Hi I’m Tanya Shastri. A little big of background on myself before we get into the presentation. I’ve been plugged into the big data world for a while. A relatively short 5 years ago, there were no conferences, let alone conferences of this scale. Unconferences back in the day. Though not surprised at all to see the size of the conferences today. There is a lot of promise in data and tools like Hadoop are helping deliver on that promise.
And while I’m not at Natero anymore, they are…
I’ll be talking about how having access to big data can provide some actionable insights to reduce the cost of customer support and improve customer experience and customer satisfaction.
Here’s how I’ve structured the presentation.
First some context Then I’ll go into the methodology – and some sample analytics
For the previous part we’ll assume that all the data has is automagically prepared.. So a spotlight on data prep because any of you who’ve worked with data will know that data prep is often if not always the harder part.
We’ll end with some considerations and learnings…
To provide a little bit of context for what kind of company this would apply to. A company that provides consumer products whether hardware or software.
The can be extended to the internet of things in general.
Disparate sources, geographically distributed, last thing you want to do is add another source.. A source that is “big”.
Combining a couple sources In some cases even using a source of data that isn’t typically used
Thousands of products – each having its own troubleshooting page. The intention of the troubleshooting page was to enable customers to self-help. There was no insight into how these troubleshooting pages were performing. If they were able to know which pages were not performing well, they could improve them and reduce incidents filed. For this two sources were used – the web-clickstream data from the support website and the incident database.
Thousands of products – each having its own troubleshooting page. The intention of the troubleshooting page was to enable customers to self-help.
There was no insight into how these troubleshooting pages were performing.
Enterprises are looking for ways to improve customer satisfaction and reduce support costs, but often do not have actionable insights. Traditional approaches and tools fall short, often based on small biased datasets and requiring long turnaround times. This talk will cover the steps involved todevelop a big data solution for support through the example of a leading vendor of electronic consumer peripherals. Topics will include: The methodology and metrics developed:Metrics to track and improve self-help through support sites Metrics to track the end-to-end support process to find delays in processing of incidents, escalations, etc. Methods to identify problem areas based on data from discussion forums Data-driven discovery of paths that customers prefer for support The data wrangling required to implement the solution using a big data analytics platform:Big data analytics techniques to track customer behavior across channels Preparing the data for analytics: joining, merging and enriching the diverse datasets Validating the parameters and techniques used for analysis Considerations for an iterative analytical approach to get results with the highest confidence interval Metrics at various granularities to meet the needs of various business decision makers Automation for maintaining and tracking up-to-date results
1. Customer Support
Big Data Era
2. What will be discussed
Customer support context
From data to business benefit
Preparing the data
Considerations and learnings
3. Customer Support Context
4. Customer Support Data Sources
Incident databases, call center data
Customer self-help website
In-product or in-app data, call home data
Disparate ● Disjoint ● Silo-ed
Structured ● Semi-structured ● Unstructured
Transactions ● Events ● Logs
Volume ● Velocity ● Variety
5. Methodology and Metrics
Tracking and improving self-help
Troubleshooting page score
In-product proactive support
Customer behavior based learning
Improving support efficiency
6. Tracking and improving self-help
ScoreC = #incidentsP1
#“P1 help page hits”
ScoreD = (#incidentsP1 - #rmaP1)
#“sessions P1 help page hits with t>10s”
RMA Return Merchandise Authorization
t duration of time spent on the page
9. In-product proactive support
Support integral part of product development
Customer behavior driven analysis
Analysis based on a sequence of actions
#A1 reach here
T2 spent here
#A2 reach here
T2 spent here
#A3 reach here
T3 spent here
10. Analysis of sequence of actions
Action 1 Action 2 Action 3 Action 4
#taking this action Avg Time (sec)
11. Discussion Forums and Reviews
Discussion Forums and Reviews
Term frequency trend
Time at each step in actual support process
12. Preparing the Data
Standardizing Product nomenclature
Enriching events/logs with traditional data
Session based analysis, session segmentation
Product hierarchy for drill-down to various levels
13. Be aware that…
The process is likely to be iterative
Data prep is a big deal
With “big data” simple analysis can be valuable
Analysis can sometimes feel like “hindsight is 20/20”