A content provider’s process that took 20 person days to complete now takes 0.5 person days of effort. A 40X cost and time reduction on central data processing, while taking the accuracy to more than 95% from earlier 75% rate.
The Context & Challenges
- The content provider has a wide agent network and a centralized call center infrastructure to collect content from over 100 cities of the US on every day basis.
- Manual agents and call center personnel prepared data in spreadsheets and sent to central processing
- The provider also gets content from over 25 providers over automated feeds in various formats.
- The massive centralized process took 20 personnel working through out the day to enable everyday content syndication from its syndication servers.
- The process was tedious, error prone and time consuming. It led to low employee satisfaction (high attrition), and customer satisfaction challenges with provided data.
- The centralized process cost overhead was extremely high
The Solution
Manual Data Capture Process
- An hosted data capture application that could be accessed by distributed agents to enter data solving spreadsheet tedious and error prone entries
- An automated suggestion system (for hosting application) while a person entered data to avoid spelling and formatting errors.
Automated Data Capture Process
- Vendor feed gateway that processed various feeds and harmonized the formats, corrected errors, normalized content and created a consolidated feed in on format.
Data Consolidation & Normalization Process
- Intermediate feeds were driven into an human assisted and continuously learning artificial Intelligence system to automatically detect errors, correct errors and normalize variations to provide homogeneous output to distributed content servers.
- Progressive learning to progressively minimize human intervention points and increasing process efficiencies.
- Operational dashboards to get content comprehensiveness, accuracy and timeliness metric
Technology
- Linux (OS)
- MySql (Database)
- PHP / Perl / Java

