Data-Driven Businesses Are Better Prepared to Emerge from COVID-19
It’s been suggested that COVID-19 has rendered investments in data obsolete. The assumption is that we can’t rely on past data because what and how consumers behaved pre-COVID doesn’t matter anymore. From a data analytics perspective, this makes sense. After all, when COVID-19 arrived, trend charts suddenly flipped upside down.
But data scientists look at the world through a different lens.
Many will confuse “data science” with “data analytics.” Data analytics provides visibility on the past and the present. It provides factual answers to questions like, “Have we reached customers in China?” and “Are we getting more traffic this summer than last summer?”
But data science is about the future. It leverages massive computing power to help answer “Should we…?” questions like, “Should we spend more of the budget in Indonesia?”
While the skeptics throw out all of the pre-COVID data points, data scientists use them as benchmarks to anchor new insights into how consumer habits are changing. For data scientists, change rarely nullifies data — in fact, change is exactly what data science is good at.
Uncertainty accelerates the need for adaptation.
Everyone agrees consumer behavior has changed significantly since March. Managers need to make crucial decisions in the absence of clear trends in their data: Should they stop advertising altogether? Should they focus only on important channels? Should they run more promotions?
Here are a couple of tips on figuring out the best strategy for your company:
Find Out Exactly What the Effect Has Been (Analytics)
Are all areas of the business equally affected by COVID? Is any marketing channel performing better than another? Are certain keywords performing far worse than others? Is there a difference between audiences?
If your business has already invested in Business Intelligence tools, you are ahead of the game. No matter how vast your datasets, big data tools are advanced enough to deliver insights in a matter of hours (including query construction). All major cloud players offer serverless tools that make it easier than ever to get quick answers: BigQuery (Google Cloud Platform), Redshift (Amazon Web Services), Azure Synapse (Microsoft), and independent player Snowflake.
All that is needed is someone comfortable writing SQL queries — an analyst, a data engineer, a data scientist — and they’ll quickly send over the graphs you need to see.
At this stage, you have visibility. With visibility comes new questions. For example, do some audiences react worse to messaging than others? Are certain audiences not using your app because they no longer commute to work? Just be mindful that before making crucial business decisions, you’ll want to make sure that you’re not following some spurious correlations …
Make Sure You Trust What You’re Seeing (Using Stats)
If you see a difference in COVID response between audiences, is that difference meaningful? You’ll want to use statistical tests to make sure the effects you are observing are not random or a side effect of something else.
Statistical tests are OK in the hypothetical world, but I highly recommend running experimental tests as well. For example, if a segment of the user base stopped interacting, is there a promotion to nudge them back? Run a test on a small number of your audience (or lookalike audience) and estimate whether the effect is likely to hold outside your test population.
This stage requires more statistical rigor than the analytical one. With that said, many analysts and most data scientists & statisticians will be well equipped to design tests and evaluate the results. They’ll just need a bit more time to deliver their reports.
Again, many cloud tools lend themselves to running statistical tests, like GCP AI Platform Notebooks (serverless Jupyter Notebook environment running Python and R), Amazon Sagemaker, Microsoft Azure Notebooks.
Automate What’s Working…And Do Less of What Isn’t
This is where machine learning (ML) comes in. For example, analyzing the last five months of advertising data could reveal that some of your most loyal customers prefer not to be contacted. A business could use machine learning to categorize users into four quadrants: the “do not disturb,” the “lost cause,” the “sure thing” and the “persuadable.” Then, concentrate your communication on those who like to speak while giving others more time.
Models can be built quickly using any of the big cloud players. Typically, you’d want someone proficient in Python or R to build the models, test performance, and embed them into wider data pipelines, but models can be built quickly in SQL using Google’s BigQuery ML, making ML more accessible to Analysts and Engineers.
Doing More With Less
Businesses don’t need years of data to get a clear understanding of consumer behavior. And gaining these insights doesn’t have to be extremely complicated either. While data science is a complex field, new cloud products are making it easier to derive business value from the existing data. Businesses already using data science to understand consumer behavior can quickly adapt to the new norm, and those that aren’t can get started with fewer operational overheads than existed five years ago.
About the author: Christoph Niemeyer is a London-based data scientist with MightyHive, a data and digital media consultancy based in San Francisco. Christoph has an academic background in cognitive science where he wired people up to neuroimaging machines and modeled their brain responses. After spending some time working in early tech startups, Christoph joined MightyHive in 2019 where he works on a variety of projects that blend knowledge about people and machines in order to improve marketing.