AI Explorations of a Former Media CTO (With No Prior Machine Learning Experience)

One of the challenges many enterprise leaders face is how to experiment and quickly create business value from machine learning without establishing large teams with specialized skillsets and infrastructure. As a CTO, I found that having access to Amazon Web Service’s pre-trained AI Services enabled our developers, even those with no prior machine leaning experience, to build AI-driven applications quickly. Even more empowering personally was how easily it allowed me and other senior leaders on the team with somewhat rusty hands-on coding experience to experiment and think about meaningful use cases. This is where AI has tremendous power to unlock business value in every industry, not just by lowering the cost of entry, but also by making it accessible to all skill levels within the enterprise.

Over the holiday break, I set a personal learning goal to explore one of these AI services: Amazon Textract, a service that uses machine learning to extract text and data from documents. My goal was to feed the curious builder in me, and also to figure out potential high-impact use cases that leaders can experiment with. I played around with a number of different types of documents, including standard government issued IDs, bank and credit card statements, tax documents, and of course…my cable bill!

As I started to explore more, I was fascinated by the opportunity to reimagine so many paper-based workflows involving manual data entry. There are two potential use cases in media—specifically television—that I want to share. They impact two of the largest revenue-generating areas for a TV company.

Advertising — Traffic Copy Instructions

By various estimates, annual TV advertising spend in US is north of $70 billion. The process roughly works like this: Advertising time is purchased in advance based on the goals the advertiser has to reach a specific mix of audiences. As the year progresses, agencies who represent the buyer (advertiser) will issue copy instructions to TV networks specifying which ads to run in which timeslots, the flight dates, and the rotation percentage. These copy instructions are sent via email or even fax with very limited standardization. The commercial operations teams at the TV networks will then take these instructions and enter them into the traffic system to schedule ads for the broadcast. Mistakes made because of this heavily manual workflow lead to lost revenue. In some cases, the network must issue expensive makegoods, resulting in even higher lost opportunity costs. A single thirty-second prime time spot can cost six figures! Over the years, TV networks have hardened this process by having multiple levels of checks and reviews, but the process is still extremely labor intensive and error prone. There are also some rather expensive Optical Character Recognition (OCR) solutions in place, but they mainly provide a review and approval workflow, making the manual process smoother rather than eliminating it.

I believe experimenting with Amazon Textract to automatically extract and ingest the instructions into the traffic system can eliminate costly errors. More importantly, it will free up valuable resources from the ad operations team and allow them to focus on differentiated work.

Affiliate Fees — Billing

A large part of a TV network’s revenue, besides advertising, is affiliate fees. Affiliate fees are what the multichannel video programing distributors (MVPDs), or cable companies, pay per month per subscriber to the TV networks. For the billing process, cable companies send monthly subscriber reports in PDF to TV networks based on different packages that carry their channels. This data is then manually entered in the affiliate sales and billing system at the TV network to calculate the charges based on the subscriber numbers and the rates. The final invoices based on these charges are in turn sent to the cable company. This subscriber data is very valuable, not only for billing, but also for analyzing trends and forecasting revenue.

I believe this is another use case where Amazon Textract can help by automatically ingesting data from these PDF documents. This should not only cut down manual effort, but also provide insights into subscriber trends faster.

There are a number of other AI-based strategies that can create tremendous value for a media company, like extracting rich metadata using Amazon Rekognition or automatically creating closed captioning and translations using Amazon Transcribe and Amazon Translate.

There are a few things I find very compelling here:

• the ability to utilize existing developers and skillsets;
• the ability to start experimenting very quickly without upfront investment or infrastructure; and
• unlocking creative ideas without a barrier to entry.

I would love to hear your thoughts on this and any other ideas about creating business value using AI.

– Ishit

@ishitv

vachhraj@amazon.com