The Unexpected Impact of the Washington Post’s Non-Endorsement on Their Data Team
The US presidential election is just under a week away, and as such we’re seeing many institutions, celebrities etc. endorse their chosen candidate for the presidency — a decision that can be benefit both the presidential candidate and the endorsing person/institution. Don’t worry, this is not another data guy trying to predict the outcome of the election, or trying to use data to nudge you to vote one way or another — this is far more interesting.
One institution that has chosen not to endorse a presidential candidate is the Washington Post. The publisher has stated that, for the first time in 36 years, they will not endorse a presidential candidate and don’t plan on doing so in future presidential races (more here: )
This decision has drawn criticism from many people and resulted in the Washington Post seeing a large number of subscription cancellations, which various sources suggest amount to around 250,000 people, or about 10% of their active subscriptions base.
So, what’s the data problem? It’s, “What is the value of these lost subscriptions?” Which on the face of it seems like a straightforward question.
While we can’t come up with a concrete answer without their internal data, we can discuss how we would go about this. We’ll also look at why this may be a much bigger problem than it appears, and why it may cause bigger problems for the Washington Post’s data team in the weeks and months to come.
Before we get into it, let’s remind ourselves that any brand with a subscription offering will have a wealth of user-level data. This will include everything from when a user signed up, how often they login per day, the features they spend most time engaging with, the genre of content they prefer, which notifications drive increased engagement, etc. From this data, the data science function typically builds multiple models, ranging from cluster analysis and descriptive modeling through to predictive modeling on the next best action to engage users.
One of the largest pieces of ths framework is usually the Lifetime Value (LTV) modeling, which can range from simplistic models to predictive LTV models that essentially give the brand a view on how much each subscriber is potentially worth before they even sign up, based on data the company can access. This, in turn, helps them set the right budget for an ad they might serve to a potential customer.
Ok, detour over, back to the Washington Post. They have multiple tiers of subscriptions:
- All-Access Digital (Monthly or Annual): $12 per month, $120 per year
- Premium Digital (Monthly or Annual): $17 per month, $170 per year
On top of this, they have first-year incentives:
- All-Access Digital (Monthly or Annual): $4 per month, $40 per year
- Premium Digital (Monthly or Annual): $6 per month, $60 per year
The subscription type a user chooses is another example of a variable that gets fed into a Lifetime Value (LTV) model, which would then be used to give an estimate of how much revenue the brand can expect to get from a subscriber.
So, how would the Washington Post assess the value of the mass subscription loss? Well, they would look at all the predicted LTV figures for every individual who unsubscribed, add them all up, and provide an estimate back to the leadership team.
But if we can’t wait, and wanted to do something very quickly, we could make some assumptions and do quick math to get a feel for the scale of this decision.
- Average subscription length: We don’t have access to any concrete figures, but streaming services see 50–60 months, so let's use this as a proxy
- Minimum cost: All-access Digital is $120 per year (we’ll ignore the first-year sign-up incentive, as this has essentially already occurred for this user base and is not relevant to our estimates)
- Max cost: Premium Digital is $17 per month
- Subscriptions lost: 250,000
So, some very quick calculations in Excel show that per year, if all subscribers stay for at least 1 year and pay the minimum, full price, the lack of endorsement has cost the Washington Post $30 million. If the majority of subscribers were paying closer to the maximum cost, this is closer to $51 million per year. Extrapolating this out to an average subscription length of 50 months, we see the potential cost of this decision rise to c. $125m of lost subscription fees (and this sticking with the minimum subscription costs).
But this is where it gets interesting. The cancellation of subscriptions has been very public, with many people posting about their cancellations on social media, often with screengrabs of their cancellation confirmation. Can anyone see what might cause the Washington Post and their data team’s model some unexpected problems?
Well done if you spotted the issue.
Through highlighting their cancellations on social media, people have inadvertently shared the Washington Post’s retention offer. Most people are aware that companies use various tactics to stop churn. Yes, there is a small proportion of people who actively try to game the system to get these offers. But I guarantee, not many people know the exact offer.
So, this could cause a large number of people to game a system they never consider gaming before and get nice low subscription offer. This may already be happening, which is an additional cost the Washington Post may not even be considering (i.e. they think their retention offer is working, when in fact it is the mass subscriptions that have caused people to seek the retention offer).
But it gets worse! Now they have to reconsider how the retention offer itself is determined. To calculate the ideal retention offer, WP’s data science team will have conducted a significant amount of A/B testing, generating extensive data to model and predict the best retention offer. This retention offer is based on the behavior of people before this cancellation surge and the public sharing of the retention offer — so the optimal retention offer their models predict can no longer be relied upon and may end up costing the Washington Post far more than anticipated.
So what should they do?
This isn’t quite so straightforward. They first need to perform a deeper analysis of the possible value of this impact. This will likely involve running a suite of scenarios that highlight the differing scale of impact based on the number of people who might now start looking for retention offers. They could remove the retention offer altogether and essentially start the testing and model-building process over again. The latter will carry a cost as well, but if it’s less than the potential impact, it could be worth it.
This could also have downstream impacts on the quality of journalism, marketing budgets, innovation investments, etc.
Overall, it’s unlikely the Washington Post considered this potential impact as a by-product of their lack of endorsement for a presidential candidate. Would they have done something different if they had known? Who knows, as endorsing the wrong candidate could have also lost them millions and potentially a lot more.
Stepping back from all this, the situation highlights that brand decisions can have unexpected ripple effects on customer behavior, which suggests brands need to carefully consider both the short- and long-term implications of such high-profile actions.
While not necessarily relevant to the case at hand, this is where having an experienced and vocal data leader as part of your leadership function pays dividends, as they will nearly always recommend running through some scenarios to assess the impact of a decision. It’s not saying you have to go with what these scenarios suggest, but at the very least they will make you question and justify your choice.
