Predicting the future
What epidemiology can teach us about how to manage brands
Chief Business Strategy Officer
Head of Strategy, EMEA
OK. We know what you’re thinking: we’re all armchair epidemiologists now. But bear with us ...
In recent months we’ve seen a lot of coverage of simulations, built by epidemiologists, used to understand the impact of COVID-19 and guide policy-making. When the UK shifted strategy from mitigation to suppression, it was based on an Imperial College simulation of infections and fatalities. When Donald Trump claimed to be “saving 2 million American lives”, it was based on a simulation of how many would die if we did nothing. And it took a Washington Post article demonstrating how social distancing works for most of us to understand what ‘flattening the curve’ actually meant.
Simulations help us understand “what if”, not just “what”. They show how certain choices or changes play out: who wins, who loses, and what the world might look like in different scenarios. Epidemiologists and policy-makers use them because they describe a range of possibilities that are sensitive to our actions.
As marketers, we face similar challenges right now. In stable times, forecasts based on data from the recent past (for example, the outputs of marketing mix models) give us a good guide to what we can expect in the near future. Indeed, most of the time we’d be mad to expect the ROI of our advertising to double or halve overnight. But normal forecasts break down when the world shifts around us. We’re seeing that right now: we don’t know how long recent changes will last, or how people will behave when conditions improve. Because we can’t forecast a single right answer, we use simulations to plan for different possibilities.
At Essence we have worked with data science specialists Sandtable (now part of WPP), to build models that help answer some of our client’s biggest strategic questions. These simulations (also known as agent-based models) are a virtual view of the marketplace, a marketing “sim-city” made up of economic, marketing and brand-specific data, where we can test different strategies over months or even years.
These simulations let us see “what if short-term changes in behaviour persist?”
We used a model based on Kantar’s Worldpanel data to look at supermarkets’ market share and shopper loyalty depending on whether people continue buying groceries online, or switch back to shopping locally as restrictions are lifted.
The preference for online shopping during the height of the lockdown way outstripped supermarkets’ capacity, so any retailer’s share was more dependent on its capacity than its brand - meaning bricks-and-mortar-only stores like M&S could steal share by introducing an online offering.
But a returning preference for shopping local, and reducing drive-times to stores, would have a big impact on the retailers that do best, with convenience-focused brands like Co-op the biggest winners, and those without small-format stores, like Asda and Morrisons, hardest hit.
We have also built a model of the telco networks and handsets markets across the US and UK, incorporating BrandZ’s equity data, Kantar’s Comtech panel and macroeconomic trends to investigate the long-term impact of post-coronavirus economic pressures.
We looked at what would happen to brands’ market shares if consumers downgraded to lower spend tiers: not much, as most would trade down within their preferred brand. We also investigated who would win if certain groups shifted back toward pre-paid options, perhaps to mitigate the financial “risk”, or if people started prioritising value over quality.
But most interestingly, it let us explore potential category disruption if small brands in the category went bust, or how much of a threat a brand like Amazon or Google might be if they bought an existing player: quite a lot, it turns out, if it’s Amazon.
Simulations don’t just show us what might happen, they help us find out how to win by “trying out” different options.
In the run-up to Christmas last year, L’Oréal wanted to launch two new fragrances without cannibalising their existing brands. We used a simulation model based on their brand tracking data to identify which associations would be most valuable to build for each brand, to make them appeal to different segments and defend against a range of competitors. This helped us work out the best positionings for all the brands in their portfolio, and how to launch the new ones. Both YSL Libre and Idôle by Lancôme cracked the top 10 in the critical pre-Christmas period.
Each of these examples used very different simulations. Some contain vast quantities of data and have huge predictive power; others are simpler, rules-based models that show us how different forces interact.
What they share is the ability to play out future worlds, and see the impact of different choices, so we can pick the route to the best overall outcome. They help us learn from the past without assuming the future will be the same. As statisticians say, "All models are wrong, but some are useful" - and right now, we’ll take all the extra utility we can get.