Bing

The good folks at Bing Canada were kind enough to invite me to their Bing.ca launch last week.

They're good people over there. Very friendly and genuinely warm.

It's taken me a week to really formulate coherent thoughts that I could write here.

From a search perspective, Bing has very good percentage margins. Of course, I don't bank percentages, I bank dollars - and therein lies the problem for Bing: getting the volumes while maintaining the clickthrough and conversion rates. In the end, of course, search engines dominate through relevancy, and it's through relevancy that Bing will win volume.

It was observed at our table of quantitative search folks that Bing was more consumer oriented. It wasn't necessarily designed for engineers. It has visually nice pictures on the home page. It's usable and feels uncluttered. It's optimized for the top 3 most relevant results to be there. There is a preview mechanism so you can get a flavor of the site before you visit. The results, we speculated, were to be optimized based on consumer relevancy.

It reminds me of the 1990's in a way. I used to use Yahoo! for Entertainment and Alta Vista for business search. Different search engines just seemed to be better at different things. Along came Google, which worked really well for both - and Google Scholar: where they house all their academic searches - and that was pretty much it for Yahoo! and Alta Vista.

If the Bing play is really to 'search and decide' (or put in a more web analyst language: "search and convert"), then it really has positioned itself to be a consumer search engine. So, much of the booing and buuu-urnsing you hear from engineers, developers, academics, doctors - and yea, even web analysts - might very well be predictable. Perhaps Bing simply isn't optimized for that category of people. Maybe people out there - consumers - really are feeling "overwhelmed by search" (a statement that I must admit I still don't feel like it applies to me.) Perhaps it's not about me after all.


It's by being relevant to the consumer that we might see Bing make inroads. It's kind of reassuring (from a creative destruction perspective) that Bing will, like Google, live and die on its algorithm.

And let the best algorithm for the best market win.

Has anyone really been far even as decided to use even go want to look more like?

You read that right.

The intersection of text analytics, social analytics, and neuroanalytics is incredibly interesting and useful.

It's an old meme, and you read about it's origin here.

"Has anyone really been far even as decided to use even go want to look more like?" is a 4Chan-ism. I suspect that it was written by a linguist. Awesome troll is awesome. It can be roughly translated to:

"Has anyone really decided as to even go that far in wanting to do to look more like so?"

Subject - Anyone Verb - decided (modified by "really" adverb) Direct object to "decided" - "that" (pronoun modified by "far") Noun clause that clarifies "that" - "wanting to do" (gerund phrase) Do what? - "to look more like so"

Meaning:

"Has any video game company really taken such measure to make a game so realistic?"

This is the kind of stuff that if a human has a hard time interpreting, a machine is going to have especially hard time codifying and returning some sort of valid output.

This all goes beyond just identifying words in a stream of text and trying to assign some sort of value to them. To be sure, volumetric measurement of mentions is an important first step. Yet, buried in words is what a person is like, how they're feeling, and what they intend the reader to feel.

Copywriters know how to write for a Grade 5 reading level, a Grade 10 writing level, and a university reading level based on a relatively simple algorithm. It follows that since words are machine readable, they can be treated very similarly to numerical input.

Take, for instance, the sequence of words:

"Butterfly violet breeze fizzy"

and:

"Papilio #800080 easy sparkling"

They each individually mean the same thing. Of course, they don't emotionally mean the same thing. The words have different shapes. The speaker of the former would be a normal person, maybe trying to write some poetry. The latter would be some sort of biologist programmer.

Words could be broadly categorized into different buckets, with great analytical effect. But it goes beyond just words. Verbs are where it starts to really get tricky.

If you want to really torture yourself, try reading "Investigations in Universal Grammar" and "The Stuff of Thought" in the same week. Take this quote from page 66 of "The Stuff of Thought":

"Some intransitive verbs resist the intrusions of a causal agent:

The bay is crying.
The thunder is crying the baby.

The frogs perished.
Olga perished the frogs.

My son came home early.
I came my son home early.

And some transitive verbs resist the attempt to strip their causal agents away:

We've created a monster!
A monster has created!

She thumped the log.
The log thumped.

He wrecked the car.
The car wrecked."

It's fairly hard to teach a machine how to interpret things that humans can hardly interpret themselves. Or grammatical rules that only seem to make real sense to the mother tongues' ears.

It's worth figuring out and applying.

Take landing page copy:

The purpose of a landing page, to a direct marketer at least, is to get the person to convert: to take a desired action. Good copywriters know how to use words and tone to compell people to continue reading down the page, like a slide. The theory is that if their head starts nodding at the top, they'll slide down the page, they'll continue saying 'yes' right into a sale. The copy, ideally, should ressonate with who the customer is intended.

I'm fairly certain that certain classes of words are better and convert more than other classes of words. Beyond that though, certain classes of verbs and tones are better at converting than others, in different contexts. The answers could mean the difference between 5% conversion and 20% conversion.

This is one of the thrusts with sentiment analysis. Useful and relevant.

The Culture of Analytics

Jim Novo has really stirred up the hornet's nest now. His post on "Analyze, Not Justify", is a great read.

Down in the comments Jim links to another post "Fear of Analytics". It's another good read.

It all goes to the culture of analytics.

There are people who are fail tolerant and people who are fail avoiding. I don't see how people can survive without a healthy balance of failure and success. Repeated success is required for confidence building and repeated failure is required for learning. Like everything, there's a downside too. Repeated success can lead to arrogance. Repeated failure doesn't guarantee that somebody will learn, either. The fail avoiding behavior, if it persists too long, results in stagnation and, ultimately, long term fail.

An organizational culture that tolerates repeated failure without learning is destined to collapse. An organizational culture that doesn't tolerate failure, even it means learning, is destined to stagnate.

Experimenting with various First (tweaks) and Second Order (major) changes to products and experiences over time, if there is an accumulation of knowledge, ultimately leads to short term and medium run commercial success. The losses from small failures are offet by the gains from an accumulation of knowledge. Some believe that multiple Second Order changes leads to a paradigm shift and to commercial renewal over the long run. (We know that certain countries from a public policy perspective can do this. This idea of 'innovation drift' is kind of interesting.) Admittedly, there are several decisions that somebody at the C-suite only get to make once.

A culture of analytics is a culture that is failure-tolerant and actively manages the risk of innovation.

Website Morphing

Recently I wrote a review on Website Morphing for the Web Analytics Associations' Research Committee.

You can go ahead and take a look at it if you want. I'll wait.

Website Morphing represents a coherent method for automating incremental optimization. It's not perfect. Morphing will require a heavy amount of human creative and analytical inputs. It's the social technology that's one of the big problems with Morphing, not the physical technology under the engine.

People with diverse skill sets often have a hard time working together. It's hard to communicate complex concepts with people who don't share your vocabulary. Sometimes it's like being an English Speaker in Germany, an increase in volume doesn't equal an increase in comprehension.

Those skills have always been hard, and they've been vitally important since the dawn of the information age. They're going to become that much more important as we move from designing static-one-size-fits all experiences, to designing custom, delta-heavy experiences.

In the end, if we're not growing we're declining.

Practical Social Analytics at NetChange Tomorrow

I'm presenting "Practical Social Analytics" at NetChange (Twitter search: #netchange ) tomorrow.

The challenge of the session will be for charities to figure out how to practically measure the effectiveness of their social objectives, using social media.

It's going to be a great. I'm looking forward to meeting people who are new to me (just because I haven't met them yet doesn't make them 'new'), and hopefully - preferably, building some bridges.

There's some trolling going on. I have been spending a disproportionate amount of time figuring out social media measurement over the past quarter - and an even more amount of time over the past three years on goal alignment strategies: so I come with a point a view and a strong desire to get beyond some of the hype.

Looking forward to seeing many of you there. If you can't make it, I fully anticipate Joe Dee using his camera to capture some of it (and subsequently posting some of it online).

As a preview to the talk, here's a picture I'll be using:

Neuroscience, The Power of Weak Ties, and The INFORMS Marketing Science Conference

First and foremost:

Sentiment Analysis, Anyone? is the continuation of an ongoing push to bypass all the pain and suffering ahead of us on the social analytics front and move straight onto the good stuff. I want to avoid a lost decade scenario, and just bypass the trough in the Gartner Hype Cycle. It's a lot to ask for, I know, but please - could we just make the decision this time to jump to the good stuff?

It's worth a read and it lays out a very specific challenge.

Next is this theme of the "Power of Weak Ties". There's an early paper (1954 I think) on Word of Mouth marketing which proves a strong tie between two individuals in a homogenous group is less likely to produce successful referral behavior than two individuals who belong to two seperate heterogenous groups who are linked by a weak tie. A presentation of an unfinished paper yesterday by Christian Barrot confirmed that finding.

So think about it: In the Web Analytics community, if somebody were to reccomend a web analytics product to you, and they're just like you - a member of the web analytics community, you are less likely to follow up on that referral with a purchase than if you were member of the data mining community. Why might that be? Well, if somebody comes at me and says "product X is absolutely awesome", and I look at them and say, "wtf are you talking about, product X < Y because of A, B, and C" - well, two experts are unlikely to agree.

If, on the other hand, you're a data miner, and you know nothing of Product X, and you don't even know about Y, Z, or W: you effectively don't know what you don't know - and you happen to know a web analyst, and that web analyst tells you Product X - you're that much more likely to buy Product X.

I'm certain that there are counter-examples. I know the importance of brand dominance for early adopters in a self-referential social network: but why the two theories are colliding - I don't know.

But carrying that 'weak tie' reference going and applying it to the INFORMS Marketing Science Conference here in Ann Arbor:

Much of the work, perhaps 95% of it, is really quite good. It's applicable to what Web Analysts and this emerging discipline of Social Analysts are tripping over. They don't speak our language.

They have their own world here, replete with hierarchies of authority, social networks, sheeple, and hidden colleges. They have their own jargon and their own biases. It's their own culture.

Every community has one. Whether they want to admit or not.

So are there major differences between the Web Analytics Community and the Marketing Science Community?

• The goal of the Marketing Science academic is to publish in one of the big journals.

• The goal of a Web Anaytics Practitioner is generally to prove value.
  

• A Marketing Science academic is focused on answering one specific question effectively.

• The Web Analytics Practitioner is focused on answering every damn question efficiently.

• A Marketing Science academic focuses on the method and the model.

• The Web Analytics Practitioner is focused on the toolset. (STILL!)

• A Marketing Science academic is likely to troll another based on the elegance of a model.

• A Web Analytics Practitioner is likely to troll another based on validity of a conclusion or the accuracy of a tool (STILL!).

• A Marketing Science academic will gloat about the size of his dataset.

• A Web Analytics Practitioner will gloat about what new media s/he is trying to measure now (the EXPANSION of the dataset).

• A Marketing Science academic will casually mention they have a Monte Carlo sim running for them at home, and that they'll check it when they get back.

• A Web Analytics Practitioner will casually mention they are running a VOC on one of their sites at work, and that they'll check it when they get back. And they also have a copy of The SIMS 3 running back home. And they'll check that too.
  

One of the biggest, and ongoing challenges for me, as a weak tie here, will be translate a large volume of this material into a format that #wa and #waw people can use, debate, and I really hope, start to measure. These people in Marketing Science are languishing with very old datasets. Web Analysts languish with very young, and frequently, very large datasets.

I'm kind of dealing with a world of extremes here.

Same base subject material. Two different cultures.

So, we'll give it the old college try and see what falls out.