Vice President, Product Management and Marketing
In this excerpt from the New Product Innovation & Development Chronicles, Morgan Hays explains how TiVo improved operations and results by automating many data processes and analyses and combining machine learning with platform mediated work techniques.
Data is King. The gap between companies using their data effectively is widening. Hear how TiVo is framing their data analysis and acquisition strategy to monetize their organization’s unleveraged data resources.
- Best practices to analyze and decompose the quality of your data
- Insight into data acquisition strategies that ensure ongoing data flow and quality
- Examples of organizational challenges to shift to a Data-Centric mindset
Metadata is data taken in from thousands of sources and aggregated. For TiVo, the data is comprised of titles, cast, logos, channels, and other information related to entertainment shows and movies. In the early days, the data was entered into the system manually to generate the weekly TV Guide magazine. Manual entry was sufficient at that time because the market was so much smaller than it is today and viewers watched on only one type of device, a television. Today, in one recent year, there was a 171% increase in overall content in the home viewing market. Further, customers expect the same content to be delivered on whichever platform they are using to view the content.
Today, the market expects more scale, higher speed, lower cost and accuracy.Responding to trends in the market, it was imperative for TiVo to make a series of changes in its workflow. TiVo wanted to differentiate itself by its technical prowess and how that impacted the customer.
There are five points in the TiVo workflow:
- Rationalize – sources of cost
The rationalize point is the most important. What are the sources of cost and how can TiVo generate more income? Guidelines:
- Segment data by quality in relation to its ability to be automatically processed
- Combine machine learning with platform mediated work techniques to drive cost reductions and processing speed improvements
TiVo takes in a variety of data in large quantities, with no standard for how it is delivered to TiVo, e.g., word file, email, excel, etc. Some content has x data points, other content has y data points, additional content has z data points, etc. The goal was to unify all the data and the key was to segment all the data by the quality of the data they were taking in.
Data Segmentation – The data were segmented by quality. The high-quality data were addressed by a machine learning system and the lower quality data were addressed by a platform mediated system (i.e., bringing in temporary contract workers.) The quality of data was determined by randomly selecting data over time to create a true set using the manual processes already in place. The true set was used to make an algorithm, which was repeatedly tested against the manual processes, and tweaks were made to the algorithm to improve machine learning. The goal was to avoid false positives, e.g., to avoid labeling an adult movie as a kid movie.
Content Decomposition – The team set out to determine which variable (e.g., cast, genre, and title) is most often tied to accuracy. Content decomposition was accomplished by first determining how they wanted to segment the data and then breaking the data into quintiles. Quintile-based segmentation drives pricing. By decomposing the data, they were able to determine profitability and pricing. Higher quality data allowed higher pricing and speed and, consequently, improved the customer experience.
Platform-mediated work – For the low confidence data, TiVo augmented machine learning systems with human-based creation on demand. Temporary workers identified on-demand worker platforms that allowed TiVo to bring in the work force, with correct skills, at the time they were needed, and at scale needed. They produced a training manual, created a series of job descriptions that would be posted, and invited freelancers to sign up. This allowed TiVo to dynamically scale the needed workforce very fast and brought the right people into the business at the right time; when their specific skills were needed.
Outcomes for TiVo: Their largest success was increased speed and decreased cost. Delivery speed went from 48 hours to 1 minute; operating efficiency decreased while accuracy remained high at 99.4%. High confidence data could be automatically pushed through because TiVo had high confidence the data were correct. Low confidence data was not worth the risk. Allowed company to make investment decisions and encourage customers to improve the quality of data they provided at start.
Rationalize – Segment Data by Quality To Perform Matching at Scale
Match requests across thousands of sources worldwide. Use high-quality machine learning:
- Randomly select data from across the dataset across world over a period of time
- Apply sampling process in place of manual to acquire a true set
- Provide the true set to the data science team to find an algorithm
- Repeatedly tested it – data in, run through manual process and machine learning to test match
- When meeting quality needs, figured out which inputs were the best at leading to best match (e.g., title, genre)
The Tivo key to success:
Sales enablement and content decomposition. They told their sources to get the best data and were transparent with customers.
- Segment your data. Separate into high and low confidence. Low confidence data is not worth the risk. By separating into categories, Tivo could decide where to invest in machine learning
- High/low segments drive speed and cost reduction
- Measure and avoid false positives (Example: Don’t add an adult movie when something else is expected)
- Follow the Hippocratic Oath: Do no harm. TiVo measured where they were and when they were ready to move forward
- Quintile-based segmentation drives pricing
- Understand profitability based on customer
TiVo moved from a manual model to a mixed manual/automated process. In doing so, there were some significant improvements in ability to compete in the market. TiVo aligned their operations team to coordinate their efforts by data assets; better alignment meant better planning and allowed the company to know where it wanted to invest.