Five Ways to Bridge the Gap Between Data Science and the Communications Industry
Drew Shives, Data Scientist
At Marina Maher Communications, we are continually struck by how much the modern data science practice has to offer businesses in all fields. Cutting-edge practices and techniques have left the realm of academia for industries at an exponential pace. No longer is the question whether machine learning or artificial intelligence has a place outside of scholarly research; but how we use these technologies to reach innovations and insights faster and more efficiently.
Think, for example, about recent advancements in digital advertising. Algorithms now control nearly every aspect of buying and selling ads, shifting the process away from traditional insertion orders to programmatic supply and demand optimization. Google, Facebook and other advertisers synthesize millions of individual data points to figure out who should be receiving which ads, factoring in social engagement, add-to-carts on direct-to-consumer websites, and even bricks-and-mortar store visits.
The PR/Communications industry still has upside opportunity in our application of machine learning and other advanced analytical models, although I’m proud to say MMC has been making progress and innovating on a number of fronts: first, an approach for clustering audience segments and testing message effectiveness using a statistical framework built on conditional probabilities; second, the development of two proprietary Web apps that enable our influencer marketing team to leverage deep understanding of historical campaign performance to forecast future campaign performance.
Below are five ways we've improved the connection between Communications and Data Science to build predictive analytics capabilities for our clients:
1. For communications, better data collection.
The crux on which every data-based project hinges is the idiom, “garbage in, garbage out,” which serves to mean that if the input data is less than pristine, than any model’s output will be as well.
Within the communications landscape, there remains several opportunities, from resolving the cold start problem of not having enough initial information, to improving data centralization and cleanliness, to standardizing metrics terminology. One way we have been tackling the cold start problem at MMC is by investing our own first-party data assets. Instead of delivering reports as email attachments, we have started making it standard operating procedure for teams to aggregate their data into a centrally managed data warehouse. This builds our own set of first-party data assets for historical bench marking, predictive modeling, and even web-based data applications on top of so anyone at MMC can interact with the data.
So, if the initial question is whether or not to collect data, always lean in the former direction, of course in conjunction with all PII guidelines and requirements. All data is inherently useful until proven otherwise. And the more data that is collected, the better chance there is to gain deeper, unknown insights.
Data collection, however, is only the first step.
2. For data science, understanding the capabilities and scope of given data, teams, and goals.
As data scientists, we constantly try to push the envelope further. The more advanced and complicated the underlying math, statistics, and programming, the more excited the data scientist.
As such, we often have a strong desire to try and replicate the latest and greatest advancements for every problem. Can we utilize deep learning for this? Is reinforcement learning appropriate here? This admirable attempt to push further, however, can lead to answers that are misaligned with a given question. By first looking closely at the data and aligning with the team and the goal, we can then apply prescriptive solutions rather than generic ones. Thinking about your team’s capabilities: strengths, weaknesses, knowledge, and gaps – will then help shape what insights can be uncovered. If not for the MMC Performance Analytics Team’s interest in Bayesian data analysis, we likely would have not arrived upon our methods for audience clustering and message effectiveness; or been stuck with standard tools that might not prove as helpful.
Goals require a deeper discussion about success overall.
3. For communications, better definition of what constitutes success for a given problem.
Thinking about the desired end action of a problem is an essential component to a successful data science initiative. Do we want to classify the data and determine if it fits into categories? Or is the goal of a model to see how much a change in one variable affects another? Each question raises a different solution, but sometimes the answers are not as easy to reach.
In advertising, there are clear goal states. Clicks on an ad, add-to-carts and many other actions define what success looks like for digital advertisers. These actions are discrete and trackable, making it much easier to build models around since the end goal is well defined.
For communications, however, these end goals can more nebulous, given our content is often produced and distributed by third parties like editors or social media content creators, and consumers respond back in a two-way dialogue.
By firming up how communications tracks and measures success, data scientists can get a better understanding of how to implement models, build forecasts, and answer questions relating to business objectives. Which brings me to my next point for data scientists…#4.
4. For data science, clearer understanding of business objectives.
If left to our own devices, we data scientists can ramp up the complexity of a project quickly. But complexity without a clear perspective on key business objectives will only leave stakeholders with confusion. It is incumbent upon data scientists to seek out a broader point of view on what is ideal and what is sought after by other teams and clients. At MMC, one of the things that grounds our data science practice is our work with influencers, and our partnership with the client leadership and digital innovation teams. Across both consumer and healthcare verticals, we utilize influencers to communicate our clients’ messages and raise awareness, credibility, and relevance of their brands. Our clients expect we track and optimize in real time, so having the apps that give us a pulse on performance expectations is critical to staying one step ahead and adjusting as necessary. All of that though would have been for naught if we all did not consider #5
5. For all, thinking bigger and being willing to dream.
The future is boundless and despite the practical realities of working with specific data and limited resources, both communications professionals and data scientists need to remember to spend time thinking expansively. The only way innovation will continue, and for the data science revolution to touch communications in meaningful, long-lasting ways, will be to dream of that potential.
If you are interested in either data science or communications and want to talk more about how we’re working to bring these two fields together, say hi at email@example.com.