Others Online Blog

What We Do (in LOLCat / LOLspeak)

Too often companies describe what they do in ways that the majority of people just don't understand. We're no different. So here's an attempt to describe what Others Online does in a really simple language called LOLSpeak.

Growth Chart for Others Online

Neal and Sam just sent me this chart on the growth of affinity profiles since July 2008 (when we started within a new data center). Pretty impressive, if you ask me.

When Does Behavioral Targeting Work?

A lot of publishers in the advertising/media business don't understand the situations when behavioral targeting will almost always result in better eCPM rates than contextual targeting.

For the sake of this post, let's disregard the fact that there different methods of behavioral targeting, each with their strengths/weaknesses. We'll stay with the standard definition of behavioral targeting -- when you target online consumers based on their past online behavior. Contextual targeting is when you target them based on the current page they're on, and what that page is about.

My rule of thumb is that contextual targeting always performs better than behavioral targeting, if the context is clear and of commercial value. I've got over 5 years experience in the contextual/behavioral landscape and have proven this over and over in various tests. However, if browsing context is poor -- either no context whatsoever (only images on the page, for example) or there's a high signal/noise ratio (social media) -- or of no commercial value, then behavioral targeting will always perform better than contextual targeting.

Let's consider several types of Web sites, indicating for each the targeting that generally works better (big "X"), never works better (no "X"), or sometimes works better (little "x"):

Type of Site	Contextual	Behavioral
Search	X	x
SEO	X
Community/social		X
News & reference		X

Search sites are all about current context. Contextual targeting almost always works better, but there are exceptions -- such as when the search keywords have no commercial value and the user is searching images. In that situation, contextual advertising doesn't work and you're better off targeting the person based on their previous interests.

I define SEO sites as those that focus entirely on search traffic. They optimize their content and links specifically for indexing within Google search results. And as a result, >95% of their traffic comes from Google -- users click on the link within the search results, go to the page, then usually leave within 1-2 pageviews. Google AdSense tends to work best on these pages, offering "secondary search results". Behavioral targeting doesn't work at all.

On community or social media sites, page context is usually very poor and many sites require logins, which means Google can't index the pages to determine context. So contextual advertising programs (like Google AdSense) don''t perform well. Behavioral targeting can work much better, if done right. People tend to spend quite a bit more time on these sites, and provide a lot of good information about themselves and their interests. Our affinity profiling software knows how to weed through the noise and figure out what people care about. And recent tests have shown 70% lift in eCPM rates using that for targeting instead of page context.

The content in news and reference sites is usually of poor commercial value. Contextual targeting can therefore be difficult, and can also result in badly placed advertising. So targeting the user based on his/her interests (instead of the page) will almost always work better than contextual targeting.

Summary

Targeting someone based on their previous behavior/interests doesn't result in better performance if the user is on a page narrow in context and commercial value. The reality though, is that there are a lot MORE pages viewed these days that are very poor in context -- blogs, social networks, videos, images, news, dating, email, family pages, etc. In that situation, targeting the person is much more effective than targeting the page.

Audience Targeting vs Site/Category Targeting

Interesting comment from a recent Adotas blog post regarding targeting people instead of pages:

In a recent eMarketer study, 90% of advertisers said they plan to use ad networks in the coming year and the top two factors that they will use to differentiate networks were the quality of inventory and the audience targeting. Networks that have quality inventory and rich audience targeting capabilities should be some of the first places marketers and agencies look in 2009 and beyond to keep revenues flowing at lower costs and greater efficiency.

Thank You for Making 2008 Our Best Year!

I've never been one to think in terms of calendar years starting/finishing -- rather I view new years just as I do new days or months, as simply a progression of time. I suppose it's because of this that I've never embraced either resolutions or predictions. But I do like to recognize achievements now and then.

Others Online started building our core technology in 2006, and the company pretty much flat-lined for most of that year as we dealt with an emotionally upset and litigious [former] consultant. So 2006 pretty much sucked. In 2007 we got back on track, added some amazing investors, launched consumer services and grew that user base nicely by the end of the year. But by January 2008 it was apparent that our affinity profiling technology was more valuable to online media companies (B2B) than online consumers (B2C). So 2008 was the "year of the pivot" for us.

As I look back on 2008, I am extremely proud of what Others Online has accomplished, notably:

• Successful shift from B2C to B2B, earning valuable customers, real revenues and an AMAZING pipeline for 2009!!
• Quietly launched our Affinity Profiling Platform in mid-July, moving our data center infrastructure from our own servers to an elastic, cloud-computing infrastructure -- thereby allowing us to scale much more rapidly to meet demand.
• Since July, expanded our cookie footrpint to nearly 80M.
• Since July, developed over 52M rich profiles comprised of over 3B individual affinities, based on almost 300M Web searches and well over 1B page views across nearly 1M sites.
• Drove targeting for >500M ads, resulting in an increase in eCPM rates exceeding 50%.
• Added Sam Tingleff to the Others Online team.
• Added a half-dozen incredible angel investors -- the caliber of people whose accomplishments humble mine.

All this despite the declining economic conditions in 2008! So while I'm not one to generally congratulate ourselves on a job well done -- rather I'm much more inclined towards the "only the paranoid survive" school of thought -- I am quite proud of what we've done as a team in the last year, and I'm ready to face the challenges of 2009. In fact, I believe 2009 will be a much better year for this company than 2008!

I'd like to especially recognize Mike Dierken and Neal Richter for their dedication and hard work throughout the year. These guys are machines -- they continually challenge me, themselves, and each other to outperform -- and it's because of them that we've been able to accomplish so much as a small team.

Happy New Year, and thank you all for your continued interest and support. May we truly add value to your business, and money to your coffers.

- Jordan Mitchell / Founder and CEO

Behavioral Data: Category vs Affinity vs Intent Data

Your audience tells you a lot about themselves in their actions and what they pay attention to online. As an ad network or online media company, how you gather and use that behavioral data is critical to your business. But given the sea of data available, what is the right behavioral data to gather and use for targeting?

There are three different types of behavioral data along the "behavioral targeting continuum": category, affinity/interest, and intent data. This post will define and explain the value of each.

Since this is not a short post, let me just start with the key takeaways:

• As with all targeting, reach/scale becomes more limited as you apply narrower targeting.
• Category data provides the shallowest targeting, least ROI yet broadest reach.
• Intent data provides narrow targeting and strong ROI but it's one-dimensional targeting with very limited reach. 
• Affinity (interest) data provides the best balance of breadth of reach and depth of targeting. Think about yourself: of all the Web pages you browse to, only a small portion exhibits purchase intent whereas all of it reflects your interests!
• Affinity data offers greatest ability to tune (via the affinity score), multidimensional utility, most flexibility, and therefore the greatest lift across the largest % of your inventory.

Not All Behavioral Data is Created Equal

By definition, category means "a collection sharing a common attribute". Category data allows you to put users into conventional and broadly-defined behavioral "buckets", with labels such as travel, business, entertainment, etc. A user is attributed to a behavioral category if they've been to a site that falls within the particular category, perhaps within some sort of time frame. The value of this approach is breadth in reach and impression volume, but the targeting is quite shallow and definitions can be blind AND arbitrary -- what does an Urban category really mean and what are the sites contributing to that definition? As a result, category-based behavioral targeting (or retargeting) may not perform any better for advertisers than site-based (contextual) category targeting.

By definition, intent means "the planning or desire to perform an act". Intent data results from an explicit indication of purchase intent -- these users are in the market for specific goods/services. A user is indicating "intent" if for example they've added goods to a shopping cart, calculated a mortgage loan, checked airfare prices, etc. Clearly there's value to targeting based on intent data, but the challenge is the limited reach and scale. As a simple demonstration of that point, what % of your online time or page views are YOU indicating purchase intent for big ticket goods and services?

By definition, affinity means "attraction or kinship to something". Affinity/interest data allows you to understand what your audience is interested in -- brands, activities, products, hobbies, places, etc. A user is said to have a high affinity to "iPhone applications" if they've explicitly indicated it (typed-in within a search field, social profile or blog comment) or gone to a number of Web pages about iPhone applications within a short period of time.

There are several attributes to affinity data that differentiate it from category and intent data:

• Affinity data is comprised of keywords and phrases, the lingua franca of SEM and SEO. There isn't a marketer on the Web who doesn't "speak keyword" or know what keywords work best for them, and those definitions are not abritrary. With keywords/phrases at the core, affinity data is then rolled into categories, personas, or custom channels/clusters.
• Each affinity attribute is coupled to an affinity strength score which allows you to target based on a user's level of interest. This score evolves continually based on time and intensity, much like the attention span of users, allowing you to achieve the right balance of scale/reach and ROI.
• Affinity profiling is deeper, since it involves the ongoing collection/analysis of what users are searching for, reading about, and publishing (through their blog, social media, etc.).
• Affinity data therefore has much broader utility, since you may use category ("travel") AND keyword specificity ("aspen, skiing"). Affinity data can also supplement other forms of targeting.

Achieving eCPM Lift and Volume

The average US broadband internet user views 138.1 pages per day, and is online at least 24 days each month. Say each page has 3 ads on it -- that's almost 10K ad impressions for each Web user each month! That's a lot of inventory per unique, especially if your company is a top 50 US ad network.

Let's say you are an ad network. How do you offer better targeting, and optimize both advertiser ROI and your own eCPM rates? Assuming the cost isn't overly prohibitive, leveraging intent data will work but at low volume -- resulting in maybe 100% lift across less than 5% of your reach. You're already offering category targeting (site-based), so it's not a big step to also offer category retargeting across your network. If you do, you'll achieve maybe 5% lift across less than 50% of your reach (you'd only offer category retargeting for your best categories, after all). Both of these approaches are one dimensional approaches however -- targeting single, binary, behavioral attributes.

To get the most lift across the greatest % of your audience, you're best off to layer in affinity data across your entire audience. This provides you with most targetable attributes per user, the greatest flexibility in how you use it (narrow vs. broad targeting, single or multifaceted targeting, etc.), and the ability to tune until you've achieved the right balance of ROI and volume.

Here are a few "for examples" highlighting what you can do with affinity data:

• Tighten your current category targeting ("consumer electronics", for example) with affinity targeting, by filtering to only those who have searched for "iPhone applications" (very narrow), or have shown any interest whatsoever to "iPhone" (broader).
• Reinforce the value of contextual targeting by targeting both current context and previous affinity to the same category. This method of "context disambiguation" can dramatically increase the value of your contextual categories.
• Further segment intent data (such as "in-market auto buyer") with affinity data, narrowing the targeting to hobbies ("skiiing", "climbing") or personas ("young professional").

Of course, the strength of your affinity targeting depends on the breadth and depth of affinity data collection. That's why Others Online offers a shared data model, allowing all participants to pool affinity data collection and scoring so that every participant achieves a wider footprint of touch points than they have access to themselves.

Affinity Data Offers Depth, Breadth and Multifaceted Targeting

Your audience tells you a lot about themselves in their actions and what they pay attention to online. If you're not collecting and using that valuable information, you're leaving money on the table -- especially these days when every advertiser is focused on eliminating wasted impressions.

Category and intent data offer single dimensions that allow you to expand your targeting capabilities at varying levels of resulting eCPM lift and volume. To really move the needle though, both for your advertisers and your business, supplement your targeting with multi-dimensional affinity data to offer the broadest AND deepest level of targeting across your network. You'll be glad you did.

Apologies for the Comment Spam

Looks like a company called WideCircles has been auto-submitting comment spam to this blog on a regular basis. More news about WideCircles comment spam ...

The Coming "Ad Network Shakeout"

This post was originally posted over at MetaMuse, Jordan's blog ...

Any of you who keep up on the online advertising market through related blogs, articles, etc. have certainly noticed the rapidly developed "ad network shakeout" meme. The story goes like this -- as the economy worsens, companies will continue to decrease their marketing/advertising budgets, and the hundreds of new ad networks that have popped up over the last few years will have to compete more vigorously for the diminishing dollars available, thus resulting in a shakeout where many won't survive.

While I'm in full agreement, and feel their survival depends on technology, targeting and optimization (an ROI differentiation), that's another post altogether. But I did want to note a few smart comments I read recently, made by folks within ad networks. Philip Smolin of Turn notes in this article:

Some media buyers still think of contextual and behavioral targeting as separate ‘silos’ – you choose one or the other, but you don’t use both at the same time. And in fact most advertising platforms don’t enable you to combine them even if you want to. But in reality both are very powerful predictors of advertising performance – you get better results when you combine the power of the two.

Some of our algorithms at Others Online combine the power of both contextual and behavioral, and we've also found this to improve results. Contextual targeting (behavior now) generally outperforms behavioral targeting (past behavior) except when page context is just really bad. And it's just logical to reinforce your contextual targeting with behavior you've seen in the past.

Who’s going to survive? The platforms with unique technology or differentiated solutions that make advertising more effective – they’re going to thrive. Behavioral will be a key component because of its effectiveness in display advertising.

Behavioral has always been about display advertising, and yes it will be key. But what about text ads and other keyword-based (not category-based) targeting? I find it interesting (and exciting) that Others Online is the only company today offering behavioral targeting technology for keyword-based ad inventory.

Another interesting comment made by Joe Apprendi, CEO of Collective Media at Behavioral Insider (link requires password/registration, unfortunately):

It used to be people bought media with a site-centric mentality. It was a legacy of the print planning and buying business. You looked at what sites indexed or reached your target. Today that site-centric mentality probably represents about 70% of the overall display media spending. But there really is a shift going towards this audience-centric buying methodology. Marketers and largely brand advertisers are recognizing that there are alternative ways to reach target audiences. We believe that no matter what the overall demand for display media, there is a shift happening from site-centric buying, which is usually more expensive and the softest area of the display market. It is going from 70% to 50% very fast. A lot of that is reliable behavioral targeting, not absolutely dependent on what site you are running on.

Couldn't agree more. If I'm an advertiser, I'm only interested in reaching an audience with certain product/market-oriented affinities -- otherwise it's just a waste of an impression ("spray and pray"). I hear this a lot from ad networks, and it's frustrating them because most of them have NOT indexed their audience -- they've only indexed the sites in their publisher network.

Rapid Scaling of Affinity Profiles

Since our last company milestones post, we've continued to expand our footprint. In the past few weeks we've more than doubled the number of affinity profiles in our system -- to more than 12 million! And we should double that again over the next few weeks. In the last week we've signed 3 new partner contracts representing over 70M uniques per month. Plus, as a result of a productive trip to Ad:Tech in NY last week, we've got many more interesting deals in the pipeline and further deployments planned over the course of the next 1-2 months.

Say Goodbye to the Others Online Toolbar

We've made the decision to turn off the Others Online toolbar, and no longer offer it to consumer/individual users. We've already begun de-listing the toolbar from a few download sources, will notify current users via email, and then pull the plug.

It wasn't a very difficult decision actually. The core of Others Online has always been affinity profiling, and the toolbar was an experiment to see if a consumer product connecting people by their affinity would get enough usage to (a) seed our initial affinity profiling algorithms with lots of good data, and (b) develop into a successful business model. The former very much happened while the latter never did. And we never intended to spend a lot of time/money to "push" toolbar distribution, only to see if we could create "pull" (meaning lots and lots of people download it via word-of-mouth rather than our marketing).

Given the current state of the economy, we knew we had to get that "eye of the tiger", be fleet of foot, and cut unnecessary costs. The toolbar services were a dog -- we didn't have enough users or earn enough toolbar-only revenue to justify the toolbar-only costs. And at the same time, we're seeing great early results offering our Affinity Profiling Platform directly to medium/large publishers and networks.