Facebook: Not the Place to Acquire New Customers

PetFlow shared their frustrations with Google AdWords and Facebook advertisements alike; namely when it came to the cost of acquiring new customers. When originally targeting the customer of PetFlow products, they started with Facebook and were spending $300-400 each to acquire a customer. “Facebook sucked” said Alex. “So, we went to Google Adwords and focused on building the right audience for our product.”

They gained the target audience knowledge from adding a short survey to the follow up post-cart interaction page asking shoppers to answer a few short questions. Once PetFlow found their target market in women ages 35 and above, who were fans of Amazon (See: buy products online), and were also shopping higher end retailers like Bloomingdales, Nordstrom and others and were fans of those pages. Eventually, PetFlow came back to the Facebook drawing board and had great success: they brought down the cost of a fan to 35 cents.

Facebook: Using the Platform to Build and Extend Community

Where Facebook really did help, was in solidifying the PetFlow relationship with these newly acquired fans.  The first step was to rename their Facebook page from simply “Pet Flow” to a name and description of their service “Pet Flow – Pet Food Delivery.” They also included other customer experience add-ons like shipment confirmations, promotional emails, contests, blog posts and they even asked customers to send photos of their pets in via Facebook. They also ensured their Facebook page logo was included across all of these customer touches.

Not only was the PetFlow Facebook page growing from new customers, but adding fans who liked the funny animal photos they shared on a daily, (excuse me –20 times a day!) basis. The PetFlow team began posting funny shots of cats “MA, the meatloaf!” and dogs alike and drew more interest in from friends of their target market browsing through their Facebook feeds. The social result is a pet food delivery service that delivers in more ways and more channels than one.

The post Best Practices in Social Commerce appeared first on Baynote.

Tablets and Mobile devices

Today stores are displaying more and more “tablet” offerings like the kindle, e-reader, and phablet.  Gian noted that there are now over 60 million tablet users and over 137 million mobile users. Internet use via desktop computer in online up 20% whereas internet use via mobile device is up a whopping 389%.

What’s most important to the customer?

With passage of the recent sales tax bill, the typical internet retailer is understandably worried about the impact to online sales.  In fact, the ComScore study found the top reasons customers bought online were free shipping, online deals, no sales tax, fast shipping and in-store pickup. ComScore did a quadrant analysis to determine which items were in high-derived importance and high satisfaction during the shopping experience. The three most crucial pieces of the shopping experience? Ease of checkout, variety of product and ability to track products online.  23% of consumers in the study also indicated that “showing me other items I might be interested in” during check out was in the top 10 most important features.

Commerce vs. Marketing

Companies like AT&T and Procter & Gamble are actively engaging in advertising through commerce powered ads resulting in a tight feedback loop where “commerce has become marketing and marketing has become commerce”(Rishad Tobaccowala). Most importantly noted, the expectations of consumers today are at an all-time high in terms of control and convenience; particularly in a seamless experience, thus adding complexity to the lofty consumer expectations of all online retailers.

The post The Pulse of Online Shopping at IRCE 2013 appeared first on Baynote.

Machine learning is supposed to be about exploration – the ability to let the machine run and freely explore the various connections between data points.  When additional logic is introduced that limits this exploration, it can bias the output.  This might mean the machine is told to look at a specific input or a specific feature in the data.  Usually this is unintentional.  Here is how it might happen: Let’s say a data scientist works in-house at one of the new Bay Area labs recently established by big retailers.  If they have an inkling that a certain input or feature might create any given outcome, then what they are really doing is degrading the organic capacity of the machine to explore.  Their instructions will send the algorithms down a more defined path – one where it is looking for something and once it finds it, machine learning exploits it.  For some applications, this is a decent outcome – for others, not so good.

So if you plan to hire data scientists any time soon, make sure you philosophically agree on the level of exploration you are willing to accommodate.  While it is still early days for many big data thinkers, my suspicion is that we will actually become much more biased, by domain over time, if we are not careful to allow our data scientists the room to roam freely.

The post Beware the Bias appeared first on Baynote.

What data should algorithms consume in order to provide consumers with a personalized experience without infringing on personal privacy? What role do humans play in determining how technology leverages data? The articles detailed below address these subjects and present interesting perspectives that must be part of the larger debate surrounding personalization and privacy

“Careful: Your Big Data Analytics May be Polluted by Data Scientist Bias,” GigaOM – Every day we see articles exalting the power of big data. But with great power, come great responsibility. This article by Baynote’s very own data scientists, Haowen Chan and Robin Morris, explores how companies leveraging big data must be aware of potential data bias. From only gathering easily collectable data to improperly ruling out data due to preconceived notions, data scientists must constantly ensure they avoid introducing bias into analytics. If they don’t, the intelligence produced may not only be misleading, but also inaccurate. To help prevent data bias, Hawoen and Robin recommend four strategies: employ domain experts, look for white spaces, open a feedback loop and encourage data scientists to explore.

“UncommonGoods Reveals More of its Online Catalog with Automated Recommendations,” InternetRetailer – UncommonGoods, as the name suggests, is in the business of selling one-of-a-kind items ranging from home décor to gag gifts. As such, identifying affinities between items in order to provide relevant product recommendations is not an easy task. In this case study, Internet Retailer’s Amy Dusto examines how UncommonGoods turned to Baynote to help increase the number and relevancy of product recommendations offered to shoppers. With Baynote, UncommonGoods implemented algorithms to quickly search through entire product catalogs in order to identify and supplement recommendations handpicked by the merchandisers. The result is a more than twofold increase in recommendations showed to consumers and a 19.9% increase in conversions when customers click on a Baynote generated recommendation.

“The Value of Big Data Isn’t the Data,” Harvard Business Review – In this article, Northwestern professor and Narrative Science’s CTO, Kristian Hammond, explores the importance of the human link between big data, the story it tells and the action it leads decision makers to take. Currently, businesses are focused on putting technology in place to perform big data analytics. While this is a necessary step, the true value comes from taking the results of the data and developing a concise narrative that can then be turned into action. Kristian argues that, “To get scale from data interpretation, we have to embrace the power of the machine to extract and explain the data that it and it alone is in a unique position to analyze and then communicate.” At Baynote, we certainly agree that big data and machine learning play an important role in communicating business intelligence, but we must remember the importance of the human element – both developing the algorithms that machine learning runs on and the potential pitfalls of data bias discussed in the above GigaOM article.

“Personal Information is the Currency of the 21^st Century” AllThingsD – Information is power. Just ask Google or Facebook. While these companies may sell products just like any retailer, these organizations’ true value lies in the huge quantities of consumer data. In this article, Tom Cochran, CTO of Atlantic Media, details how information is the new currency driving our daily lives. He concludes, “There is a zero-sum relationship between personalization and privacy. To get the personalized digital experience you want and have grown accustomed to, you have to accept the loss of your privacy.” Is this a fair exchange for giving up your online privacy?

The post May Personalization Roundup appeared first on Baynote.

But what about the business needs?  Ecommerce companies must also focus on maximizing key metrics like engagement, revenue, conversion rate and average order value. Hence, personalization should display that content which meets the visitor’s needs but also seeks to maximize key metrics that the business cares about.

Personalization is a means to an end, not the end itself. The logic goes like this: If a single, generic experience for all users produces a given set of results like a 1.5% conversion rate, and targeting content by segments produces better results like a 2.2% conversion rate, then delivering unique content or experiences to each user should produce the best possible results to maximize key metrics.

This logic is reasonable but based on a false premise that one to one personalization is required to see this sort of conversion rate lift.  Delivering the most relevant content for a given user in their current session can produce great results.  Whether this content or the experience is entirely unique is irrelevant.

Website Personalization, Personalized Email and Recommendations Deliver ROI

Baynote recommendations users typically convert at anywhere from 2 to 6 times the rate of non-rec users.  So it follows that the more an ecommerce site delivers personalized content or product recommendations, the greater the chance for increases in conversion rate becomes.   All retailers have the opportunity to personalize recommendations throughout their site.  From the home page, to category, product detail pages and on through to cart pages and follow up emails, there are many opportunities to deliver a cross sell, an up sell or even a special promotional item option to your shopper.  Why let those opportunities slip by without personalizing?  In each case, with contextual personalization, the recommendations are dynamic and follow your shopper on their journey – facilitating a satisfying experience for them, and improved KPIs for you.

Part 1 of 3: Personalize for the “Now” 

Part 2 of 3: Personalize Based On User Type

The post A Personalization Manifesto – Why Bother? Part III of III appeared first on Baynote.

The next step up is shoppers that you recognize and have seen before, but are still anonymous.  We call these known users.  With these shoppers, start by leveraging current intent and then look at what you may know from past experiences. For example, a user arrives on the homepage by entering the URL directly in their browser.  Baynote personalization knows what they were looking at the last time they were on the site and uses that information to show relevant products. If the visitor then navigates to a different area or the site or searches for something else our technology instantly recognizes the change in patters and starts looking at their current intent instead of what they were looking at last time. So start by welcoming the shopper back by letting them know where they left off, and then follow them through their current shopping trip to see what is different and personalize accordingly.

E-Commerce Personalization Has a Name

Finally, we hit the mother lode – the named visitor.  This is someone who may actually need to log in, has a profile with you and some purchase history.  The key here is to entice that second purchase (or third, or fourth – but the second is the hardest!) Personalization for these shoppers starts by first using current intent and then leveraging insight about their segment membership, past purchases or user attributes provided by a CRM or marketing automation system. Past purchases can be used to engage the visitor in a dialog around complementary products for their recent purchase or to show them new arrivals that are similar to past purchases. Baynote also uses segments and attributes to filter results. For example, a user is identified as a high value purchaser who typically spends more than $500. Using this information the Baynote system can filter results that are returned to show higher value items.  For the named visitor, you must treat them as familiar, but be careful not to cross the line.  No one wants to feel stalked or followed.  Keep it light, keep it contextual and keep it appropriate.

 See Post 3 of 3: Why Bother?

The post A Personalization Manifesto – Personalize Based On User Type Part II of III appeared first on Baynote.

A user’s current intent is the best indicator of current needs. Intent is what the person is trying to achieve at that moment. Intent is expressed through behaviors of the user including the search terms they use to navigate to a site or in onsite search, the links they choose to click on, and which products or content items they engage with.

Personalize Experience for the Now, Not the Past

Many supposed personalization systems rely on customer profile data to deliver personalized content. These systems suppose that available user data such as demographics, home location, or past purchases are sufficient to predict what a person is likely to be interested in today. While profile data can tell you if the individual is more likely to purchase a certain brand, spend over a given amount, or what’s popular in their area; profile data cannot tell you what the person is looking for right now.

In addition, profile data is not particularly helpful when you have never seen the visitor before (anonymous visitors), the person is shopping for someone else (gifts), or when the person is buying something new and they deviate from past purchase behaviors (business traveler buying a vacation package).  So when thinking about personalization strategies, be sure that you look for solutions that observes each user’s behavioral cues and takes advantage of onsite search terms to determine what the person is looking for right now.  This step should be backed up by machine learning/data science that determines patterns in behavior and product affinities on a 24X7 basis.  From there the system can infer what content are most likely to be interesting and then act in real-time to deliver the compelling content, in the moment.

See Post 2 of 3: Personalize Based On User Type

The post A Personalization Manifesto – Personalize for the “Now” Part I of III appeared first on Baynote.

Video recommendations are similar to product recommendations, but are more interactive and a good tactic to use on a product detail page. Zappos uses them on almost all of their product pages.  In our recent webinar, with Sucharita Mulpuru from Forrester, she commented on her favorite shoe website and discussed how the site played short clips of women walking around in the shoes, so the shopper can see them in “real-time” before committing to a purchase. But videos can also be informative such as how to install an item, or best ways to use a product.

If you think about it, providing video content to your shopper is like adding another layer of trust. You provide them with more information to make a conscious decision.  How do the shoes really look on someone’s feet?  Making the purchase may be easier when one has theoretically “walked in another person’s.”  Also, when you add in video recommendations, you have taken the user to a more interactive page.

Similar, but not as scientific and intent-based are YouTube recommendations of other videos you might be interested in – like after watching the “world’s cutest frog”, you may find other people who watched the entire frog video also watched “how animals cross the road.” That way you’re still laughing, and you’re also still on YouTube. It’s a win-win.

Not only “dressed” for retail, video recommendations can offer content focused websites a cleaner, more helpful interaction with their visitor.

Learn more on Baynote’s solution: content recommendations.

The post Video Recommendations at Face Value appeared first on Baynote.

Type 1: The proportion of users who use the recommendations is the same for groups A (KPI-optimizing recommendations) and B (traditional item-based collaborative filtering recommendations).

Type 2: The proportion of users who use the recommendations is lower for the new KPI-optimizing algorithm than for the more traditional algorithm.

However, for both types of sites, the total revenue was the same or higher for the group that was seeing the recommendations from the new KPI-optimizing algorithm.

Think about this for a minute.  For sites of type 2, a smaller percentage of the users use the recommendations, but the site’s revenue is improved.  This can happen one of two ways, either those using the recommendations are converting more often and/or spending more per purchase, or those users who do not use the recommendations end up converting more often and/or spending more than those who do use the recommendations.

Recommendations Provide Alternatives

Isn’t this the opposite of what we’re trying to do with a recommendation system?  Don’t we want the users to use the recommendations, as that will lead them to find the items that they want to buy?  Yes and No.  What happens if the user has already found the item they want to buy?  It’s right there in front of them, on the page they’re looking at.  Traditional item-based collaborative filtering recommendations at this point will show them alternatives to the item they’ve chosen, which may cause them to start browsing again, and possibly slow conversion to purchase.  KPI-optimizing recommendations show different sorts of items – ones that in some sense “go with” the current item – but which may be less distracting to a user who has already found the item they want.  Net result – users who don’t use the recommendations convert more.

I should also point out that the KPI-optimizing algorithm does also work as intended – those users who use the recommendations coming from the KPI-optimizing algorithm typically convert at a higher rate and with a higher average order value than those interacting with the recommendations coming from the more traditional algorithm.

That’s what’s going on with sites of Type 2.  What about Type 1?  Why did the proportion of users who used the recommendations not change on these sites?  Answer – The merchandising layer.  These sites had filters applied to the output coming from the recommendation algorithm that restricted what was shown to the user.  So the distraction factor was still there, but the users were being distracted by items which had the highest expected KPI – net result, conversions and revenue were both up for these sites.

An Important Combination

It has often been argued that the particular algorithm behind a recommendation system has little effect on the overall system.  We’ve found that the algorithm is in fact important, and can increase lift, but exactly how it does it is a complicated combination of the algorithm, the merchandizing, and the psychology of the web-site user.

The post Algorithms, Merchandising and Psychology – Part 2 appeared first on Baynote.

But instead, I want to write about recommendation systems, where the recommendation algorithm is only one part, and where merchandizing, presentation and psychology also play a role.

The Layers of Recommendations

A typical recommendation system has lots of parts, but the three main parts when serving recommendations are the recommendation algorithm, the merchandising layer and the presentation layer.  All three are important to the success of the system.  The recommendation algorithm provides the basis, identifying the potential items to show to the current user.  The merchandising layer allows the website owner to customize the recommendations, applying filters, blacklists and pins to modify the output from the recommendation algorithm to fit a set of desired marketing criteria.  Finally, the presentation layer puts the recommendations in front of the user, at which point psychology takes over.

So let’s look at each of these parts in turn, and then I’ll discuss some really interesting results we’ve had recently which show how they interact in unexpected ways. (See part two of this piece)

Key Performance Indicators

As written in earlier blog posts, I’ve been developing a set of recommendation algorithms that explicitly target customer Key Performance Indicators (KPIs).  So, for example, a customer could specify that they wanted recommendations that optimize for conversions, or more typically, revenue, and the algorithm would generate a set of items with the highest expected revenue when presented to the current user.  This is very different from more typical recommendation algorithms, which tend to find items that the current user would rate highly, or which users similar to the current user have bought — the hope being that the user would then go on to buy one of these items.  My new algorithm does not just hope, it makes the KPI explicit.

The merchandising layer is, from a data science viewpoint, a bit of an oddity.  We would hope that the data should tell us what the best items are, but often a website will want a little more control over the users’ experience.  They may want to ensure that the items shown on a page are alternatives to the current item, or are a mix of alternatives and complimentary items.  Or they might want to pin things like gift cards at particular times of year.

The presentation layer is the website, and it makes a difference where on the page the recommendations are presented, and in what format.   But, beyond discussing best practices with the website owner, there’s not too much we can do on the presentation end of things.

It is the combination of all three of these that is finally shown to the user, and that’s where the psychology comes in.

We’re running A/B tests of the new KPI-optimizing algorithm vs a more traditional item-based collaborative filtering algorithm on a number of sites and we have found some interesting differences.  Read my next blog post to find out how KPI optimization differs from collaborative filtering later this week.

The post Algorithms, Merchandising and Psychology – Part 1 appeared first on Baynote.