American broadcast radio is a highly competitive industry that is, yet again, undergoing significant change. Traditional broadcasters are being pressured to adjust long-established business practices by a growing number of rivals in digital, satellite, and Internet radio, including streaming services such as Pandora and Spotify. More than 15,000 stations vie for $18 billion in annual U.S. radio advertising spending — a figure projected to grow by less than 1 percent annually through 2021.

To generate revenue, broadcasters must increase listenership. To attract more listeners, they have to play music by artists that their target audience wants to hear. Traditionally, radio playlists were created intuitively, based on personal experiences or industry connections. Today, however, programmers engage in very careful curation, tapping data from relevant research to customize playlists that will ensure maximum listener satisfaction.

We needed to deliver a solution that focused on driving ​workflow alignment and data standards for 800+ program directors and 4000+ disc jockeys​ using cloud-native applications.

The challenges radio stations face are determining which data sources to trust, and then deciding how to put data into action. Stations, themselves, collect data; but, it is often stored in analog fashion — on spreadsheets, in file folders — making it difficult to analyze using cloud technologies at scale.

Laying the Groundwork

Team
Designers: Director of Experience Design: Salina Brown
Principal Experience Designer: Raymond Macari

Engineering: Principal Cloud Architects, 2 Senior Cloud Architects, Business Analyst, Project Manager, and 3 Software Engineers

Timeline: 8 sprints

To get started, we led an Experience Design (XD) Discovery Workshop with key stakeholders — including the CEO and music industry executives — to unlock key goals and problems the company was facing.

Key Players — The Personas

We identified the industry personas —​ “power players,” “industry front-runners,” “program directors,” and “industry veterans​” — each of whom had different motivations and different experiences leveraging analytics and technology.

• Power players​ sit at the top of the food chain, they focus on where to invest capital based on an artist’s traction in a target market
• Industry front-runners​ care about “What’s next?”
• Program directors​ have to balance brand value and listenership. “Is this song on brand?” and “Will it increase my listenership?”
• Industry veterans​ seem to be driven by tradition and can, at times, struggle to make sense of the data

Observation

We interviewed key music industry personas, including programming directors, and shadowed them, documenting their day-to-day responsibilities. Most of the personas we observed used data as a secondary tool to validate assumptions, often resulting in confirmation bias. Also, because so much was done on paper or in Excel, outcomes could not be tracked back to the data.

Other key findings:

• On average, assistants spent most of their day compiling data from many sources to deliver a one-sheet on the desk of an executive by the following morning
• Every day, record label executives spend most of their day, promoting their artists, on calls with program managers at radio stations, social media influencers, stream platforms, and large playlist programmers
• On average, music executives are responsible for 1 — 3 artists who may be at different stages of their careers but have similar characteristics, such as genre
• Record label executives believe that they have a scientific formula for determining success
• Everyone has access to the same data, and there is a lot of it
• Decisions are still being managed on paper or Excel

Aligning on Assumptions

Our observations supported the business’s goals — We needed to drive workflow alignment and data standards for 800+ Program Directors and 4000+ Disc Jockeys through cloud based application.

We used real data in our prototype, collected using Google Sheets and JSON, to explore visualizations of a song’s trajectory against an artist’s life cycle and a song’s life cycle. We explored several visualizations using simple bar graphs, choropleths, and trend lines to better understand what metrics could be correlated. This led us to the discovery of new metrics that improved the benchmarking of artist and song life cycles.

Using their existing worksheet model, we prototyped a collaborative web app that displays daily song metrics and allows programming managers to rank and organize songs for radio airplay. Analyzing the programmer’s playlist decisions is crucial. It leads to insights about how programming decisions are linked to listenership and brand position, and it improves the business’s ability to track outcomes to decisions more directly.

Delivering Solutions

We delivered a minimum viable product (MVP), this solution consisted of four components:

• Music Analytics Dashboard — providing views into artist metrics and song life cycle
• Music Research Worksheet — combining historical data from multiple sources, ongoing research, and predictive metrics so local stations can collaborate, develop music schedules, and make better data-driven decisions in a single unified view
• Serverless ETL pipeline — meshing data from disparate sources that, in turn, powers the application
• Cloud Search — enabling an easy form of navigation and discovery across the MusicLab user interface

Measuring success

Client Success: One year later, the client reported a positive shift in focus of program directors (OpEx savings) + Increased listeners on every radio station that piloted the program.

Team Success: Design-led $250K pitch turned into a multi-year, multi-track, $12M software delivery engagement.

Techstack

The Music Analytics Dashboard application follows a traditional three-tiered architecture with a presentation layer, a business layer, and a data layer. The presentation layer is an AWS Elastic Beanstalk application, running Node.js, and built using React framework components. The business layer is in line with the current trend of serverless architecture — written using Amazon API Gateway and AWS Lambda, along with custom authorizers, and integration with AzureActive Directory via their OAuth interface. The data layer is a multi-availability zone AWS RDS instance using the PostgreSQL engine, due to the relational nature of the data.

The serverless ETL pipeline is a MapReduce pipeline built using serverless components instead of Hadoop. Key features include abstracted data ingestion that connects to FTP, S3, Google Cloud Storage, SQL databases, SOAP services, Windows network shares, and HTTP data sources; Node.js modules for all things ETL; and async streaming between different data protocols.

The Amazon CloudSearch component provides the application with an autocomplete search bar, as well as a full search results page that includes fuzzy matching.