After 416,350 words in posts and 845,128 words in comments on this blog, thus far, there is always more to explore, illuminate and share. Hence every once in a while I flip the tables and ask you for challenges you are facing.
It is a great way to stay connected to what's most important to you (and keep the blog and its content relevant!). This past Monday I asked for your questions and you were kind enough to share some awesome questions. Thank you.
I'm going to try and answer all of them here. But since the questions alone made up more than 1,000 words, I am going to try and keep the answers as pithy as I can while trying to give you an answer to chew on.
To keep things a bit organized the questions are organized into four buckets. They are: the tactical "How do I?", the strategic "How can we?", the abstract "How come it's not that way?" and finally, the surprising "How can I possibly answer that?"
It is going to be fun. Let's go!
Nose to the grind "How do I" questions:
David Walizer: How do you sell the value of web analytics to a skeptical client in 30 seconds or less?
By saying this:
"Web Analytics will help add $2 million worth of additional Economic Value by recommending specific ways to delight our customers and improve the way we sell"
You can say that in 10 seconds.
The challenge is that you should have done enough work upfront to know what's important to the business, got a rough sense for things you can fix right away and their value, and then done some back of the napkin calculations about the Economic Value your fixes will add. That takes a few days of pain. But there is no alternative.
If you are in a company this is easier to get done as you have access to people and at least some data (even if the site is not tagged). If you are a consultant then identifying opportunities is a smidgen harder, but you can use your experience with other clients to quantify value.
We all look for shortcuts, in this case some magical words from an enchanted fairy who lives in the mythical city of Oz.
Sadly David, nothing like that exists. I know you know that. :)
For help with identifying opportunities and how to do business analysis please see this post: The Beginner's Guide to Advanced Web Data Analysis
Nilaye Thakrar: What is the best way to attribute an offline sale to an online assist?
By doing multichannel analytics!
Your problem is the primary key. So use unique phone numbers (specific to campaigns if you want granular details)… leverage unique coupon / campaign / offer codes… get good at geographic targeting… become a God of controlled experiments.
More detail on those techniques, and three more, are in this blog post: Multichannel Analytics: Tracking Offline Conversions. 7 Best Practices, Bonus Tips
With 89% of all shoppers using the web to research information, it is ever more critical to quantify the impact of your online existence to your offline outcomes.
Did I say achieving greatness takes effort & desire? I think I did. :)
Why do I see keywords from Google with 0 visits in Google Analytics? A bug?
It is hard for you to share a lot of detail in a tweet so let me just make up one example and answer your question. I open GA's keyword report. I see this:
KW: "Avinash is a awesome" Visits: 2
KW: "Avinash is not awesome" Visits: 0
There are two reasons this could happen.
Paid Search. For AdWords accounts that are linked to Google Analytics in very rare cases it happens that "click" data is available from AdWords for certain keywords but there is no Visit data available in Google Analytics. Perhaps because the tracking code did not fire off or, more likely, there was no tracking code on the landing page. Then you'll see some zeros in the visits column. One way to verify is to click on Traffic Sources > Campaigns > Clicks tab to see the non-zero impressions from AdWords.
Organic Search. Let's use this example.. You come to the blog on the search phrase "Avinash is awesome," you land on a irrelevant page on this blog, you hit the back button and go back to Bing, you try a different search phrase "Avinash is not awesome," you land on the right page, you love it, you read lots of post, you leave. So how does GA decide which keyword to assign that visit to? Should each keyword show one Visit? That would not be right. Should it be the first search phrase you came on? The second one?
GA will show you both, but put a zero for the second. Essentially it is assigning "credit" (attributing) the visit to the first keyword ("Avinash is awesome"), and a zero in Visit for the second keyword ("Avinash is not awesome"). But each keyword gets "credit" for other metrics. So if you had seen three pages on the keyword "Avinash is awesome" then it will show one visit but it will show three page views. And if you came back again, in less than 29 mins on the keyword "Avinash is not awesome" and saw ten pages and converted then that kw will show zero visit, 10 pages and one conversion.
IMPORTANT: This type of behavior is rare so you should not see 0 visits often.
[ Update: Technical explanation via Cristina Chetroi: What happens is that when you access the site for the second keyword, the __utmz cookie gets updated with the new keyword, but, as the visit had already started when you accessed the website via the first keyword – the 1 visits gets attributed to that initial one - the first one. This way GA will show you both keywords while the aggregate for visits will not be skewed – as technically it is just one session.]
Landin Gee: Besides compete.com, where else can I access free competitive intelligence (excluding social media tools)?
For website visits, traffic sources, countries, keywords, and other sites visited, you can use Google Trends for Websites.
For organic search keyword analysis (share of shelf, unaided brand recall, competitive indexing, industry analysis) use Insights for Search.
For paid search keyword reporting (volumes, competitive performance, CPS's etc.) use the AdWords Keyword Tool.
For demographic and psychographic analysis (age, gender, income, education etc.) use the DoubleClick AdPlanner.
There are many, many paid tools in the market (like HitWise and Compete and Netsuus). Above are just the free ones whose datasets are is large enough to provide good data. They are all Google tools
Benjamin Rodde: I'd like to predict our Daily/Weekly/Monthly churn using the #GooglePredictionsAPI with unique visitors from #GoogleAnalytics.
The best option is to hire a statistician with experience in data modeling and forecasting. In case she/he does not have programming experience, hook them up with a technical person who has modest technical skills.
Two other quick things…
Churn is a term most closely associated with customers you have acquired (and then failed to retain) and not so much to "fly by night" Visitors on your site. The latter, except in rare cases, is hard to do predictive analytics on unless you are a stagnant business.
If you want to do churn analysis, as defined above, then you don't want GA as much as you want your back-end ERP / "Orders" database that has your customer history, and just a couple of acquisition signals (campaign and source come to mind). You can and should do that now (along with customer lifetime value type valuable computations).
Brian Krick: Best way to measure and communicate "available demand" from available channels (social, search, display) for forecast modeling
Please see the advice above.
Additionally, it is exceptionally difficult to measure available demand because 1. the web is so insanely fluid, 2. there are so many variables that drive demand (not just online but also offline and events etc.), and finally 3. because the data is so very not available.
(These are also reasons why click attribution to multiple campaign prior to conversion is such a thankless exercise. Implicit in that is the assumption of infinite demand, among other problems.)
I have personally had a lot of success using Controlled Experimentation techniques, such as, say, Media Mix Modeling, to understand both current available demand and also segment conversion effectiveness. And this has to be a continuous approach and not discreet. If you have Web Analytics 2.0 please refer to the controlled experimentation section, page 205, in the book for more.
Peter – IJsbeer: How to gain insights to improve competitive keywords rankings of websites? And to visualize it in a report..
Use tools like Insights for Search, Compete (Search Analytics reports), HitWise, SEOmoz Tools etc to understand your performance in context of your competitors.
Without knowing what you want to show it is hard to make a recommendation as to how to visualize. There is no upper limit to effective ways to visualize data.
But if it helps I love Tag Clouds, Keyword Trees, and Motion Charts. Also while it not so much a visualization, I love how effectively you can go from thousands of rows to just the few that matter using Weighted Sorts.
Find more inspiration at David McCandless's blog Information is Beautiful. [Thx Liz!]
Stephan Belanger: Analytics gives quantitative data; surveys: qualitative. What's the best way to aggregate to understand clearly how to improve?
If there was a pat answer to for this we would all be Analysis Ninjas!
My advice would be this:
1. Truly embrace Multiplicity and Web Analytics 2.0 (the idea that there is no one right tool or one right place to go all the time for every answer). See the first part of this post: Best Web Analytics 2.0 Tools: Quantitative, Qualitative, Life Saving!
2. Get really, really, really good at honing your data analysis skills, your brain strength in business analysis. You can start with specific ideas here: Beginner's Guide To Web Data Analysis: Ten Steps To Love & Success
3. Practice, practice, practice. See: Web Analytics Career Advice: Play In The Real World!
Over time you'll find that it will get easier and easier to answer your question above. I wish there was an easier path, if there is one I've yet to find it. : )
There are many I could nominate for this honor… so it is hard to just pick one… I want to go with the completely useless % Exit metric… but I think I'll go with Pageviews.
Total page views generated or even average page views per visit is just so useless almost all of the time.
I might let you slide if you show segmented depth of visit expressed in pages consumed, but only if you can show a clear line of sight between that and your company's net income.
Dan Rice: How can competitive intelligence be used to estimate competitors' conversions?
I am sorry but this is almost always a complete and utter waste of time.
See #1 in this post from 2006: Competitive Intelligence Analysis: Metrics, Tips & Best Practices
If you want to index against your competitors there are a ton of other valuable analyses you can do. They have been extensively covered here and in both my books.
Again my apologies, I wish I did not have such a depressing answer for you.
Dan Richardson: How to balance conversion as a factor / ratio of increase (or decrease) of traffic. i.e. Con +2%, T +10%. Con overall is lower?
Do regression analysis that factors in all the available variables (and be really thorough here) to isolate the relationship, as Wikipedia says, "between a dependent variable and one or more independent variable."
Chris Rinaldi: What's the most accurate way to discern a competitor's web traffic (at least 'visits')?
The highest boost in accuracy will come from educating yourself on precisely how data is collected by competitive intelligence tools. You can then understand how to choose the right tool – depending on the location of the competitors (U.S. or non-U.S.), the rough size of the traffic they might get (panel based systems suck at any analysis of sites that might get less than one million unique visitors a month), and line this up with your ability to pay (you go from free to really expensive, for example, as you move from Trends for Websites to Compete to HitWise).
Alex Cohen: How to optimize with sparse data!
Such a short question from my friend Alex, and so hard to answer.
You have to be able to recognize when there simply is not enough data, and in our WA context, this happens a lot. There are things like conditional logistic regressions that you can apply to tease out some insights. It is hard.
I have personally learned to switch to "What else can I do?" So for example, if the site has a very small sample, I could switch to a simple one question on-exit survey to get some VOC to get more hints as to what's up. Or maybe just dump the whole data from visitors and switch to expert analysis (conceptfeedback.com type analysis). Or maybe simple usability research. Or, worst of all when the gun is to my head, "best practices."
None of these are perfect but I like multiplicity because it lets my experience triangulate various sources and I can fill in some of the gaps in the sparse data from which I can't find insights with enough confidence.
Strategic "how can we" questions:
Simon: How to "sell" GWO testing (+Analytics) to Managers, very limited time/budget/staffing (btw, we already use Analytics and Adwords, but GWO is proving harder to get prioritised and understood, maybe worry of complexity).
If there is not budget or time or staffing then you might be hitting your head against a brick wall and sometimes recognizing that this is a waste of time and waking away is the most optimal thing to do.
But if there is an opening, then there is one sure way to convince almost anyone (to use analytics or testing or WebTrends or surveys or whatever): Compute the economic value of following your recommendation.
For that you have to know the goals, some goal values, and have some sense for how much improvement you can drive.
Today: Visits 30k. Conversion Rate 2.13%. Economic value: $2 mil.
Post Testing: Visits 30k. Conversion Rate 2.4%. Economic value: $2.2 mil.
That always works. Of course it is not easy to compute, but that is why you and I and the Ninjas are paid big bucks (to do the hard work!).
Jon Whitehead: Why does the public service have such a massive aversion to measurement? Silos, budgets, internal comms all influence.
There is no tradition of accountability in almost every country when it comes to public service (there are some exceptions like Singapore).
No accountability = very little desire to measure.
There has to be fundamental massive change to a bureaucratic, siloed, politicized institution populated by non-relevant people at the top environment. Else you and I, Jon, will slog in vain, or at best, move in minor increments along a local maxima.
Chaudhry Javed Iqbal: In a non profit organisation, how to convince business managers that their is more to analytics than just unique visitors?
#1: See above. : )
#2: Create a crystal clear Web Analytics Measurement Model well before you ever talk about the data or tool.
With a WAMM in place I promise you that they will care and want good metrics because they would have contributed to the top part of the WAMM which gets them to put skin in the game.
Webbingyourway: If you had one wish that would force all analysts to do one thing perfectly all the time when analyzing, what would it be and why?
Get really good at measuring Primary Purpose & Task Completion Rate, and then segment that data and analyze.
It is exceedingly rare that we understand all the reasons people come to our site. It is common for our leaders to have a non customer-centric view of the world. Both of these flaws result in our analysis being significantly more narrowly focused than it should be.
Use 4Q, use KissInsights, use something else you like. Get into measuring this regularly. Bring customer-centricity to your analysis, take your understanding of that data and now identify all the Macro and Micro Conversions in your web analytics, measure holistic success and celebrate the massive acceleration of your salary!
Reason enough to do this? : )
Suzy Sandberg: Easily understanding the incrementality of all marketing channels without having to compare order IDs.
I LOVE doing this.
Simple answer: Media Mix Modeling or Marginal Attribution analysis.
Both require patience and love, the former a bit more than that latter.
More details in Chapter 12, pages 366 – 368, in Web Analytics 2.0.
Zyxo: Do you know 1 company that mixed web data and customer data for marketing purposes ? Results ? Thx.
I personally don't know any company that has done this: "Let us build a massive data warehouse of multiple years of clickstream data with all our customer data and look, we have orgasmic insights." 100% of those efforts have lead to failures.
VERY, VERY IMPORTANT: This could just be a blind spot for me. There very well could be hundreds of companies that have done this and are thriving.
I have seen quite a few companies that have taken a selective subset (typically a handful of values related to acquisition dimensions) and put that into their corporate data store and used that to do remarketing, or computing customer lifetime value, or other marketing purposes. The centrality was not web data, it was the corporate multi-channel customer data. The web data happened to be one small part, but an important one.
Diogenes Passos: I got problems on communicate under performing micro conversions [content based websites]. Would love to hear practical advices.
Diogenes I am afraid I don't understand the question.
But if getting people to take action on your recommendations is a challenge then I encourage you to look at the two spreadsheet examples in this post, Barriers To An Effective Web Measurement Strategy, and use that type of a framework, especially the last column with red font which quantifies the impact of non-action.
It always works as a swift, but lovingly-delivered, kick to the rear end.
Sarah E. Bowser: Does web analytics help the internet field overall or hurt the user and privacy rights? Where's the line?
Analysis of data of their website by owners of that site unquestionably helps them and the internet field and the users.
With regards to user privacy. . . I was recently asked about it and our (analyst) behavior and analytics tools. Here's my answer:
As a web analytics vendor you should provide choice to the businesses that use your tools. Don't be one of those lame vendors that use third party cookies to "opt out of tracking." Have the courage to develop better opt out options like plugins. If a business's website visitors have choice they are more likely to trust the website they are on, which is great for everyone.
Consumers should worry about privacy. They should always have a choice. We (consumers, site owners, vendors & analysts) should ensure that choice exists, and we should work hard to earn the trust of website users.
Hope this helps.
Analytics Ninja: The best way to create a multiple touch attribution model for conversions.
I wish I could give you a pithy answer.
Perhaps you already have a copy of my second book, just jump to Chapter 12.
Christopher Hart: Why linear trending alone, without data correlation to triangulate causation/answer are deadly mistakes waiting to happen?
If you plot the number of doctors in the city and the number of murders in the city you'll notice that the correlation might seem pretty tight. So you, or I, could declare "Reduce murders, eliminate doctors!"
We do that in web analytics every day. Our only salvation is to consistently seek to establish causation.
Tara Dunn: My one question would be, where do you draw the line between using data for answers vs. using your own logic? It seems like I often run into scenarios where the analysis will only take me so far but it still doesn't tell me WHY. So then I have to switch roles, from analyst to scientist, and build hypothesis and test them. I would be interested to hear when you switch from analyst to scientist, and if you think there are other "roles" necessary.
I answered this in a recent comment where Erica asked me: Do you get insight from data or find data from your insight?
Here's my answer:
The answer of course is: Both. This is why *analysis* is so important, and reporting is just something we have to do to survive in this world.
When you are doing analysis sometimes you are in an exploratory mindset, you know the goals of the company and you are exploring trends and patterns in data to find insights. At other times you'll have a hunch or a gut feeling about something or (my favorite) a hypothesis about something, and then you'll don your analyst hat again and you'll analyze data to validate your hunch / gut feel / hypothesis.
So insight from data or data from insight depending on your starting point.
To summarize Tara, the answer to your question is: No line. :)
Abstract "how come it's not that way" questions:
I requested Dorota to define what she means by predictive analytics. She kindly defined it: As using past visitor data (web analytics) to generate actionable insights about future visitor behavior, whether through modeling, etc.
When I wrote that post (Predictive Analytics On Web Data Works? Nyet!) I defined it as such in my mind, but in hindsight, I did a poor job of making that explicit in that post. I should have and I regret that oversight.
To answer Dorota's question…
In context of that definition… I am afraid that I don't think we are there yet. We have not solved the problems outlined in the 2007 blog post. Wait. That's not right. Our technology is still the same, the data we collect is the same (with the same fragility), the holes that existed still exist, the environment we have to do PA is not any better (and it does not matter if you have a massively complex data warehouse).
It is important to realize that there is one "predictive analytics" you can and should do if you are a largish ecommerce (outcomes) type website. The behavior of those who actually buy. You have loads of info about them (including PII), you can tie it to their other purchases / contacts (offline), you can try to "predict" repeat purchase rates, attach rates, likelihood of this or that centered on outcomes. This will cover 2% of the traffic, but of course it is all your revenue.
What people will share publicly is changing fast. A certain social media company whose name starts with F and ends with ook :) is pushing boundaries of what data is captured and then shared with partners and how it is used. This is very different from the data we capture today in the web analytics world. In the near future this will, I think, change what analytics vendors consider data. We are entering a world in which we can tie a visit to a person with supreme confidence, a person who is not just a cookie but we know likes to listen to the Jonas Brothers, wears pink underwear, has three iPads and has just bitched about Delta airlines. Just think of what you can do with that.
The challenge is not that we don't have algorithms, the challenge is not that we are not bright… the challenge is the number of and cleanliness of known and unknown variables we can input into the data-set we have. This will change with time.
I hope never. It would be horrible to be "absorbed" by BI as it is today.
My experience in the world of Business Intelligence still surpasses my experience in the world of Web Analytics. I have not only lived there, but done the down and dirty, and have the bruises and some trophies from my time in BI.
BI is massively IT-centric, slow moving, controlled by a centralized team of report writers who fulfill requests based on a painfully prioritized processes on a monthly schedule from data sources that are have the agility of a turtle carrying a thousand ton weight, powering big decisions infrequently.
WA is, for the most part, owned by business teams with data stores in clouds with little corporate IT involvement, the coolness of capturing more fascinating data faster and, and, and ability to analyze the data with the agility of a turtle with no weight, powering small and medium decisions every week.
WA moving into the above-described IT-centric environment would be the kiss of death.
BI does have the benefit of corporate buy in (how else would you pay sixteen million dollars to Oracle each year for your "backend" and five million a year to SAS for the "frontend?"). They have the golden data (Ohh PII I love you so!). They have an established history of proven algorithms and models and mathematical techniques and all those approaches Jim Novo and Kevin Hillstorm keep talking about!
WA's got none of that. Okay okay so we are just a 4-year-old child and BI a 65-year-old.
The old "offline world is dying," its way of doing business is dying. The web is current, digital and its demands of how business should be done are the future. I hope that a new field emerges, let's say called Cutroni Analtyics, that absorbs the discipline and the analytical methods and rigor (and money) of BI and adopts the agility, cloud based non-kill-me-now-rather-than-wait-for-9-years-for-IT-to-implement-something-death-grip way of doing business-centric analysis.
My friend Matthew Tod called it Customer Analytics. That's a nice name.
In the end, I personally don't care what its called. I pray to Jesus and Allah and Krishna that it is a hybrid that contains the best of both worlds. That's what we need.
Joe Teixeira: What will it take to finally (FINALLY) "make it" to the big time as an industry?
It will ultimately take becoming central to every company's existence. Perhaps it takes the hybrid I have described in reply to Justin's question above.
But I don't see that happening anytime soon, the worlds of traditional business intelligence and web analytics are populated by people and mental models as different as. well to borrow a popular metaphor. Mars and Venus.
I do not believe we (taking a web centric view for just a moment) are that far off from big time. If every single Analyst throws off the yoke of being a Reporting Squirrel and focuses her/his work on tying every single analysis they do to Net Income then I think we will become BFFs of Sr. Leaders very quickly.
At the very minimum that will give us a really solid shot at being relevant.
Surprising "how can I possibly answer that" questions:
Ben Hofstetter: What If Google Analytics had custom domain URL shortening and a URL builder as1 http://goo.gl/ & http://goo.gl/crrY9#utm_that
Google Analytics (and WebTrends and CoreMetrics and everyone) already support easy tracking of shortened URLs. All you need to do is make sure that you are adding well thought out campaign parameters.
See how that works in the Social Media Segmentation part of this blog post: 3 Advanced Web Analytics Visitor Segments: Non-Flirts, Social, Long Tail
Naqaash Pirani: What best practices in web analytics can be applied to social media? Are there any similarities between the two mediums?
They are same and they are different. They are same in that an obsession, an absolute obsession, with outcomes is mandatory. They are different in that the desirable outcomes have to be rethought (from a website analytics context).
Start with specific ideas here: Social Media Analytics: Twitter: Quantitative & Qualitative Metrics
Geruza: What is the best tool for a web analyst?
Any. No that's not right. All. No that's not right. Omniture! No that's not right either.
The tool that is right for you is the best tool for you.
Figure out which tool is right for you by asking these three important questions: Web Analytics Tool Selection: Three Questions to ask Yourself
Then ask the vendor you're going to choose these questions to ensure you are choosing the right one: Web Analytics Tool Selection: 10 Questions to ask Vendors
Dominic Parker: What do you think is the best web analytics package out of Google Analytics or Site Catalyst?
It is foolish for anyone to state that one package is better than another one for everyone. If someone from Google Analytics is saying that, they are exhibiting a lack of critical thinking brain neurons. Ditto if someone from Adobe or IBM says that.
We are blessed with a diverse set of tools in the marketplace. Spend some time trying to figure out which one is right for your current needs not what you might need 96 years from now. (See the 3 Questions link immediately above.)
To base your decision on what I think, or an "industry analyst" or your lover is really. what's the word I am looking for. hmmm. let's just say not good.
Timur Khamitov: What is the MAIN value driver of over-all conversions: rankings, rich content, UI, IA etc? Guess what I mean is, in your experience: what is the 1 thing that delivers results above all else.
The one thing that consistently delivers results is delivering customer delight.
That could mean doing all the things you mention in your question well, or delivering some of them really well and not totally sucking at the rest.
At the very minimum, ensure there is a high degree of overlap between Customer Intent & Website purpose. See the first part of this post: Six Tips For Improving High Bounce / Low Conversion Web Pages
Sanchit Somani: How to define the perfect dashboard?
Hmmm… here's something tweet sized…
Perfect dashboard: No data pukes, just insights with actions to be taken based on a critical few (3) business KPIs.
Here's one idea to get you started: The "Action Dashboard" (An Alternative To Crappy Dashboards)
33 questions is a lot is it not? : )
If you are interested in equally complicated questions (and hopefully equally delightful answers) then please see these two posts where I've also answered user questions:
If you are interested in a video version of a Q&A series then check out Web Analytics TV, it focuses only on Google Analytics:
So much stuff for you to read, learn and become awesome. The ball truly is in your hands.
Before we go… I want to thank you for being engaged and for asking such wonderful questions. You force me to think harder, you keep me connected to reality, you help me become a better Practitioner of the art of business analysis. Gracias. Arigato. Merci.
Okak its your turn now.
Which question did you find most astonishing? Which answer did you find surprising? Which answer do you think is completely off the mark? How would you have answered Alex's sparse data question? How about Justin's BI question? Do you agree with my perspective and tough love to Analysts on privacy?
Please add your wisdom, critique, feedback via comments.