Send time optimisation

The send time optimisation feature in the broadcast builder sends all the emails in the selected variant at different time slots.

Based on the (human) open rate of the emails delivered in the various time slots, optimise send-time iteratively optimises your send times so that emails are sent at the time that maximises the chance of engagement.

This algorithm doesn't assume that the recipient opens the email immediately after delivery - for example some contacts may have habits that lead to the optimal send time to be some time before they wake up and see their email notifications.

The send time model

Ometria trains a send time model for each contact that learns from their email deliveries and opens.

The model does not count pre-fetched opens. 

The aim is to predict the best time to send an email that will result in the email being opened by the recipient. 

The model uses the following information to calculate optimal send-time:

• The time emails are sent from Ometria
• Whether or not the email was opened.

There are different models trained for campaigns sent on:

• Weekends
• Fridays
• All other weekdays (Monday - Thursday)

Each day is divided into 12 two-hour slots, and for each slot a probability of success is calculated for each contact (based on their past behaviour) using a beta probability distribution.

If Ometria doesn't have any information about a contact (i.e. they’re a new lead), a model trained with account-wide data is used.

Ometria uses multi-armed bandit to balance "exploration-exploitaiton", meaning that there's an element of randomness which makes the highest performing slot not necessarily the one used for each contact, so that we get more data for potentially unexplored periods of the day. 

Training the model

The send time model learns on a contact-by-contact basis, and requires at least 10 sends to a contact in order to have fully accurate data. 

If you aren’t targeting the same audience each time, this will impact the number of total sends needed.

Limiting the range of time will also reduce the efficacy of this feature, as the model will have less information to learn from.

Impact on campaign sending

Campaigns using optimise send-time send in batches, with the appropriate number of batches being generated depending on the send time limit you enter:

A batch is created for every slot we have contacts for - each contact is allocated to a slot with the highest score for them. 

Contacts who don't have enough data (delivered or opened emails) for the algorithm to confidently choose a time slot are allocated an account preferred slot (based on account averages) within the specified time period.

The confidence increases based on the number of email deliveries and the number of explored time slots.

​​Skewed send time data

If you have always sent emails at specific time frames, this can potentially skew the optimisation model.

For example, if emails have historically only been sent at 8am on weekdays and 1pm on weekends - for contacts that opened many of their emails the model will assume that 8am on weekdays and 1pm on weekends is the optimal time.

This is not necessarily a problem - if those contacts are actually opening many of their emails, then sending at those times is optimal for those customers.

For contacts who rarely open their emails, the model will work out that 8am on weekdays and 1pm on weekends is a bad time to send the email and in this case it will look at other time slots.