If retailers were to rely exclusively on recent historical data, they might conclude that UK consumers buy their annual supply of toilet paper and pasta in March. Why? In early March 2020, UK consumers were expecting the country to go into lockdown. This drove panic buying, especially for these staples. Today that information remains in companies’ customer insights and marketing databases alongside other predictive buyer behaviours like turkey sales in December or pumpkin sales in October. But which behaviours are likely to repeat, and which are anomalies?
The issue continues in the (fingers crossed!) post-Covid-19 period as some sectors enjoy a substantial bounce back. Should companies, having seen 50-100% year-on-year growth, believe it to be a trend and plan for the same in the coming year? Doing this might be sensible, but it could equally mean missed forecasts and excess stock sitting on shelves.
And the problem extends beyond supply and demand and customer engagement data. For example, I see anecdotal evidence of virtual event attendance levels dropping off alongside hesitancy to commit to in-person vent attendance — an unwelcome double whammy.
A colleague recently suggested that while it might be obvious to people what data represents a trend, it may not be so obvious to an algorithm. This issue is highlighted in Forrester’s 2022 Global Marketing Survey, where respondents identified the rapidly changing business environment making historical data meaningless as the top measurement and analytics challenge.
Marketing leaders must carefully consider how to interpret and use data that may be anomalous. Consider this example from my first job: While some clients were visiting my company’s network operations centre, the massive digital network map lit up with red warnings. The visitors were excited to a network crisis response first-hand; however, the staff carried on like nothing was happening. When they queried the presenter, he said that what we don’t want in a network centre are the proverbial headless chickens who react to data without thinking (this was pre-ML/AI). Diving into a network and making immediate changes based on a short-term change in the data was often detrimental. The policy was to wait to see if the data point became a trend and only then take action. As if to prove his point, the red alerts started disappearing from the screen and the network went back to normal.
There is no magic formula that marketers can apply to decide what to do with the last three years of customer and market data. All businesses are different, and each will need to think about their own circumstances and what impacts the pandemic had on their customers’ and markets’ purchasing patterns. Based on this information, companies can decide whether the data that you’ve captured remains valid, needs to be calibrated, should be quarantined, or should be erased. What’s important is that you make conscious decisions to avoid sleepwalking into automated data-driven mistakes that could seriously impact business performance or balance sheets.
