What can COVID-19 teach us about modeling business outcomes?

2020 started a new decade with a bang- a worldwide pandemic to be exact. Coronavirus has forced many of us to change the way we live in just a matter of weeks. People are not only struggling with the most top-of-mind consequences of Covid-19, but also with understanding new phrases that weren’t always part of everyday vernacular. Phrases such as “exponential growth”, “asymptomatic transmission”, “R0”, “logarithmic growth”, and “lagging CFR” are being used more often as they become normalized media jargon.


These terms are already difficult to digest on their own, but the mathematical relationships underpinning them make it that much harder to fully understand what they mean. That’s the danger of something like exponential growth, it doesn’t seem like a crisis until it’s almost immediately a crisis. Jude Law’s character in the movie Contagion best explains exponential growth when his conspiracy theorist character proclaims.


While ironically, he’s not technically right about the time for the doubling rate of R0 of two, he is right that exponential growth is the doubling of the sum of all the outcomes previous to this point.


The reason this is so hard to intuitively understand is that it is non-linear and as humans we are used to experiencing the world linearly, indeed we are biased towards seeing the world that way.


One not so linear concept that we’re dealing with during the COVID-19 crisis is the idea of a lag. Simply, the amount of time between experiencing an outcome (in this case the recovery) and the case onset (the number of days that passes). The problem is that when people want to understand a ratio, say between recoveries and cases, they compare the totals in the same period; i.e. today we have 100,000 cases and 2,000 recoveries, so the recovery rate is 2%. This is clearly mistaken, as it doesn’t take into account the different lags that exist in the underlying cases; some started 10 days ago, some started today.


Perhaps the most egregious understanding of this is when the news reports that the number of cases of people succumbing to Covid-19 have not decreased since the “lockdown” five days ago and are still growing! The analysis of the efforts ends up not only misunderstanding the problem, but falsely reporting the efficacy.


In the above, the epidemiologists are not focused on the number of deaths, or the absolute number of new cases, but the rate of change in new cases, or the logarithmic growth. This is the flattening of the curve.


In order to understand what is likely to happen in the next four weeks and how to react appropriately today it is important to be able to model what we know about how the disease progresses and to measure the right impacts at the right time in order to mitigate the full impact of this pandemic.


What does this have to do with business modeling you ask?


The first thing this teaches about business analytic is that we need to prioritize understanding the characteristics that underpin our business. What drives new customer growth? What are we measuring when we look at driving growth? What is the lag between key points in the customer journey (first engagement -> sale)? If our marketing actions are focused on driving first engagement, we have to be careful of ourselves and others measuring the wrong actions. For example, if we have a 10 day conversion window between new engagement and sale, if we measure the daily ratio of both our conversion rate will go down. So notice here that setting the right measures is always critical.


Marketing to drive new customer sales is likely to have a lag condition- the sales cycle. Understanding the conversion cycle is critical to measuring the ultimate impact of the campaign. Setting upper funnel, more immediate measurements, that we have proven as strong leading indicators will allow real-time campaign performance management.


The second thing we learn from the pandemic, is about how modeling the behavior between consumers and advertiser marketing efforts is this more esoteric conversation of non-linearity. Because the relationship between marketing and consumer behavior is rarely linear. The following are some simple examples:


  1. Adstock: A media exposure today has an influence on a customer decision tomorrow. This is basically consumer memory. If I had 100 customers today from an ad I placed today, and 50 tomorrow, then 25 and so on, I would say “the adstock is 50%”.

  2. Diminishing Returns: The more investment is placed in media the less effective the incremental sales are. SO, if I spend $100 and go 10 sales, then spend $200 and get 19, $300, 27 and so on, I would say the diminishing return is 90%.

  3. Offer Pull Forward: Promotions don’t just bring in new sales, they can pull the buying behavior forward. Imagine we had 200 sales per day and we do a 50% offer and get 600 sales! But then in the following days we notice that right after the sale we only get 100 sales, then 125 the next day and so on until 5 days later we return to normal. In that period we lost 250 sales, so overall the campaign only generated 150 more sales, not 400. This modifies the elasticity of the offer quite significantly.

  4. Demand: It helps. No amount of advertising or offers selling precut Christmas Trees on July 4th is convincing me to buy a tree, though I might consider buying ski boots that are 50% off. As demand (or seasonality) increases the likely response to your media campaigns will also increase and the need to reduce price with discounts also decrease. In modeling terms, we refer to this as synergistic, and will generally model this as a complex function that conveys the “amplification” that occurs between these events. To carry on the simplistic example, as the market can bear more ski boot advertisers spending more to reach customers, because more customers are available to respond; demand is higher. In modeling terms, or diminishing return for increased spend will be less.


There are many other “transformations” that will take place in building a robust model of your business, all of which are essential to allow the business to quantify the impact that media, seasonality, and demand shocks have. It is important to understand what is going to happen next and how to react and plan.


In the coming weeks we will go into more detail about how to build a business model and how it can be used to manage the business in less certain times