EMAIL OPTIMIZATION FOR PREDICTED RECIPIENT BEHAVIOR: SUGGESTING A TIME AT WHICH A USER SHOULD SEND AN EMAIL
Techniques are described herein for predicting one or more behaviors by an email recipient and, more specifically, to machine learning techniques for predicting one or more behaviors of an email recipient, changing one or more components in the email to increase the likelihood of a behavior, and determining and/or scheduling an optimal time to send the email. Some advantages of the embodiments disclosed herein may include, without limitation, the ability to predict the behavior of the email recipient and suggest the characteristics of an email which will increase the likelihood of a positive behavior, such as a reading or responding to the email, visiting a website, calling a sales representative, or opening an email attachment.
1 . A system comprising:
a memory;
one or more processors coupled to the memory and configured to:
for sent each sent message in a plurality of sent messages:
determine a sent time for the sent message;
determine an opened time for the sent message;
train a model based, at least in part, on the sent time and the opened time for the sent message;
determine a desired opening time for a draft message;
determine, based on the model and the desired opening time:
a first predicted time at which the draft message should be sent to increase likelihood the draft message will be opened by a first recipient at the desired opening time; and
a second predicted time at which the draft message should be sent to increase likelihood the draft message will be opened by a second recipient at the desired opening time.
2 . The system of claim 1 , wherein the draft message identifies the first recipient and the second recipient.
3 . The system of claim 1 , wherein the first predicted time is different than the second predicted time.
4 . The system of claim 1 , wherein the one or more processors are further configured to:
receive, at a server computer, a first instruction to send the draft message at the first predicted time to the first recipient; and
in response to the instruction, send the draft message, from the server computer, at the first predicted time to the first recipient.
5 . The system of claim 4 , wherein the one or more processors are further configured to:
receive, at the server computer, a second instruction to send the draft message at the second predicted time to the second recipient; and
in response to the instruction, send the draft message, from the server computer, at the second predicted time to the second recipient.
6 . The system of claim 5 , wherein the first instruction and the second instruction are included in a request received at the server computer together.
7 . The system of claim 1 , wherein the one or more processors are further configured to:
for each sent message in the plurality of sent messages:
determine a feature associated with the sent message;
wherein training the model based, at least in part, on the feature associated with the sent message;
determine a particular feature based on the draft message;
wherein determining the first predicted time at which the draft message should be sent is further based on the particular feature associated with the draft message; and
wherein determining the second predicted time at which the draft message should be sent is further based on the particular feature associated with the draft message.
8 . The system of claim 7 , wherein:
the feature associated with each sent message in the plurality of sent messages is based on content in the sent message; and
the particular feature is based on text in the draft message.
9 . The system of claim 7 , wherein:
the feature associated with each sent message in the plurality of sent messages identifies an intended recipient and the feature is associated with the intended recipient; and
the draft message identifies a particular recipient and the particular feature is associated with the particular recipient.
10 . The system of claim 1 , wherein:
the model is a machine learning model; and
the one or more processors are further configured to generate the machine learning model based, at least in part, on a multi-layer perceptron.
11 . The system of claim 1 , wherein the predicted time is limited to be within a particular window of time.
12 . The system of claim 11 , wherein the particular window is 40 hours from when determining a user requests the predicted time.
13 . The system of claim 1 , wherein the one or more processors are configured to round the desired opening time to the nearest half-hour, wherein:
determining the sent time for each message in the plurality of messages comprises rounding the sent time to the nearest half-hour;
determining the opened time for each message in the plurality of messages comprises rounding the opened time to the nearest half-hour; and
determining the predicted time comprises rounding the predicted time to the nearest half-hour.
14 . A method comprising:
for sent each sent message in a plurality of sent messages:
determining a sent time for the sent message;
determining an opened time for the sent message;
training a model based, at least in part, on the sent time and the opened time for the sent message;
determining a desired opening time for a draft message;
determining, based on the model and the desired opening time:
a first predicted time at which the draft message should be sent to increase likelihood the draft message will be opened by a first recipient at the desired opening time; and
a second predicted time at which the draft message should be sent to increase likelihood the draft message will be opened by a second recipient at the desired opening time.
15 . The method of claim 14 , wherein the draft message identifies the first recipient and the second recipient.
16 . The method of claim 14 , wherein the first predicted time is different than the second predicted time.
17 . The method of claim 14 further comprising:
receiving, at a server computer, a first instruction to send the draft message at the first predicted time to the first recipient; and
in response to the instruction, sending the draft message, from the server computer, at the first predicted time to the first recipient.
18 . The method of claim 17 further comprising:
receiving, at the server computer, a second instruction to send the draft message at the second predicted time to the second recipient; and
in response to the instruction, sending the draft message, from the server computer, at the second predicted time to the second recipient.
19 . The method of claim 18 , wherein the first instruction and the second instruction are included in a request received at the server computer together.
20 . The method of claim 14 further comprising:
for each sent message in the plurality of sent messages:
determining a feature associated with the sent message;
wherein training the model based, at least in part, on the feature associated with the sent message;
determining a particular feature based on the draft message;
wherein determining the first predicted time at which the draft message should be sent is further based on the particular feature associated with the draft message; and
wherein determining the second predicted time at which the draft message should be sent is further based on the particular feature associated with the draft message.
21 . The method of claim 20 , wherein:
the feature associated with each sent message in the plurality of sent messages is based on content in the sent message; and
the particular feature is based on text in the draft message.
22 . The method of claim 20 , wherein:
the feature associated with each sent message in the plurality of sent messages identifies an intended recipient and the feature is associated with the intended recipient; and
the draft message identifies a particular recipient and the particular feature is associated with the particular recipient.
23 . The method of claim 14 , wherein:
the model is a machine learning model; and
the method further comprises generating the machine learning model based, at least in part, on a multi-layer perceptron.
24 . The method of claim 14 , wherein the predicted time is limited to be within a particular window of time.
25 . The method of claim 24 , wherein the particular window is 40 hours from when determining a user requests the predicted time.
26 . The method of claim 14 , wherein:
the method further comprising rounding the desired opening time to the nearest half-hour;
determining the sent time for each message in the plurality of messages comprises rounding the sent time to the nearest half-hour;
determining the opened time for each message in the plurality of messages comprises rounding the opened time to the nearest half-hour; and
determining the predicted time comprises rounding the predicted time to the nearest half-hour.