Exploring Marketing Technology

Category: Data Marketing

Using the power of computing to draw a straighter line between a business and its customers, to the benefit of each

  • AI boosts ecommerce conversions by compressing evolution

    AI boosts ecommerce conversions by compressing evolution

    Every sale on your ecommerce site travels through four phases: Attention, Interest, Desire and Action (AIDA). Before AI, optimizing each meant months of randomized controlled tests, slow feedback loops, and lots of guesswork. AI compresses that personalization cycle dramatically. Marketers using it have a distinct competitive advantage

    To explain how, I’m pulling from two creatures found in nature. They evolved over millennia to do exactly what your ecommerce experience needs. One of them might be the strangest and most instructive thing you’ve never heard of.

    I described those two creatures recently on another blog site, Dyslexic Data. The title: Using data to visualize evolutionary forces. There I explained how the pair have evolved to catch and eat the optimal number of bugs, in a feat of natural selection that evolutionary biologists have visualized.

    Although you should not expect either showing up in an upcoming Pixar feature, in my mind’s eye I do see them, above, fitting in nicely outside The Haunted Mansion. (I describe the significance of this Disneyland attraction in another post, also on Dyslexic Data). 

    First, the spider: A creature so reviled it even has a phobia named after it. I asked AI to make it look friendly. Still, all those legs! It’s hard to avoid the creep factor.

    The other isn’t shown, because, 1.) It is genuinely hideous, and 2.) It spends its phase of life hiding in loose sand, at the bottom of funnel-shaped depressions like the three shown in the foreground. 

    This shy, larval stage of the lacewing, is called the ant lion.

    Growing qualified visits by optimizing attention and interest

    AI is rewriting the economics of all four phases of AIDA. The first two, before modern LLMs, were addressed by the human optimization of paid and organic search, as well as digital display ads. Algorithms helped, but decisions based on feedback loops were slow and data was thin.

    You can think of this process, pre-AI, as similar to the silk spun by the very first ancestors of modern spiders. They weren’t very efficient. 

    The computer simulator described in my evolution blog post makes a case that natural selection and food scarcity inevitably drive generations of spiders to spin webs optimized for the lowest calorie expenditure yielding the greatest caloric intake. If you ran that NetSpinner simulation a thousand times, over multiple generations, each pass would iteratively, gradually evolve to what you see above: the classic Charlotte’s Web bug-catcher. 

    Every single time. Now that’s optimization!

    Similarly, modern machine learning works to optimize your campaigns for better attraction of attention and generation of interest. Here are just three of the mechanisms:

    1. It does a better job of stitching identities in your customer data platform (CDP)
    2. It refines your organic and paid campaigns by providing more actionable data
    3. It continually interprets that data to launch and test multiple ad variants

    Those actions include look-alike modeling, to find more prospects similar to your best customers, and suppression models, to ensure you aren’t “wasting silk” — in this case, your ad dollars — on people who are already customers.

    In these ways and the others below, AI compresses evolution by quickly learning the best placement of its own “sticky silk strands” to capture the attention and interest of ideal prospects.

    From search rankings to AI recommendations: the new attention engine

    Organic search marketing, also known as search engine optimization (SEO), is still important. 

    AI helps there too. 

    It can optimize your product description pages to rank higher in search engines for the phrases your customers use to find the products.

    Rapidly emerging as a counterpart to SEO is agentic engine optimization (AEO … sometimes called GEO for “generative engine optimization”). AEO / GEO acknowledges that attention and interest take place in ways we marketers find difficult to quantify, in recommendations made by ChatGPT and other chatbots.

    Agents such as Optimizely’s Opal can build the data schemas within your product description pages that are invisible to human visitors but instructive to bots.

    Those schemas serve as training datasets for the LLMs. They help persuade chatbots to recommend your products over competitors when consulted by your prospects.

    Measuring Success

    Success metrics in a world of AEO / GEO can be tricky. 

    In the old world, interest was measured by the raw number of people who arrived at your site. Since the emergence of Google Gemini, ChatGPT and their ilk, marketers are noticing significantly fewer visitors. That’s because consideration for many is happening on search engine answer sections and the recommendations served up by the chatbots. 

    Here are two metrics that still work:

    1. An increased return on ad spent (ROAS)
    2. A reduced number of clicks leading to conversion, since people are generally arriving more knowledgeable about your offerings, with fewer questions

    Boosting desire and action through optimized conversion funnels

    I’ve known about ant lions since I was a morbid little boy. As a digital marketer, I recognized they’ve evolved to produce a literal, perfect conversion funnel.

    The ant lion hangs out at the bottom. When it senses an ant has stumbled onto the rim, it flings its shovel-shaped head to toss grains of sand up and over.

    The unwitting ant works against the avalanche as it begins a downward slide. The ant lion counters with more targeted shoveling. In stages, similar to the steps of conversion funnels that are your site’s shopping experiences, any slip-up would mean escape, and a ruined dinner.

    The Action at the end of the Desire phase is, of course, a sale. Improving your conversion funnels to optimize sales, before AI, was slow and imperfect. 

    True, funnel-optimizing A/B and multivariate testing succeeded in the pre-AI world, by using complex math to measure improvements. But today, AI makes these calculations and decisions in an instant, based on more variables than you could ever hold in your mind (the segment of that prospect, the visit source, day and time, and purchase history, to name just a few attributes).

    AI ensures that every method to reduce conversion funnel attrition is put to practice in real time. The result? A greater share of your prospects pass from Desire to Action.

    Measuring Success

    What metrics would tell you if your conversion funnels are performing like a well-dug ant lion trap? Here are two:

    1. Conversion rate, as measured from sales divided by entries into the relevant product description page.
    2. Shopping basket size, since AI can also supercharge your “You may also like” recommendations

    Better webs, smarter funnels

    AI doesn’t change what a sale is. It compresses the time it takes to get there … to serve up the right message and experience for the right person at every stage of the process. 

    Chances are some of your competitors are already using AI to spin better webs and dig more efficient funnels. The question isn’t whether to adopt AI. It’s which phase of your AIDA cycle you optimize first. 

    Start there.

  • All experiments should end with a bang

    All experiments should end with a bang

    In a perfect world, all A/B and Multi-variate Tests you conduct should be as expectantly observed (and ghoulishly revisited) as the one conducted last week, pictured above.

    Before the test launch of SpaceX Starship, Elon Musk said there was a one-in-three chance it would not survive to a successful landing. He was wise to set expectations. It exploded upon contact with the earth.

    I loved that Musk called the mission an “awesome test,” noting “We got all the data we needed.”

    Spoken like a true scientist.

    How do you handle and publicize your failed tests? I hope you celebrate them for the data you gathered. Put yourself in Musk’s shoes: He paid dearly for a pile of burned shards, but in return he got measurements throughout the flight that will help him rocket above the competition (sorry) to claim the profits in manned space flight he is chasing.

    The only true failed experiment is the one where the data gathered is not applied to future efforts.

  • How standard email marketing metrics fall short

    When you’re trying to optimize the profits of your business, most web metrics are unhelpful at least, and deceptive at worst. But what about the world of email marketing ? Kevin Hillstrom, in his excellent Mine That Data blog, gave this example to illustrate how conventional email metrics look at the wrong things:

    Say you have a list of 500,000 e-mail addresses. You send your standard campaign on a Monday. Later in the week, you tabulate your results:

    • 500,000 recipients
    • 20% open rate = 100,000
    • Of the opens, 20% click through to the website = 20,000 visit website
    • Of the clicks, 5% convert and buy something = 1,000 orders
    • Average Order Value = $100
    • Total Demand = 1,000 * $100 = $100,000
    • Demand per Recipient = $100,000 / 500,000 = $0.20 [per customer]

    He compares these finding with what you’d get if you did something called a mail/holdhout test. You compare a control group that does not get the email with a test group that does. For instance, he suggests this breakout:

    • Mailed Group = 400,000 Recipients, $300,000 spent = $0.75 per customer
    • Holdout Group = 100,000 Held Out, $45,000 spent = $0.45 per customer
    • Incremental Lift = $0.75 – $0.45 = $0.30 per customer

    Much more insightful!

    This is why I’m not a fan of open/click/conversion. A mail/holdout test proves the actual value of an e-mail marketing campaign. In this case, we observe $0.30 lift, whereas open/click/conversion yields $0.20 lift. E-mail marketers, why would you not want to know that your campaigns are working 50% better than when measured via opens/click/conversion?

    Kevin goes on to provide other interesting observations from his years of doing this sort of testing. You can’t go wrong by following his blog, and trying his approach to data-driven online marketing.

  • Time is the only solution to privacy objections to behavioral targeting

    Everyone involved in technology and marketing has had this conversation: They’re in a social setting, and the subject comes up about how technology is crafting messages to match consumer behavior. Someone pipes in, “Oh, like in Minority Report? That’s so wrong!” Although Gmail’s ads are customized based on the content of email messages, I’ve noticed that few object to that practice. Instead, they’ll complain about the ads appearing in Facebook — ads that clearly are using information users have given the network about themselves.

    Big Brother Is WatchingIt’s a Catch 22 for online marketers. In order to boost response rates, we need to know more about the people viewing our ads. This work of behavioral targeting sounds like a win/win: “We’ll only provide you with the ads that you will likely care about.” But in practice, consumers get spooked.I’m reminded of direct response research done years ago. It was a survey to find out how people like to be hit up for contributions to non-profit causes.

    Here’s how consumers responded:

    1. Least favorite: Personal asks
    2. 2nd Least: Telemarketing calls
    3. 3rd Least: Direct mail

    What researchers at the time found telling was the direct correlation that existed between disliking a method of asking for donations and its effectiveness in getting them. In other words, personal asks — your sister-in-law selling Girl Scout cookies for her daughter — are most effective in terms of closing rates. The closing rates of telemarketing (back when this was a more viable medium for fund raising) were nearly as good. And a distant third in terms of effectiveness was direct mail.

    So what’s really going on here? The accepted theory is this: We all have limited money to contribute to causes, and we would prefer to put off making decisions about where that money should go. Therefore, the most effective ways of forcing a decision are the least preferred.

    Similarly, we love DVRs, because they allow us to zoom past commercials. They give us a way to avoid participating in commerce. They can’t touch us because we’re averting our eyes.

    Behavioral targeting, if done properly, presents ads that also touch us. Thus, we look for reasons to hate the practice. Privacy is as good a reason as any and better than most!

    So what’s the solution?

    I do think that attitudes are slowly changing, and this change will eliminate privacy concerns as a reason to hate behavioral targeting.

    Here’s an example. Consumers using social media are getting more comfortable with the various personas that they present on networks. They’ll show their “all-business” persona on LinkedIn, and their more casual facade on Facebook. Both are true depictions of the user, but they’re single facets of a full personality. Context determines how consumers behave on these sites. They are becoming accustomed to being watched by friends and business associates.

    Reading Online Body Language

    These same consumers are seeing how they can watch their friends right back. They are learning to “read” a friend’s feelings and preferences, based on online behavior. Consumers are getting accustomed to the online equivalent of body language. Or maybe that’s too strong a word. We’re all quite conscious of what we’re conveying, and body language implies unconscious action. Maybe what we’re doing on social media is more akin to a Kabuki dance.

    Reading these dances is what marketers behind behavioral targeting are learning at a mass level, and turning into surprisingly accurate algorithms. But when it’s done by machines, in the service of a sale, many consumers still insist it’s “so wrong.”

    Perhaps not forever.

    To predict how people will react in the future — especially in areas such as privacy — I try to look at real world analogies. It’s not hard to imagine a web site as similar to a retail store. A while ago I wrote about how stores are analyzing shoppers using arrays of cameras. One system, called PRISM, is finding some unexpected insights into shopper preferences and behaviors. Shoppers aren’t getting freaked out. The prime reason: Most are oblivious to the surveillance.

    But what if the people behind the cameras sprung up from their chairs and raced into the aisles, to say things like, “I saw you were looking at those lawn rakes. Can we interest you in some yard waste trash bags as well?” There would probably be lost sales, at the very least!

    In the online world there are intercepts like this, but they are automated. This automated “intercept” is something people will become more accustomed to over time, as a generation weaned on social media, and used to their online movements being watched, comes into the majority. They will be able to understand that their behavior is being measured by machines as well as their online friends. They’ll realize there is no man behind the online ad “curtain” … just a predictive model.

    Blurring The Lens To Reduce The Creep Factor

    Another trend that will help the acceptance of behavioral targeting is a move toward more explicit boundaries. For instance, I expect an eventual backlash to camera surveillance such as PRISM. But before this reaction can take hold, the boundaries of store monitoring will likely improve. This improvement in technology will, for once, be toward discreetly blurring the “analytical lens” — instead of making it sharper.

    Radio tomographic imaging is a new way to study store traffic. It uses arrays of extremely inexpensive radio transmitters and receivers, placed around buildings, to display people moving within. Below is a demonstration of the technology.

    The benefits of this technology over cameras, at least to marketers, is cost — both in equipment and the labor of monitoring. Because people become moving “blobs” of color, it’s easier for computers to analyze traffic patterns and behaviors. Less can sometimes be more. Combine this information with RFID signals and you have a way to track a shopping basket all the way to check-out.

    Imagine how this could be used:

    Before he leaves that store, a consumer might someday pass an “intelligent end cap.” This smart store display knows — based on radio tomographic imagery, enhanced by RFID data sent to it — when to light up. The end cap would know the consumer has a yard rake, and offers him the trash bags he forgot to add to his list.

    This hypothetical consumer will probably be grateful, since he really did need the bags for his yard project. And after all, he was just a blob of color moving through the store.

    Ironically, this is the level of detail being used for most online behavioral targeting. This is the “privacy invasion” that is causing such a fuss on Facebook and elsewhere.

    Over time, consumers will become more comfortable with behavioral targeting’s perceived betrayal of privacy.

    That will leave only one valid objection the technique: They just don’t want to buy more stuff.

  • Employers of marketing and PR pros are undervaluing a key skill

    Online newsroom specialists iPressroom recently surveyed businesses to see what sorts of skills they are looking for in their marketing and PR pros. The survey had a small sample size, as many of these do, and this report’s many charts read far more into the results than can be reliably concluded. But I credit its authors for noting something that jumped out at me as well:

    Rather than focus on attracting or pulling visitors to their website by publishing high quality content and researching popular language, organizations appear to be more interested with pushing out messages to “friends” through social media, even though, in many cases, those messages include hyperlinks back to their own websites. Until these organizations learn to develop a more sophisticated approach to building and managing landing pages and web content management on their websites, they will have difficulty evaluating their return on investment for these emerging channels.

    (Emphasis mine.)

    I found this report by reading a trendy headline somewhere. It proclaimed that marketing and especially PR executives are expected to possess skills that most are still scrambling to master. Here is a sample chart showing the data behind this assertion:

    The digital skills expected of marketing and PR executives

    The employers surveyed should be commended for understanding the pressing demands of social media. However, they’re overlooking an equally important skill in their communications hiring checklist. They must hire people who understand the importance of good site content and how to measure its value. This is essential to making long-term gains from social media and search engine efforts.

    It’s not enough to know how to attract eyeballs. The owners of those eyeballs had better find something on a site that’s worth experiencing and sharing.