even during this very difficult year, some $215 billion will be spent on ads in the U.S.
But do they work? It’s conceivable advertisers are in the pincers-grip of wishful thinking and fuzzy math. Having spent some years in the ad measurement business, I can tell you that measuring return on ad spend (ROAS) is not as easy as it looks, and it doesn’t look easy.
Oddly, digital channels make measurement harder – not simpler. When a majority of ad spend flows to platforms that do their own measurement and don’t release raw data, in the name of privacy, complex equations are required. The good news is that analysts who have worked these equations, over time and on a large scale, generally agree that advertising works. That is: brands that advertise more can be shown to have higher sales, on average, and the impact of ads is incremental.
What about Tesla? Sooner or later, the anti-ad crowd always mentions this brand. After all, four of the five biggest car makers – all major ad spenders – are losing share. Tesla is growing, despite spending almost nothing on ads.
My response to this is simple: come up with a revolutionary product that disrupts a 100 year-old industry, add a charismatic chairperson, and you too will enjoy so much free publicity that it would be redundant to advertise.
What Else Have You Done for Me, Lately?
Demand is its day job, but advertising has more mojo. There is the industry itself with its quarter-million hard-working employees. Many of these people have impressive (subsidized) side-hustles in the arts and sciences. And there is the indirect impact of additional demand, as companies sell more and then hire, buy equipment, take out leases, and so on.
To add all these direct and indirect effects up is a complicated exercise that has been attempted over the years. Each time, conclusions were impressive. A 1999 study overseen by a Nobel Laureate found advertising accounted for about 2.5% of U.S. economic output.
A decade and a half later, IHS Global Insight concluded that every dollar spent on advertising generates $22 in economic output (sales), and every $1 million spent supports 81 American jobs. An update done in 2015 found that advertising impacted up to 19% of U.S. GDP.
There are a lot of caveats, of course. IHS is a respected research firm, but the reports were sponsored by ad industry groups. The reports’ definition of advertising was very broad, e.g., including direct mail. But even if the numbers are discounted, they still point to a major impact on the nation’s economic life.
And we haven’t even mentioned that ads support content. It appears that advertisers spend about $35 per month to reach each U.S. adult online, which should make us feel pretty good. That money pays for news, entertainment and utilities that in an ad-free world either would not exist or would cost us something. Newspapers have lost 70% of their ad revenue since 2006, with a similar decline in reporters and content. Oft-mentioned counter-examples such as the New York Times’ rising subscription revenue are really just poignant outliers.
On the other hand, somebody is making money on ads. Two of the five most valuable companies on Earth are almost entirely ad-supported. Content on Google (including YouTube) and Facebook (including Instagram) is free, and the prolific ad experience they offer clearly isn’t turning people away. So the “advertising-industrial complex” may not have ended but just changed its owners.
Let’s admit that ads aren’t going away. The business needs to adapt its data and approach to a new reality. And many of the industries’ problems — from oversaturation to intrusive targeting — are actually self-inflicted.
Don Draper gives us some solid advice: “You want respect? Go out and get it for yourself.”
If you want to see the dramatic impact of COVID-19 on a world-class marketing department, look at Unilever. The consumer products powerhouse met a challenging global environment in Q2 and is making rapid changes in tactics, reviewing all marketing spend “to ensure it’s effective and appropriate,” according to CFO Graeme Pitkethly.
Meeting chaos with agility, Pitkethly points out that the company is “dynamically reallocating” budgets as consumer behavior shifts, moving resources out of outdoor ads (no traffic), and TV production (not safe) into areas with higher immediate return-on-investment (ROI) – such as skin care, home, and hygiene.
Unilever’s quick shifts and shimmies are mirrored by our other customers, many of whom are charged with increasing marketing ROI with fewer resources. The World Bank predicts a baseline 5.2% contraction in global GDP in 2020. Gartner’s CMO spending survey released in July showed 44% of CMOs expect midyear budget cuts.
What’s the best way to improve marketing ROI in today’s challenging landscape? By increasing one of these three things:
Effectiveness – get more revenue from the same investment
Efficiency – get the same (or more) revenue from a lower investment
Optimization – a combination of these through better resource allocation
And how do you know where to start? Data.
Keeping this framework in mind, and highlighting examples from our customers, here are three ways you can use data to increase your marketing ROI.
1. Dial up your digital (Effectiveness)
At a time when most of us spend more time than we’d like staring at screens, digital channels are the best way to reach us. Many indicators from time spent on mobile to time wasted – um, spent – playing networked video games prove that a lot of our lives are now online. And from a commerce perspective, McKinsey said consumers “vaulted five years in the adoption of digital in just eight weeks.”
Digitizing just as fast, marketers are ramping up their technology investments to manage customer data and use it effectively. And if you’re not, you should be. In addition to providing greater control over channels, data-driven investments include analytics to improve customer segmentation, message personalization, and targeting methods such as lookalike modeling. Despite an overall decline in enterprise tech spend, Forrester forecasts a rise in marketing technology investment. “In some cases,” says VP Principal Analyst Shar VanBoskirk, “technology may offer greater efficiency than relying on manual effort.”
Case in point: Orvis, an outdoor clothing retailer, saw pandemic-related store closures lead to tighter budgets and a mandate to improve engagement. They used Einstein Content Selection to automatically choose when to send standard messages and when to deliver content from its values-focused “Giving Back” campaign. They also took advantage of Einstein Send Time Optimization, which uses artificial intelligence to predict delivery times that align with when each recipient is more likely to open an email. Both approaches together led to a 22% higher email click-through rate.
2. Send fewer messages (Efficiency and Effectiveness)
Yes, that’s right: less can be more in digital marketing. All of us know the experience of being hounded by a brand to the point that we stop engaging, progressing from tuning out to turning off (hello, “unsubscribe”). During a time of message saturation, keeping your communications clear, on point, and not too frequent can cut costs (efficiency) and raise responses (effectiveness).
Last year, more than 40% of U.S. consumers said they were “overwhelmed” or “annoyed” by the volume of marketing content they experienced daily, according to Gartner. That reaction has only sharpened this year. In fact, Gartner calls this message-stress syndrome “COVID Fatigue.”
Treating the syndrome takes a good source of customer data and links among call centers, ecommerce, marketing, and other systems. With the right pipes in place, you can execute ROI-boosting tactics like:
Suppress social ads to people who have an open case
Merge customer records
Send fewer messages to people who are overwhelmed
Cable network Showtime did just that, starting before the pandemic. It used Salesforce Audience Studio to suppress spending on existing subscribers, then shifted budget toward those who had recently canceled, for the win back. This more efficient approach to their marketing helped them reach 25 million people.
3. Speed up planning cycles (Optimization)
During normal years – that is, not now – it’s common for marketers to do quarterly or even annual media budgeting. But in today’s environment, that pace won’t work.
Measurement, reallocation, testing, and optimization – these should be ongoing disciplines, not intermittent ones. Continuous monitoring allows you to move spend to higher-performing channels, cut short losses, and make the most of the resources you have. It also allows you to respond to market shifts, such as changing your tactics in areas affected by natural disasters or virus outbreaks.
Part of reimagining marketing in the “next normal,” according to McKinsey, is always-on customer data analytics: “Analytics will need to play a core role not only in tracking consumer preferences and behaviors at increasingly granular levels, but also in enabling rapid response to opportunities or threats.”
Part of this acceleration requires better data management, moving from manual reports to an automated real-time system. At Salesforce, we managed to combine marketing data from 83 sources and 182 streams using Datorama. This reduced wait time on data from two to three weeks to near real time. Most important, marketing ROI grew 28%.
Even as businesses face tighter budgets and a lower tolerance for risk, there’s still a world of opportunity for marketers to increase their ROI. You just need the right data to help you chart your course.
Despite well-known struggles keeping up with consumers’ hyperactivity, Nielsen remains the gold standard of media ratings. Some $70bn in US media are bought and sold based on its ‘currency‘. Its various ‘Total Audience’ products, introduced five years ago, layer in viewing on mobile, streaming and on-demand platforms using a combination of panels and direct data collection.
Ratings measure reach, of course, and they started with TV viewership diaries and then automated in-home ‘People Meters.’ Capturing online consumption was handled at first in a similar way: 200,000 people installed a ‘PC Meter‘ on their computers, which tracked video viewing on websites. But digital media presented frustrating technical issues. For example, in 2010 Nielsen admitted it was undercounting time spent online by as much as 22% because of super long URLs.
Don’t blame the players – blame the referee. Some of the industry pushback is a case of punishing the messenger. After all, it is happening at a time when linear TV is seeing its viewership in a state of free fall (down some 20% in five years). When she was president of Nielsen’s Watch (ratings) division, Megan Clarken observed wryly: “Like any referee we’re not always going to be loved.”
Still, every marketer is in a sense in Nielsen’s (and Comscore’s) position: having to report some kind of measurement of reach and response across an array of channels. We’ve all had to adapt to abrupt changes in consumer behavior, data availability, and tools over the years. We can learn from Nielsen as it points us to what to do next.
So where are they pointing us? I think, in three directions
Reliance on partners
Nielsen will be relying on various proprietary ecosystems, such as social networks, to provide data about consumption that would otherwise be opaque. As the company’s Chief Data and Research Officer Mainak Mazumdar admitted in an interview, “We will work with multiple parties in a significant way, which we did not in the past.”
Whatever the outcome of the current Congressional probes, few industry observers believe the open web is poised to grow. Growth is in the gardens, which already hold a dominant share of digital ad spending. In the US, about two-thirds of such ad spending goes to Google, Facebook and Amazon, according to eMarketer.
None of these ecosystems allows marketers to see user-level data, such as impression log files with IDs. Without that level of detail, marketers can’t build real multi-touch attribution (MTA) models. They can’t independently measure unduplicated reach and frequency. They’re reliant on aggregate reporting provided by the platforms themselves, or on managed tools such as Google’s Ads Data Hub.
What this means: Make a map of all the proprietary ecosystems, including big publishers, that see your audience. Develop a detailed understanding of the data provided by each one. If you have scale, lobby for more access, or ask your agency to do it.
Admitting nobody really knows what’s going to happen, COO Karthik Rao revealed a major goal of his team was to build “a flexible platform that we can adapt to new technology, data and regulatory changes.”
Everybody embraces adaptability in principle; in practice, it’s not so easy. Adaptability means that the method used must be able to accept data at different levels of detail – from national-level campaign data down to user-level impressions and clicks – depending on what’s available. This availability in turn depends on media partner policies and privacy regulations, by region.
So, we admit that a single MTA vendor “silver bullet” – so hyped in a decade ago – won’t work. Whether we want to or not, we will all need to use more sophisticated econometric and media mix models (MMM), in-house or through an agency. There are too many unpredictable variables for simple models to succeed.
Nancy Smith, president of Analytic Partners, pointed out recently in the Drum that the future of measurement falls more heavily on MMM than MTA. “In my own review of activities with marketers,” she wrote, “I’ve seen about 80% of the impact coming from MMM and only 20% coming from MTA.” Instead of a standardized approach, therefore, she advocates “user-level analyses within the channels that matter” combined with a “holistic measurement framework” that unites these channel-specific measures.
What this means: Develop a distinct approach to measuring individual channels, including big publishers. Incorporate these into a larger measurement framework based on econometric principles.
Ratings are important, but the goal of measurement is to determine impact: did the campaign or ad view cause incremental sales, or improve brand perception? In the absence of complete data at the individual level, marketers will have to execute more in-market tests to measure the incremental impact of ads.
It’s a daunting task. Back in 2013, researchers at Google published a depressing paper titled “On the Near Impossibility of Measuring the Returns to Advertising.” They pointed out that there is simply too much noise in the ad environment to make measurement useful: too many factors, like the economy, the weather, consumers’ moods, competitors’ moves, viewability, etc., that obscure the truth.
There’s still a lot of noise, but our methods have improved. And it’s reassuring to see that Google now encourages testing to determine the impact of ads. In a recent blog post, the company said marketers should strive for the “gold standard” of using treatment and control groups: “Experiments … should play an important part of an advertiser’s attribution strategy.”
What this means: In situations where data gaps are significant, tests can add information. And often, they are the only way to make sure the ads really caused the outcomes you’re seeing.
If all this seems like an admission that the future will be more complex and unpredictable than the past, that’s because it will be.
The following article appeared originally on the Salesforce.com blog on 8/6 as part of the “Moment Makers” series, supporting the upcoming launch of my & Chris O’Hara’s book “Customer Driven” (about CDPs), coming later this year.
Customers behave differently during crises, and our data strategies have to adapt. During the dark days of the financial crisis of 2008-09, when I led an analytics team at a marketing agency, I met with the CMO of a U.S.-based credit card company. She had noticed a disturbing pattern in her customer data: early-stage companies were charging more on their cards but missing more payments. Heavy late fees made their burdens worse, at the worst possible time.
Looking at the drivers, we saw the problem came down to cash flow. It was temporary and often outside their control. So the CMO made a bold choice, offering a new “start-up card” with less stringent payment terms. The company lost little on the venture and made a cohort of customers who proved to be more loyal than the company’s average cardholder.
Today, as businesses around the globe decide how and when to reopen, it’s a great time to revisit our assumptions about our customers. Like the charge card company’s, your customers have probably changed due to the pandemic.
Based on experience and what Salesforce has learned from our own customers, we recommend three basic principles to guide data-driven marketers in reassessing their approach to customer data: communicate clearly, revisit old assumptions, and practice long-term thinking.
1. Communicate clearly
Whatever your industry, your customers likely start with a level of distrust. For example, a Pew Research Center survey of U.S. adults revealed that 81% think they have little control over companies collection of their data, and an equal proportion think the risks outweigh the benefits. An alarmingly small 6% said they had a “good understanding” of how companies are using their data.
Personalized communication is less about you as a company and more about your customers – their wants, needs, and attitudes in the moment. Uncovering customer needs from data is even more important during times of rapid change, and it’s a key differentiator in customer connections. Salesforce research [recommend viewing dashboard on desktop] found that 71% of consumers say companies who show sensitivity to the current climate are more likely to earn their loyalty.
Customers also told us they are willing to share information with companies that inspire trust by clearly communicating how, when, and why they’ll use the data. In fact, 58% of respondents to our State of the Connected Customer report said they agreed with the statement “I’m comfortable with relevant personal information being used in a transparent and beneficial manner,” while only 17% disagreed.
Key takeaway: Rather than focusing only on data pipelines, customer data teams should also double down on communication, telling customers exactly how and why they’re collecting and using their data, and what they’ll get in return. This strategy is known to produce happier customers.
2. Revisit old assumptions
During the early months of the COVID-19 crisis, brands scrambled to adapt their messaging – for example, by removing things like high-fives and finger-licking that no longer fit into a world of social distancing. After that, many rose to the moment with campaigns that were more down-to-earth. For example, Toyota pivoted from a sales-promotion campaign to a series of videos emphasizing home and family with the tagline, “We Are Here For You.” Revamped campaigns better reflected customers’ changing needs: one survey revealed people want to hear messages that were “safe” and “hopeful” from brands.
Recognizing no two crises are exactly alike, researchers at Harvard studied how customer attitudes and behaviors change during downturns. They saw broad changes both in behavior and how categories of products and services were perceived. Some customers remained optimistic, while others (often, but not always, younger people) threw caution to the wind and “lived for today.” Meanwhile, some “nonessential” products such as makeup and skin care are seen as even more important by many sheltering-in-place today.
Recent Salesforce research [recommend viewing dashboard on desktop] shows how fast consumer sentiments change. Between May 1 and July 1, customer optimism rose six percentage points, confidence rose five points, and trust rose six points. More than ever, marketers need to examine customer data and adapt rapidly to shifts.
Key takeaway: Your customers are going through a lot of changes, and it makes sense that what worked pre-pandemic may not work so well now. Previous analyses into segments, attitudes, and positioning will have to change. Revisit them and develop tools such as surveys and tests to fill data gaps.
3. Practice long-term thinking
There’s ample evidence investing in fundamentals such as R&D, workforce training, and brand advertising during tough times yields long-term benefits. The same goes for customer relationships. If you can afford it, reorienting around longer-term key performance indicators (KPI) and more holistic measures of health will provide long-term dividends.
Now is a time when most B2C and B2B companies are dealing with price sensitivity. Discounts and other concessions are common. There is wholesale revision in annual targets, as relatively few companies can maintain their pre-pandemic momentum.
Shifting from a topline-focused KPI such as net sales or profits may not work anymore. However, short-term pain can yield long-term gain. Recall the components of customer lifetime value (CLTV): revenue x frequency x retention %. If you’re willing to be patient, shifting to a focus on retention and loyalty through superior empathy and service will overcome a (temporary) dip in revenue.
Key takeaway: As we gear up and return to work, we owe it to our stakeholders and customers to ensure we’re infusing clarity into our data collection and communication, we’re revisiting our pre-pandemic data assumptions, and we’re focused on making recovery work for the long run.