Frequency of mailings sent: more often not always better!

Recently, I often hear opinions that “ The volume of sending letters is the key to success in email marketing!”. At their core, they mean that sending additional emails leads to more activity of subscribers, making more money, and, in general, is better (regardless of what “better” means to you).

Their arguments are simple:

1. My data shows that the more I receive / open / click on my newsletter, the more money I earn.
2. Since I cannot magically “conjure” new email addresses, so I should send mailings to those who already are more often.

After all, if you have an address database of 10,000 addresses and each time you send an email to them, you will receive 100 orders, then by sending an email to these addresses twice a month, and not one, you expect to receive 100 more orders , right? Money in your pocket! Why not go for it?

Of course, the reasoning is correct. But not so simple.

Maybe you can increase the frequency of your newsletters to increase sales, or maybe you can not. And here's why: The activity of the subscriber (opening, clicks) depends on the frequency of sending newsletters . The more you send, the less subscribers open your newsletters and click on the links in them. And that means there must be an equilibrium point at which a certain distribution frequency maximizes the activity of the subscriber (and, consequently, your sales).

Newsletter frequency and subscriber activity are negatively correlated

Ok, the first thing we need to show is that the frequency and activity are negatively correlated , i.e. the more often you send the newsletter, the less subscribers are active with each next campaign. For this study, I used the click rate in the newsletters instead of the level of openings, because This is a more honest indicator and it is more closely related to sales for online stores.

Mailchimp has a lot of users and we can study the effect of the sending frequency on the activity of subscribers, but for some users the sending frequency does not change. They, as a clockwork mechanism, send out mailings with the same frequency, so their data is useless in this study. Also, it would be wrong to combine data from different users, as they send newsletters to a variety of databases with different subscribers who have different expectations about the different content of the newsletters. We want to avoid such differences.
Therefore, in the end, I formed the procedure that I used in my research:

1. We’ll draw out all the users over the past 2 years, who sent often (from two times a month) at least one thousand subscribers. We will receive data on the level of clicks in each of their newsletters.
2. We calculate the sending frequency for the user on each of these mailings. Instead of the exact frequency value, I used a simple moving average for every 3 sendings to smooth out the “perceived” sending frequency (the way subscribers perceive it, because for them, the moment of receipt is the moment a message is found in the mailbox), which, I guess, is slightly behind reality.
3. We eliminate all deviations in the data by applying Tukey's fence for each user for a couple of sending frequencies - click levels. For example, an email newsletter on “Black Friday” can show a surge in the value of the click level, regardless of anything, so this newsletter should be excluded from the analysis.
4. We study data from those users who have significant differences in the frequency of sendings over their history (I used the following criterion: the inter-quartile range should be no less than the average frequency).
5. For each such user, we take a linear regression of the level of clicks on the frequency. We study regression with no statistically significant relationship in the data ( R-squared > 0.2, F-test p value <0.05).

Still reading?

The result of the study was amazing: in each case in which we obtained statistically significant results, a negative slope of the regression line was noted. In other words, for all users who have a good mailing history, the rule applies: the more often we send newsletters, the lower the level of clicks we will receive in each of them .

For example, here are the results for two users X and Y:

Send more, get less activity.

Another example of reduced activity in the newsletter.

[From translator]
Here in the original article there is an example with Obama's email marketing campaign, which is used by supporters of sending the maximum possible number of email newsletters. But, as it is rightly pointed out, if you need to maximize the result to some particular moment (in this case, this is an election), then everything is ok, but for periodic e-mails without a specific end date this is absolutely unacceptable. And that's why.
[/ From a translator]

The total activity of newsletters over a long period of time is a convex quadratic function, not an increasing line.

In light of the negative correlation between the frequency of sending and the activity of subscribers, we can find out how many times a month a client must send their newsletter using the technology that hotels, airlines, car rentals and others use in their pricing inventory.

The above graphs for users X and Y are very similar to demand curves in an economy where demand falls and price rises. Therefore, we can introduce such a concept as “ elasticity of activity versus frequency ”. Consider user Y as an example. For a separate newsletter we have:

уровень кликов = -0.08% * частота рассылки + 2.5%

Therefore, ceteris paribus, the total number of clicks that I plan to receive per month, based on my frequency of newsletters, can be modeled as:

всего кликов в месяц = размер базы * частота рассылки * (-0.8%*частота рассылки +2.5%)

We need to maximize the number of clicks per month. This function is convex up and quadratic . This means that taking the first derivative and equating it to 0, we can get the optimal distribution frequency.

2 * размер базы * -0.8% * частота отправки + размер базы*2.5% = 0

we get:

оптимальная частота отправки = -2.5%/(2*-0.8%) = 15 отправок в месяц

In the general case, the optimal sending frequency can be calculated from the curve of the dependence of subscribers' activity on the frequency of sending newsletters of the form

Y = A*X + B

as:

-B/2*A

For everyone who is frightened by mathematical calculations, we will simply formulate an idea. Slowly increase the frequency of sending newsletters, looking at the indicator of the total number of clicks collected per month from all newsletters. As soon as you slip through the optimal value of the frequency of mailings, this indicator will begin to decrease and each next “extra” mailing per month will decrease it more and more. Just roll back one value (distribution frequency per month) back and stop at that maximum - this will be the very ideal balance that you are looking for.

conclusions

Please note that the above calculations of the optimal sending frequency, there is one parameter that mysteriously is thrown out: the size of the address base. Thus, all statements “You can spam people more if the size of your address base is large” are absolutely inappropriate. No matter how big your address base is, there is a more subtle way to send mailings than clumsy “swotting”.

In the case of user Y, it turned out that the best way to send would be sending in a day. In the study, there were cases when the user sent every day, but the elasticity curve of the level of discoveries versus the frequency of sending recommended to moderate the pace of sending, based on the nature of their reader activity.

In our findings, the optimal frequency for sending a newsletter depends on the user. Her addiction is based on your audience and her expectations on the content of your newsletters. Therefore, do not pay attention to any calls to send more and more, but try to find a middle ground that you feel will be balanced between the individual activity of subscribers in each individual newsletter and the total activity of subscribers in the entire period for all mailings.

============================= The
material of the article from the Mailchimp.com blog has been translated and prepared by the email marketing service Pechkin-mail.ru .