Exploring Marketing Technology

Category: Web Marketing

New ways to create and measure sites so they improve their ability to bring your best customers closer and attact other individuals just like them

  • Competing at Machine Speed: A Book Review

    Competing at Machine Speed: A Book Review

    I just read a book that describes an important use case for AI automation. If you work in the marketing of enterprise brands, you should read it too. 

    I’ve licensed the above cartoon because it aptly reflects a major challenge facing brands in an increasingly complex shopping environment. Today product description pages (PDPs) need to serve two masters: The humans who shop for a product, and the AI agents shopping on those humans’ behalf. Agents like Meta’s Muse1, a new assistant whose app was downloaded 5 million times in its first three weeks.

    Muse and other agents don’t care about carefully crafted packaging, emotional storytelling, or brand heritage. They want one thing: product facts … and plenty of them. Marketers who don’t do that work — inserting those facts into PDP data schemas — are doomed to lose progressively more share to upstart competitors who do.

    It’s only one of the challenges facing enterprise brands that are explored in the opening chapters of Competing at Machine Speed. That section delves deeply into a daunting list of brand marketing headwinds, using many real world examples. Notably, the book describes the emergence of new, post-pandemic battlegrounds where brand sales are won and lost: the retail sites. 

    Each major retail platform (Amazon, Walmart, Target, Kroger, Instacart, Albertsons, and the rest) runs its own proprietary algorithm for determining where products appear in search results, which sponsored placements convert best, and which content signals it uses to judge a product’s relevance and quality. Those algorithms are not published. They are inferred from performance data, industry reporting, and trial and error. And they change.

    That last sentence is key, because changed requirements by the platform must be addressed quickly just to maintain placement prominence. 

    Before these retail sites emerged, the marketing of your product was simpler. You promoted the product and ensured physical store shelves were stocked. Yes, there were competitors in your category on those same shelves, but their number was physically limited. You could adjust your advertising to elevate your comparative primacy and make the sale.

    Of course you still need to do that. But today important retail sites have their own “shelves”: the product listing pages (PLPs). And because of long-tail economics, the platforms have an incentive to include in their PLPs and search landing pages as many competing brands as meet their low cost of entry — dozens or more of them, all vying for your customer.

    This has turned selling on retail platforms into ruthless cage matches. Bidding determines which brand gets primacy on PLPs. The auctions erode margins and can even exert downward pricing pressures. Add to that the fact that you can see your prized brand demoted because of changed requirements I mentioned above. Those must be addressed, and often are not in a timely manner, because you and your colleagues are busy making similar changes for higher priority brands in the portfolio.

    A challenger brand with one hundred SKUs and a team focused entirely on its digital shelf can manage every product with the intensity a major brand reserves for its top 10%. Its Long Tail does not exist. Its entire portfolio is the Head.

    Competing at Machine Speed

    The primary way to win these cage matches is automation. That is the agentic use case the rest of the book covers.

    Change of this magnitude is hard, especially in an enterprise with many layers of oversight and governance. The authors knew it will take a lot to persuade the C-Suite. That’s why the book presents, near the end of that first section, ample evidence of small, nimble challengers “swarming” most brands, in most categories, and stealing their market share today. 

    Challengers use their small size as an advantage. Industry metrics show they are succeeding:

    Beating the challengers at machine speed

    The book’s solution — the bulk of the book — is to employ agents that “think” like start-ups, meeting challengers where they live:

    The swarm’s advantage is structural, not strategic. It is the advantage of being small, having little to protect while attacking everything.

    What changes the calculus entirely is when the incumbent builds a system that matches the swarm’s operational speed while applying the advantages that only the incumbent has: the brand equity, the category expertise, the retailer relationships, the consumer trust built across decades, the capital to invest at scale. The swarm can replicate keywords. It cannot replicate twenty years of consumer loyalty and clinical investment. What it can do, and is doing, is render those assets invisible by competing in an environment where the [incumbent] brand cannot show up consistently.

    The system architecture described is detailed and comprehensive. So much so that the book’s text can be used to align enterprise LLMs on how to organize themselves and get to work.

    I don’t want to give too much away, because you really do need to buy the book and study their proposed architecture first-hand. But one thing I want to emphasize is that this agentic use case is not a strategy for shedding headcount:

    This is not about replacing the humans on the commerce team. It is about reassigning them to work that their intelligence is suited for. The category manager who spent 40% of their week pulling data and updating reports can now spend that 40% developing the commercial strategy that the agents then execute. That is the shift the system enables. Operators become Orchestrators.

    Two of the authors are work friends of mine, esteemed colleagues from my tenure at Accenture. But I have no financial or personal stake in endorsing this book. You should keep that in mind when I tell you that this book blew me away. I’m confident it will become a business category best-seller. 

    If you’re a marketer at an enterprise brand, you need to understand how to meet today’s challenges. This book gives you the blueprint. Whether you’re a CMO, category manager, or digital strategist, it will fundamentally change how you think about your role in the commerce chain.

    1As of this posting, Amazon is blocking Meta’s agent. This may change. In my opinion it will, when Meta works out with Amazon ways to compensate for the likely loss of income from the leapfrogging of Amazon’s profitable Marketplace ecosystem.

  • Hey TikTok, where is my check for inventing the metric that powers you?

    Hey TikTok, where is my check for inventing the metric that powers you?

    A spooky tale about a dead metric rising from the grave to take over the world

    This 2010 blog post of mine, Content Interest Index is the “missing link” in web analytics, outlined a novel metric that fills a measurement gap in a sales journey. Six years later, TikTok was launched with an algorithm that preforms precisely this measurement. And does it brilliantly. Here’s the story of my quest, and what it teaches us about finding and filling innovation gaps.

    I outlined the business case in an earlier post. It described why Content Interest Index (CII) needed to be invented:

    The first marketing class I ever had in college taught me the AIDA model of advertising. It’s still used today, lo these many years later. The AIDA model goes like this: [To make a sale,] once you attract Attention, you must generate Interest, create Desire, and enable your market to take Action. Do all of those things and you’re golden.

    Back then the only way you could actually measure any of these (except for the last A, which was making a sale) was by employing expensive and time-consuming research. The web changed all that. It allows us to measure each of these steps — except for that pesky thing called Interest.

    When I posted that I was leading a team of web developers. Using Google Analytics 1.0, I added client-side instrumentation to measure proxies for the interest levels of visitors in the pages of a financial institution’s website. Those proxies included how many people shared a page, bookmarked it (a thing back then), or sent it to PDF for printing. 

    With those behavioral proxies we hoped to calculate the comparative “interestingness” of this client’s content. Unfortunately our contract expired and I was never able to analyze the results and calculate the likelihood of CII to predict entering a conversion funnel (Desire) and achieving an online conversation (Action).

    I soon moved on from that agency, and also concluded the client-side instrumentation available at the time would never provide enough of a signal to be useful. I considered CII a dead metric.

    Then came TikTok.

    The breakthrough of ByteDance was using newer, real-time technology to create an app that leveraged its own proxies for interest, and prioritized them in a scoring model that could revise itself in milliseconds.

    According to this Wall Street Journal investigation into TikTok’s secret — and extremely addictive — algorithm, the app prioritizes engagement signals including likes, follows, and watch time. Most notably, “We found that TikTok only needs one of these to figure you out: How long you linger over a piece of content.”

    They found what many of us have experienced firsthand: Within minutes of lingering even slightly longer on a particular video type, TikTok adjusts your feed to show more similar content.

    This methodology became a blueprint. Meta adopted similar interest-duration models, and within years, interest measurement became the industry standard. In January, TikTok formalized its U.S. operations with an investment group, valuing the platform at approximately $14B USD, a vindication of the algorithm’s effectiveness.

    The Takeaway for You: The AIDA model had a digital analytics blind spot, and for years, we couldn’t find a way to fill it. Rather than accepting that limitation, I tried to bridge the gap.

    You likely face similar blind spots in your current behavioral analytics work: metrics you can’t measure, customer signals you can’t track, or questions your tools can’t answer. That gap is where innovation lives. When you find it — and have the persistence to investigate — you may be creating the next industry-standard metric.

  • Three lessons from my 15 years on Twitter

    Three lessons from my 15 years on Twitter

    Today is my 15th year on Twitter. The medium has changed a lot over the years. More importantly, it has changed me. And society. Here are my thoughts.

    1. Twitter’s history has been a dance between innovative users and its sometimes canny developers

    Twitter’s sudden celebrity will soon become a fight for relevance

    The above post on this blog was from its very early days. Twitter was born of SMS, that 140-character text capability that was a biproduct of cell service. The character limit wasn’t alone in suppressing adoption. So was the fact that early cell phones, pre-cell-based-broadband, actually charged you once your texts exceeded a certain monthly maximum.

    I was new enough on the platform, following pitifully few people and being followed by fewer, that this never phased me. But it was a problem for danah boyd. She is was anthropologist studying (at the time) youth and technology, before she left academia to work and publish at Microsoft Research.

    boyd’s thoughts on the new technology called Twitter, which she encountered full-on at the 2007 South-by-Southwest, could be summed up in her blog post’s lead: “SXSW has come and gone and my phone might never recover. … To the best that i can tell, i received something like 3000 Tweets during the few days i was in Austin. My phone was constantly hitting its 100 message cap and i spent more time trying to delete messages than reading them.” (danah admitted at the time she was under the heavy sway of ani difranco, and her capitalization habits reflected it and still do.)

    Twitter’s developers solved this problem within the year, by developing an app-based approach to following and reading tweets. But other innovations came from the users themselves, notably the invention of the hashtag. It was proposed by Chris Messina, late in the summer of 2007:

    https://twitter.com/chrismessina/status/223115412

    Adoption spread quickly, and soon Twitter had built an easy search mechanism for hashtags that we all take for granted.

    2. Twitter has fostered grassroots revolutions and community-based journalism

    I feel privileged to have witnessed history as it unfolded. One example is the Arab Spring of 2010-12, where, especially in Egypt and Tunisia, Twitter and Facebook helped to organize protests and, in cases such as Libya, take down dictatorial regimes.

    Closer to home, I recall following the updates about the pursuit of the Boston Marathon bombing suspects nine years ago from two screens: My television, tuned to CNN, and my cell phone. As reporters on the ground, and copters circling the neighborhood where the bombers were hiding, were frantically speculating where they could have disappeared to, people I followed on Twitter were already, correctly, speculating they were hiding under the storage tarp of a boat parked in a driveway.

    3. Forging friendships and business associations I never could have had

    Especially since I quit Facebook, in light of the company’s egregious disregard for member privacy and spreading of fake news, Twitter has been my primary social media app for deepening some friendships and forging new one. Especially when I return to my old stomping grounds in Milwaukee, I often connect with people I have gotten to know on the platform extremely well. Twitter friends only, it’s been fun to meet them IRL.

    I’m reminded that “adulting” often gets in the way of making new friends. Twitter has helped me in that important aspect of happy adulthood (and helped me familiarized myself with a ton of coinages such as IRL and adulting!).

    Where from here?

    These are the three bright spots on what could be a condemnation of social media. But I’m not Pollyanna. I am reminded daily of how Twitter has contributed to the disassembly of professional journalism and proliferation of filter bubbles, leading to no shared community or belief in things that are unquestionably true.

    But what strikes me today, on the very day I first registered on the platform, is how much Twitter has contributed to my life, and the course of technological, social and geopolitical history.

  • Apollo was the god of prophecy. His namesake is a digital insights time machine

    Apollo was the god of prophecy. His namesake is a digital insights time machine

    Adam Greco has been doing digital analytics implementations since before Adobe bought Omniture (now Adobe Analytics). I go back nearly that far, and have religiously studied his blog posts on the topic. He’s saved me hours of work. So I was thrilled to talk to him the other day about a new product category invented by his latest employer Search Discovery: Apollo, an Analytics Management System. I’m impressed, but I’ll be calling it a Digital Insights Time Machine.

    Here’s why:

    Excellent Digital Governance

    If you’ve been in my shoes, and Adam’s, you know the trouble an organization can get into if it doesn’t have buttoned-up digital governance. Like when an enterprise lacks a clear insights generation strategy.

    You see, that strategy describes business goals and objectives. and from them, predicts the reports that will be needed. This measurement strategy answers the question, What user behaviors are needed to achieve our business goals? Many organizations skip this step, and go right to implementation … Often trusting analytics team members with little understanding of business (!). This leaves these implementation pros having to make a best guess at what reports the enterprise will need down the road.

    Imagine you’re someone responsible for an implementation of Adobe Analytics, and the chat window lights up with a request from your boss’s boss. Or boss’s boss’s boss. “Can we get a report on XYZ?”

    When there is excellent digital governance in place. the answer you give is almost certainly Yes. But if you don’t, you’re faced with telling someone in control of your career that the report requires metrics that are not currently measured. Worse, the implementation will take weeks or even months because a change to the digital analytics data layer will be needed, and IT has many hotter priorities.

    If you’ve faced this moment of white hot panic, you probably have wished you could climb into a time machine and make sure those metrics get implemented out of the gate.

    Apollo is that time machine.

    The first thing you’ll see, if you get a demo like the one Adam showed me, is Apollo’s best practice library of business requirements, for reporting and insight generation.

    The list is vast, literally hundreds of requirements.

    Search Discovery — relying in part on Adam’s deep experience with clients in every industry — have provided building blocks for any type or hybrid of online business . From these business requirements flows all of the metrics and dimensions that will be needed to address them in reporting.

    Then things get really interesting

    This shows part of the flow that is followed as you set up your Apollo instance:

    Everything following out of the Business Goals and Objectives of this Measurement Strategy value tree is prompted from you by Apollo. Keep in mind that I’ve avoided arrows in the graphic, that would typically connect boxes from one column to the next, but as you likely have guessed, there is a one-to-many relationship flowing from left to right, with all the relationships being contained and documented within the Solution Design Resource (SDR).

    If you state a requirement such as “I want to report how many orders are placed each day, week, month, etc.,” you select that requirement as needed and Apollo automatically adds all of the variables, data layer objects, tagging, etc., that you’ll need to track within the SDR.

    More about that SDR: Unlike all of the SDRs I’ve ever encountered, the one in Apollo is part of its relational database. That means it can be exported to Excel if you feel inclined, but lives as a dynamic document that revises itself every time you make a change to requirements, metrics or reports.

    So when you join an organization using Apollo, you never have to encounter all of the lapses and omissions that come from an SDR that is only half-heartedly updated as an Excel file (often in several versions, causing you to wonder which contains the most “honest” implementation snapshot!).

    Leveraging APIs to Adobe Analytics, Launch, and even Workspace

    How does Apollo populate your instance of Launch? It’s connected via API to your instance of of both Adobe Analytics and Launch. Since it adds Launch tags, all that’s left is for you to do is refine its work and begin the testing.

    And because most implementations require IT to install or update the data layer, Apollo auto-generates the JSON code for that data layer. This makes the work of IT easier, improving the odds of speedy deployment.

    Finally, Apollo helps you at least two ways to debug the implementation once it is deployed. It uses those API connections to identify errors, and pushes to Adobe Workspace the reports that can make visual review of the data easier.

    Speaking of Workspace, all of the reports that are specified in this digital analytics “time machine” are pushed there, so all you’ll have to do is review and refine them.

    Apollo has impressed me so much that I can’t wait to get my hands on the working system, for my first client to use it. If you’re also intrigued, contact Adam for a demo. Like me, you’ll get a glimpse into the future of our industry, where we can spend more time on strategy and insight generation, and less on wrangling code and change requests.

  • How to lie with data visualizations, Part 2

    How to lie with data visualizations, Part 2

    This is a follow-up to my post from last month. In that Part 1, I wrote about how shifting the ranges for heat maps and starting bar charts at numbers greater than zero can deceive. In this installment, I want to introduce you to the greatest source of lies in data visualization: Cognitive biases of their authors.

    I started my earlier post with an example torn from the headlines. I will again below.

    If you follow the latest news in the U.S. about Covid, you may have seen this graphic shared over the weekend, included as evidence for emergency approval of convalescent plasma as a treatment for the infection. The evidence was provided by FDA Commissioner Stephen Hahn. His chief graphic is shown in the above image.

    The graphic  appears to show a 37% reduction in deaths for Covid patients receiving infused plasma from donors who have survived the infection. Famously, Tom Hanks and his wife donated their plasma to help the research being conducted world-wide.

    Rounding down to 35% in a press conference, “A 35 percent improvement in survival is a pretty substantial clinical benefit.”

    It is, but that isn’t what the data shows.

    After an outcry from what seemed like just about everyone in Covid research whose name was not Stephen Hahn, he apologized. The drop in mortality was from roughly 11% to 7%, which is a drop of somewhere around 4%, not 37%. It was also from research on a small sample size, weirdly without a control group, among many other concerning factors.

    In an administration that is arguably more politicized than any in recent history, the impulse is to say this U.S. government official lied. Strictly speaking, he most certainly did — in both his words and his supporting data visualization. But the motivation could be less nefarious — or at least, more common.

    Cognitive biases can make us lie to ourselves

    Our brains are not the perfect instruments we delude ourselves into believing. Consider the Free Brian Williams episode of Malcolm Gladwell’s podcast Revisionist History about the fallibility of human memory. Or more relevant to this instance of (self-)deception, consider this five-minute excerpt of a talk I co-presented in 2018 with Chad Sanderson at Adobe Summit in Las Vegas. It was called Profit Through Personalization. In that talk I explained why clear, data-focused hypotheses are an important protection a data scientist has against cherry-picking or distorting data without even knowing it.

    Everyone wants to report successes. But science is hard, and replication of test results is a persistent problem. In addition to clear hypotheses, we can also conduct an exercise called preregistration to save us from ourselves.

    In conclusion

    In my first installment I talked about how easy it is to distort numbers by using bad or lazy visualizations. In this second and last installment, I want to remind you that a far more pernicious murderer of the truth is your own best intentions. I’m willing to give Mr. Hahn the benefit of the doubt that he wasn’t so much lying as practicing wishful thinking, which blinded him from many clear contraindicators of his conclusion — including the fact that randomized trials of convalescent plasma at scale is dead simple to conduct. They’re indeed being done elsewhere in the U.S., such as at UCLA — where Tom Hanks donated his plasma — and in other countries around the world. Yet no country has rung the bell on such a decisive victory over this horrible virus.

    Let his humiliation be a lesson to us all.