Posted by
Bo Borland on Tue, Apr 10, 2012 @ 03:10 PM
Today's web application audiences are in the thousands or even millions of users. Users are different in terms of behavior, usage, needs, and attitudes and marketing campaigns or promotions can be tailored to their different needs. Behavioral segmentation of web application users is driven by user behavioral and usage characteristics. Behavioral segments can be created by pre-defined business rules or created by data mining algorithms. Business rules can efficiently handle only a few dimensions and is managed by a business expert, but data mining on the other hand can analyze behavioral data, identify the natural audience segments, and suggest segments based on data patterns.
Data Mining Clustering algorithms attempt to identify distinct user groups with similar profiles so that marketing can be tailored to their needs. With clustering user groups are not known in advance and are derived based on an input set of data fields. The major advantage of the clustering techniques is that they can handle a large number of data fields and detect patterns and suggest audience segments that humans and queries cannot identify.
Pre-defined user analytics business rules can also be applied to data for behavioral segmentation. This data could include item purchases, web sessions, and product utilization or event data which are usually housed in web application databases. A first step in analyzing and segmenting this data is to pull it all together in a single user table or view. A user analytics solution such as Monetrix can combine all of this information to together into a 360° User View which uses common business rules to segment the audience by their behavioral and usage patterns. Many pre-defined audience segments can be derived by this combined user view. This view also serves as a nice data-set to apply the previously discussed clustering algorithms to because all of the data is in a nice tabular format aggregrated to the customer level. Here are common clustering algorithms that can be applied to the aggergated customer view to determine user semgents.
- K-means - a fast and efficient clustering algorithm that can handle large data sets with many columns
- Kohonen Network - provides a map of clusters on 2 dimensional grid
Posted by
Bo Borland on Tue, Apr 03, 2012 @ 09:18 AM
Data-driven marketing professionals seek personalized customer interactions through the identification of their customers' unique needs, preferences and behaviors. Customer segmentation is an important process for dividing customers into seperate distinct groups or "segments" based on similar characteristics. Once the segments are defined, differentiated marketing strategies can be tailored to segments. A company that can tailor their marketing message to the needs and wants of the individual customer and make appropriate suggestions is more likely to foster a long-term customer relationship.
Customer insight is critical to developing this ongoing relationship. Traditional CRM Software provides transactional support to capture and organize customer interactions; however, these operational systems do not deliver the data-driven, customer insight needed to understand the customer's needs and wants. Customer Analytics is about analyzing online customer information to determine customer preferences and value based on their unique profile and behaviors.
User Analytics is subset of customer analytics that deals with users of web applications. For ecommerce sites or other web applications, user analytics can support customer development by creating user or "audience" segments and matching item preferences, better targeting of item promotions to those audience segments. This segment insight gained from analytics should be loaded back into operation CRM systems and other marketing and promotional channels.
In this four-part blog post series, I will discuss these three different segmentation methods
- Behavioral Segmentation
- Value-Based Segmentation
- RFM Analysis Segmentation
for improving web application user development. These user segmentation methods can help companies move away from mass marketing to a more individualized handling of customers.
In the last decade and a half, there has been an undisputable rampant growth in online purchasing thanks to convenience, better pricing, drop-ship fulfillment models, just to name a few. Traditional retail outlets (Wal-Mart's, Macys, Target, etc.) have made their products available online in addition to keeping physical stores open. Anything and everything you can imagine can be at your door, next day with a few easy clicks of a mouse.
Analysis around profit, revenue, margin, and inventory is so streamlined that retailers are able to perform additional and more focused analysis with newly available customer data sources: click stream, search and purchase histories, Ad Sense, and customer information, all of which are captured as soon a customer browses to their favorite search engine to find what they're looking for to their order confirmation screen.
Fast forward to today. The new "last mile". Customers are now researching and making their purchases anywhere and everywhere an internet connection is available, all thanks to the increasing availability of internet capable mobile devices. No sweat. We're still capturing the same data as we were before when they were sitting at their desktops in their home or office and we can generate the same analysis as we were before to better customer experience, identify target markets, optimize our supply chain, or whatever it was we were doing with the data. But now we have additional information to add to our analytics; namely, time and location. Let me rephrase real-time time and location.
What can we do with this? The exact same thing we've been doing tailoring the online experience (ads, specials, email campaigns, and so forth) except now in 3D. That's me, Monday through Friday, walking down the street passing two dozen retailers before I buy my extra strength cup of coffee, for those of you who didn't tie "3D" back to real life.
The analytics you've built for every one of your customers (purchasing history, frequency, habits) is now being mashed up with the elementary real-time data feed below, for example:
-
time of day
-
location of customer
-
closest retailer location within 100 meters #1
-
closest retailer location within 100 meters #2
-
...
-
closest retailer location within 100 meters #N
You're now able to display an ad or notification to your customer's mobile device suggesting they purchase the shoes they've looked at from 20 different IP addresses since 2012-03-01 08:34:29:030 at your retail outlet they are 50 meters away from. By the way, you know what inventory is on hand at that location so that you don't mistakenly display the advertisement and since you're such a nice guy, you'll display a coupon (assuming all of the P/L math works out and you aren't losing money on the sale) that can be scanned at the register if they decide to walk in and make the purchase.
Maybe they're about to make a purchase on their mobile device , you're out of stock online, and the system is able to suggest a retail location nearest to price match the purchase with. Maybe 10 of your customers are on the block and the sign in the window changes to display a Groupon-ish type deal if 10 of these items are purchased in the next few hours? The possibilities to enhance your [or maybe you're partner's] customer's experience and in-store user monetization process are endless.
Posted by
Bo Borland on Tue, Mar 06, 2012 @ 09:50 AM
Imagine you are a marketing manager wanting to identify, monitor and market to a niche segment of users of your web application that currently has 10 million active users. Your company host an annual customer appreciation event and you want to run banner advertisements within your application to market the event to your user base. After running the ads you need to immediately test the effectiveness of the campaign on a niche segment vs how the campaign performs across the entire user population so that you can make adjustments on the fly to the banner advertisement in hopes of improving repeat-purchaser responsivenes. Your company has access to a poweful marketing analytics tool for profiling top users capabable of making all of this happen seamlessly.
The marketing manager defines the niche segment as users that meet the following metrics criteria:
- Monetary Analysis for users in the top 20% of all users based on lifetime revenue and
- Users with an ARPU greater than $10 and
- Who log-in from the State of California and
- Who bought an "event ticket" last year
Step 1: User Segmentation - Analyze and segment your users with a powerful and scalable big-data analytics tool. You need an analytics platform that is easy enough for non-technical marketing personnel to perform segmentation analysis on current data to determine the users that fall into the segment defined above. It also needs to be scalable and powerful to process the results at the speed of thought on top of a 360 user view.
After 1 minute of analysis, you determine that there are 10,000 users that fall into this segment. Now that you have defined your segment, you need a way to {tag} that segment for identification in future analysis or for tagged user identification in other 3rd party marketing tools.
Step 2: Segment Tagging - Create a {tag} name for this segment of 10,000 users. The tag name for this example is 'Top 20% CA Event Ticket Buyers’. You need to be able to easily create this tag within the same interface used to identify the user segment. This tag name needs to be exposed and available to ad serving platforms, email marketing tools and reporting.
Step 3: Ad Serving - Now that you have tagged this population, you enable your ad server to read this new key/value pair consisting of the tag name defined above. The ad server needs to be able to make an API call to your analytics server that houses the tag names that the marketing manager just created on the fly in case you want to serve the ads only to that user segment. In this example, however, you decide to run the "event information" advertisement to your entire user population.
Step 4: Real: Time Monitoring - This marketing manager is interested in seeing the immediate impact of the ads being served inside the application. What types of events (e.g. purchases, social events) are happening within the application RIGHT NOW after serving up the ad to both the {tagged} users and the general population. To answer this question by user segment, you need real-time monitoring reports that show how this tagged segment is currently interacting with the your application compared to the general population.
Step 5: Take Action Based on Results from Analytics: The marketing manager notices right away that the tagged population is not responding “more” positively than the general population, and she makes a management decision to create a modified ad to include a 10% discount for users who bought the 'event ticket' last year. This ad will only be served to the users a tag of ‘Top 20% CA Event Ticket Buyers’. After making the change, she continues to monitor the current day’s event activity, and she notices an immediate uptick in positive response from the tagged user segment. All of this happens within the span of one day.
The five-step closed-loop analytics example is easily achievable with the right analytics platform that can combine real-time application data-capture with powerful end-user analytics output. This type of real-time, closed-loop marketing analytics is exactly what Monetrix User Analytis was designed to do. Check it out at www.monetrix.com and register for a private beta to get free access to the powerful closed-loop analytics capabilities described above.
The Internet is a busy place, so why does it feel like you could be doing better when it comes to app and website revenue? Tracking and monitoring the monthly movement of your users is vital to gaining an understanding of what you’re doing wrong (and what you’re doing right) with your activity online – our analytics products help marketing executives do exactly that.
Useful InsightWe provide you with detailed user metrics that provide insight into your application users, which in turn help you to identify your current audience, and your target market, along with helping you to understand your application growth and monetization effectiveness.
These insights can provide you with a more compelling marketing message, which in-turn brings more qualified users who are more engaged with the app and have a greater potential to become paying customers. This results increased user monetization, and more revenue for your company.
How Does it Work?
Well, there are several movements which you can track, including acquisition, upgrades from free to paid, downgrades from paid to free, churn, and re-activations.
The most intricate part of tracking and monitoring users’ movements is defining the rule of these movements. Below you’ll find some common definitions that generally apply to users of free and paying web applications, which look at free and paying users separately to gain an accurate view of movement.
Example of User Analytics
- Registered Users (opening balance) – this is a count of the total registered users from the last month.
- New Users – A count of this month’s new registered users.
- Upgrades to paying - Count of the users moving from paying last month to free this month
- ARPU - Average revenue per user. Determined each month by calculating the total revenue that month by the total amount of unique users for the month.
- Stickyness - A mesure of engagement calculated by dividing DAU/MAU (daily & monthly active users)
- Reactivation - Users that connect in the current month after spending two of more months inactive.
- Churn – Inactive users or users that have not logged into the application in the previous month or week.
Analytics can also help you define statistics such as average user demographics (age, sex, location of your users), and who is visiting your site and when – all of which can be used to your advantage. Think of it as insider knowledge. For more information on our product, please
contact us today.
When organizations decide to embark on a business intelligence initiative, they often struggle with the question of what the proper metrics are. This frequently is solved in a bottom up fashion. Where each individual department defines their own metrics. This blog posts outlines how a top down approach works, and why it’s more effective.
A top down approach for metric definitions starts with the CEO. The CEO has a company vision, long term goals, and short term goals. The CEO has certain metrics that she/he tracks that measures these goals at high level. For instance, one of these metrics may be “revenue per employee”. The 4 or 5 most critical metrics (KPI’s) that the CEO is worried about is what should go in an executive dashboard.
From there, each department head needs to decide what are their 4 or 5 key performance indicators that support the metrics defined by the CEO. These will be more specific to the departments business, but also will insure that each department has goals congruent with the CEO. From a BI implementation standpoint, these are “drill-downs” from the CEO’s executive dashboard.
This same procedure can flow through the rest of the organization’s hierarchy. In the end, the organization gets a business intelligence implementation that insures all departments are focused on the same goals consistent with the vision from the top.
Developing a successful app means knowing who your users are, why they are there and what they expect to receive from downloading, engaging, and using your application. Understanding your app users’ through
customer segmentation and
monitoring is essential and provides visibility into the aforementioned questions. I am going to provide a series of blog posts that highlight 5 of the most important things to know about your app users that can be accomplished through customer segmentation and monitoring.
My first post will address user expectations. Before any app, or product for that matter is downloaded or purchased the prospective customer has an expectation of how the product will benefit them. Understanding and meeting or exceeding the expectation is essential to the products success or failure. A failure to meet or exceed this expectation will result in buyers remorse. Although this is common when spending any amount of money for a product the psycho-purchase term applies to “free” products as well. By “free” I mean it doesn’t cost money but does require the contribution of your time, personal information, and access to social accounts that many would rather not relinquish if they didn’t have to.
What Are Their Expectations?Different types of apps have different expectations. For example some apps are utility apps that are used to accomplish a certain task such as a QR scanner. Therefore all the user expects is for the app to work correctly when they come across a QR code. If for some reason the app doesn’t work correctly it will probably be abandoned or deleted quickly. Where as a more social app will probably be given the benefit of the doubt if a certain function doesn’t work as promised. That is because the user is engaging with a completely different set of expectations.That example just highlights the difference in expectations. If an app doesn’t work it will never be successful no matter what type of application you have.
Other apps have intricate and detailed expectations. Such as a gaming app or a social community app. There are expectations with game play, graphics, social sharing, friending, and in app payment to name a few. The more your user engages the more you need to understand who they are. The expectations of the user are highly aligned with engagement time. If the app falls short so will the users engagement. The opposite is true with better than expected application experiences.
Knowing how your user plans on engaging and using your app is at the center of developing a successful application. Make sure to set a realistic and creative application experience no matter what category you application resides in. Its important to understand what your user expects to get out of your app and therefore make sure your in app experience aligns with this perceived benefit. If there is a gap in these two it is likely that your
churn rate will be high.
Posted by
Bo Borland on Tue, Feb 07, 2012 @ 02:21 PM
Understanding an application’s users allows companies to better manage the relationship through what is offered and how the company communicates. User analysis and segmentation can deliver insight to understand drivers to acquisition, usage across features, and responsiveness to campaigns. Monetrix was developed to make profiling easy with powerful drag-and-drop analytics and segmentation capabilities, and to provide a “write-back” mechanism to take action on the insight gained from user analysis and leverage that for user monetization. This write-back capability allows tags to be created on the fly which can be fed back into marketing channels for increased user monetization.
The
360° user view is a core data model component and foundation needed to enable on-the-fly user segmentation. The single place to go for all relevant user attributes. Some examples of data sources used to build the 360° are: user profile records, purchase records, events records, session records, campaign records. Each of the sources provides a rich set of dimension and numeric data for summarizing and storing on a single user record. This greatly simplifies the reports and queries used against this model.
The thought first came to me during a financial derivative class in college.
"Why do we continue to look at years and years and years of old data to try to figure out where the markets going next?"
A few years ago, maybe we didn't have fire hydrants of information available to us on every corner of the web. So before that, we only had historical data to conduct statistical analysis on in an attempt to forecast where to place our calls and where the market is going next. While this is still very valuable, we have accessibility to near real time information (practically everywhere) to take our forecasting to the next level.
For arguments sake, let's say your current method of forecasting your company's stock price took into account 4 industry wide factors. And every time we provide a forecast, we always seem to miss the mark, either over or under. Question: how can we better this forecast? By using information extrapolated out of social media streams.
Bear with me, because a lot of things need to fall into place here before we can simply boil down to another factor, and add it to our existing forecasts.
There's a lot of buzz on the Twitter, Facebook, Qzones, etc. feeds on the product your company is releasing in the next quarter. No one but a handful of beta testers and tech sites have gotten their hands on it yet and the only information on the product available to the general public is through beta testers, press releases, trade shows, etc. Given the information, and having never used your product, what are potential customers saying about your product on social sites and furthermore, what is the general "feel" for this data?
Your next data warehouse or analytics platform should be drinking from this fire hose. I once saw on a Yelp-like site once, a gradient that tried to rate the restaurant somewhere from a red to a green based on, what appeared to be, semantically analyzed reviews from customers.
"I liked the chicken but the salad wasn't that great. Over all, I wouldn't come back again"
Good connotations: chicken. Bad connotations: salad, the contradiction "wouldn't" followed by "come back again.". Multiply this by the 100s of reviews that have been scraped from around the web, add a pinch of salt, and you have a gradient indicator.
Now, Imagine semantically making sense of the data extracted about your future product and boiling it down to a colored gradient, a numeric quotient, happy/unhappy faces (you get the point) that gives you a better feel of how everyone else feels about it!
With the prevalence of social media outlets available to organizations, this data has the potential to unlock finely tuned information that you may otherwise never had access to. Semantically understanding data in these feeds is probably piece of the puzzle that needs to fall into place based on how dissimilar the data in these streams tend to be.
The Data to Ink ratio is a concept coined by Edward Tufte, a pioneer of data visualization, which boils down to keeping information on a visualization clear and unhindered by other pieces of the visualization. More explicitly, the ratio is defined by the "amount of ink used to display data" divided by "total ink used on visualization." Anything that is not part of the numerator, should be removed from the visualization.
The one and only tenet of the ratio is that the data is always a first class citizen on a visualization. I think we can all agree that If you are quickly glancing at a dashboard, you want clarity for clarity's sake. As a side, I suspect this was also realized in a time where the tablets, displays, and mobile devices of today were not available and every table, bar graph, pie chart, etc. had to consume ink to be printed. Ink or no ink, the ratio still holds true on our modern devices.
Let's take, for example, this delicious pie chart.
Things we must have for this visualization to make sense:
- title
- legend
- color to discriminate between the two types of classifications
- percentage of each slice
Things that we could do without:
- color legend in callout of slice
- title in callout of slice
- classification in call out of slice
- bold, italicized, underlined title and background color
- background image of delicious pie
- random sizes of fonts
Now imagine a more realistic dashboard of today comprised of trend lines, scatter plots, bar graphs, time-lapse charting, mapping, etc. of which each piece of the dashboard has relating information in an adjacent graph or chart. Would you place a legend on every chart? Would the color of attributes in common across the pieces of the dashboard be the same or different colors? Would you need a title for every chart? Is it clear?
A good follow up when designing visualizations would be to keep each chart clean [by minimizing non-data ink] and then reviewing the dashboard as a whole to see where you could combine things such as legends, color schemes, or labels. Always remember to keep the data apparent by removing redundant, conflicting, and irrelevant pieces of non-data.