Case Study: Skinit Transforms Big Data into Actionable Information
Gains unprecedented visibility into customer behavior
By: Kate Bartkiewicz
Sep. 6, 2013 01:06 PM
Skinit Inc. is an eCommerce provider of personalized skins and cases for a wide range of consumer products. With over 180 licensed brands, in-house artwork and the ability to completely customize designs, the company reaches a diverse market. Skinit knew it faced an increasingly competitive market where customers were expecting a more personalized experience. Like many eCommerce companies, we had a large amount of data on what was purchased, what pages were visited, where visitors were coming from and so on. However, while we had more and more data every day, we still lacked the ability to connect all these data points and turn them in to knowledge.
The eCommerce Challenge:
The "wish list" consisted of the following:
1. We needed an analytics system that was light on implementation and high on integrity. The unique capabilities of Skinit's site made implementation management difficult and threatened data integrity. Like many companies, our development team was swamped with other projects so getting implementation on site in a timely manner was a big challenge.
We were looking at various tag managers to solve this issue. While tag managers would have surely relieved some of the implementation burden, we still had concerns about data integrity as we had experienced problems with tagging in the past, particularly with getting the data comparable for year-over-year benchmarking.
2. The data needed to be person-based rather than page-based. All our efforts were spent optimizing and improving the customer journey, and we needed data that supported this. Connecting web-behavior data to sales, CRM and other supplemental data sets was the next logical step.
While data analytics platforms allowed us to get some data in this format, it was limited and difficult to connect when visits did not lead to purchase - and non-purchasers were the target audience for optimization.
3. Data needed to be collected to enhance merchandising and personalization. With such a large catalog, understanding relationships between products was key to helping the customer personalize their device quickly. We wanted to be able to make recommendations to help the customer along. Since the product can have a wide array of attributes such as color, brand, gender affinity, sports market affinity, geography, etc. that aren't necessarily present in page-level meta-data, we needed to be able to pull data from multiple sources to better understand just what made different products appealing to different users.
We looked at third-party recommendation engines, but felt there were some limitations in the models as they were either built to analyze frequency of purchase combinations or frequencies of products viewed. This presented a challenge for us since most customers bought one product at a time, and products viewed together were proven to be self-fulfilling because customers either viewed what we merchandised to them or they left. The challenge was that we didn't feel we knew how to merchandise the right thing to them.
For example, what recommendation should we make to a customer accessing the site from the San Francisco Bay area but viewing Pittsburgh Steeler products? If we based the recommendation on category, the customer would be shown other NFL products for other teams - unlikely to be appealing since it is rare for a fan to have a strong affinity to two teams in a single sport. Do we market another sport such as baseball? And if so, do we assume they are a San Francisco Giants fan or a Pittsburgh Pirates fan? To really elevate the business, we needed a system that could effectively make decisions like these.
4. We needed to be able to quickly identify, track and examine each micro-conversion path for optimization. Ideally, we wanted to be able to quickly identify the most frequent paths to success and the most frequent paths resulting in drop off. For example, if we had a five-step cart process, we wanted to see how many users went from Step 1 to Step 2 to Step 3 to Step 4 to Step 5, and how many deviated from that path and how. Traditional cart funnels clued us in to deviations from this path, but not where or why.
We could conduct this analysis using our paid analytics platform, but we could not integrate it into our business intelligence system, which made the analysis time consuming and limited in scope.
5. Finally, we needed a better way to visualize problems or optimization opportunities throughout the company in order to get teams to act on findings. We looked at session replay technologies, but were concerned about the limitations in filtering or searching these sessions for specific characteristics important to us. This meant we would have to sample records, which was not the time saver we had hoped for.
We added up the cost of all the various services, partners and platforms we could use to meet most of these challenges. Ultimately, Cloudmeter, a leading software solution provider for transforming real-time network Big Data into actionable information for IT and business users, satisfied all of these requirements and cost 30 percent less than the alternate solutions.
The Cloudmeter Results
We created a data feed where the primary unit was the customer. Whether they converted or not, we were able to capture each navigation path using custom logic, from how they came to the site, found the product and what cart actions they took. Each of these paths could be aggregated and searched. For example, we could quickly identify the common navigation paths for a customer who clicked on a recommended product and added it to cart, or users that added an invalid coupon code. Since this data was layered into our BI system, we would examine onsite behavior differences by things like gender or market segment.
We were able to capture product metadata from the page and combine that with things like searched keywords to understand relationships between products for merchandising. We found common behavior paths in different customer segments and information, and how to quickly identify those segments - from geographic region, search interest or loyalty.
Finally, the combination of Cloudmeter Stream with web session replay added a layer of efficiency to our analysis. Not only could we email sessions throughout the team - allowing customer service, technology or marketing to see issues first hand, but we could also gain a level of insight into a problem in a much shorter time period. In the past, when an anomaly in the data was observed, it would trigger a time consuming deep dive. Through "data forensics," we tried to piece together theories for the change. Now, when we observe something unusual in the data, we immediately search sessions in Cloudmeter Stream containing the behavior anomaly and usually gain immediate insight into the cause. This simple yet powerful step frees up our analytics team to spend more time on strategy and optimization rather than on constant investigation.
As a result, Skinit is in the process of re-designing the eCommerce platform based on the insights we have learned from Cloudmeter. We expect to see some big gains from the new site, but more importantly, I think we are all excited to find the next place to continue optimization.
Reader Feedback: Page 1 of 1
Latest Cloud Developer Stories
Subscribe to the World's Most Powerful Newsletters
Subscribe to Our Rss Feeds & Get Your SYS-CON News Live!
SYS-CON Featured Whitepapers
Most Read This Week