Many TV businesses have vast amounts of archived data going back years. It might be great to have a large volume of data to work with, but it also means you need a complicated process to figure out what data the company has in a report that lists these assets in a straightforward, easy-to-understand manner. At Deductive, our data science consulting team generally counsels that companies should be primarily looking to see what business value these data have for their company, and how they can use these data to increase profitability and efficiency.
Data mining is the process of extracting useful information from data and is typically done using some predictive analytics, whether off-the-shelf or with software built by the company. We strongly recommend the latter, i.e., doing your data mining, for any company which has the data science team and the resources to do so. The goal of any initial data mining is to extract a detailed report about what the data contain and what business value they have. Only then can they accurately figure out what to do with the data to maximize their profit, while at the same time deleting any unnecessary data and pseudonymizing any PII data.
Getting the data ready
Once the team fully understands the business value of the data, it is necessary to understand the data which they are mining. This process is different to the extraction of the data itself and involves understanding where the data have come from, typically from a whole variety of different places, and how and how often the data team is collecting additional data to add to the original datasets. So data about how customers interact with an app will be added to, perhaps daily, as these interactions happen every day. The way to do this is to describe the data, noting any which are particularly important (such as PII data). Once the team has understood the data, they need to be validated and cleansed. Failure to do this would seriously compromise the data mining.
When the team has adequately prepared the data, they will be ready to be modeled by using mathematical modeling techniques for mapping the data and for the predictive analytics process. The models may not work, typically because of poorly prepared data, but also for other reasons where the data mining team will need to figure out what is the problem; successfully doing this is a significant step forward. The final stage before life deployment can happen is the evaluation, where the model and the data are tested to see what sorts of analytic insights they produce. Checking the integrity of these insights is essential. Using incomplete or inaccurate data mining insights in the business could create a worse situation than doing no data mining at all. Merely because analytic results look coherent after a brief review is not a sufficient reason to assume that they are right. When the evaluation stage is finished the data mining models are ready to go live.
Take a look at your competition
Once the data science team has completed these stages, they should then try to assess, in as far as it is possible to do so, what data their rivals, and any other companies in the same industry, both have and lack. In this way, the team can build a more accurate picture of the business value their data may have.
Building this picture should include the potential the data have to do new tasks, which needs data quality. So the company might be using a dataset in a particular way, say in identifying potential new customers, but the same dataset could also be used to help reduce churn for actual customers. The data team can only find new ways to use the data by ensuring that their mining focusses on comprehending the business values of the data.
Perhaps a rival added OTT to its linear offering in one region and, in spite of some implementation issues, the service has been very successful. What data are they using to make their VOD service better? How can ee avoid making the mistakes that they have made? Or perhaps a rival has devised a new and very successful segmentation system, marketing specifically towards sport into enthusiasts and casuals. If they are growing market share, then the question is how to acquire the same data assets to implement the same strategy?
This particular way of segmenting would have been an innovative and risky decision for the company. However, once it is clear what a great decision it was, with the data showing why and how the conclusion was successful, it's easy for a data science team to replicate the same strategy. Most of the data which helps rivals increase upselling and reduce churn is publicly available for a fair market rate and must be of interest to other divisions which want to do the same.
Mining the archive
Broadcaster's archives are a potential goldmine. If a company has tens of thousands of shows going back to the seventies, they usually syndicate these shows on OTT services both in-house and externally. However, there is also likely to be a lot of data associated with these shows, which also have business value. Descriptions of these programs and other metadata could now have great value for the website of the OTT service. Or by digitizing the scripts from thousands of drama and comedy shows, including both the smash hits and the flops we can use artificial intelligence, to help assess new content, focussing on the best ideas and techniques from the smash hits and avoiding the worst pitfalls of the flops.
This might sound like content generation by machine, but as well as using data mining to profile your customers and potential customers, it's important to focus on the applications of data to the full breadth of the company: if you're not trying it, you can be sure that your competitors are.