Automated upsells in customer conversations

1. A computer-implemented method for suggesting a product upsell to a customer service representative during a conversation between the customer service representative and a first customer, the method comprising:

receiving information about the first customer from a data store of customer information;

computing customer features from the information about the first customer;

receiving information about a plurality of messages between the first customer and the customer service representative;

computing conversation features from the information about the plurality of messages;

determining to suggest an upsell to the customer service representative by processing the conversation features with a first classifier;

computing, in response to determining to suggest the upsell, a first plurality of scores by processing the customer features with a second classifier, wherein each score of the first plurality of scores corresponds to a respective product of a plurality of products;

determining customer data indicating products provided to other customers;

performing collaborative filtering using the customer features and the customer data to generate a second plurality of scores, wherein each score of the second plurality of scores corresponds to a respective product of the plurality of products;

computing a third plurality of scores by interpolating the first plurality of scores and the second plurality of scores, wherein each score of the third plurality of scores corresponds to a respective product of the plurality of products;

using the third plurality of scores, selecting a first product from the plurality of products based at least in part on a score corresponding to the first product; and

causing presentation, to the customer service representative, of a suggestion to upsell the first product to the first customer.

2. The computer-implemented method of claim 1 , comprising:

receiving information about a subsequent message between the first customer and the customer service representative;

computing third conversation features using the information about the subsequent message; and

determining to not suggest an upsell to the customer service representative by processing the third conversation features with the first classifier.

3. The computer-implemented method of claim 1 , wherein the conversation features comprise context features.

4. The computer-implemented method of claim 1 , wherein the customer features comprise information about products previously provided to the first customer.

5. The computer-implemented method of claim 1 , wherein determining to suggest the upsell comprises:

computing a second score using the first classifier, the conversation features and the customer features; and

determining to suggest the upsell based on the second score meeting a threshold.

6. The computer-implemented method of claim 1 , further comprising:

comparing each score of the first plurality of scores to a threshold; and

wherein the first product is a product with a highest score.

7. The computer-implemented method of claim 1 , wherein determining to suggest the upsell comprises processing the customer features with the first classifier.

8. A system for suggesting a product upsell to a customer service representative during a conversation between the customer service representative and a first customer, the system comprising:

at least one server computer comprising at least one processor and at least one memory, the at least one server computer configured to:

receive information about the first customer from a data store of customer information;

compute customer features from the information about the first customer;

receive information about a message between the first customer and the customer service representative;

compute conversation features from the information about the message;

determine to suggest an upsell to the customer service representative by processing the conversation features with a first classifier;

compute, in response to the determination to suggest the upsell, a first plurality of scores by processing the customer features using a second classifier, wherein each score of the first plurality of scores corresponds to a respective product of a plurality of products;

determine customer data indicating products provided to other customers;

perform collaborative filtering using the customer features and the customer data to generate a second plurality of scores, wherein each score of the second plurality of scores corresponds to a respective product of the plurality of products;

compute a third plurality of scores by interpolating the first plurality of scores and the second plurality of scores, wherein each score of the third plurality of scores corresponds to a respective product of the plurality of products;

using the third plurality of scores, select a first product from the plurality of products based at least in part on a score corresponding to the first product; and

cause presentation, to the customer service representative, of a suggestion to upsell the first product to the first customer.

9. The system of claim 8 , wherein the conversation features comprise sentiment features.

10. The system of claim 8 , wherein the at least one server computer is configured to:

receive information about a second subsequent message between the first customer and the customer service representative;

compute third conversation features using the information about the second subsequent message;

determine to suggest an upsell to the customer service representative by processing the third conversation features with the first classifier;

determine a second product from the plurality of products by processing the customer features with the second classifier; and

cause presentation, to the customer service representative, of a suggestion to upsell the second product to the first customer.

11. The system of claim 8 , wherein the at least one server computer is configured to:

determine to suggest an upsell to the customer service representative by generating a boolean value using the first classifier, wherein the boolean value indicates whether to suggest an upsell to the customer service representative.

12. The system of claim 8 , wherein the at least one server computer is configured to:

receive an indication from the customer service representative to upsell the first product to the first customer; and

transmit a message, to a device used by the first customer, offering the first product to the first customer, wherein the message was not generated by the customer service representative.

13. The system of claim 8 , wherein the conversation features comprise context features, topic features, and sentiment features.

14. The system of claim 8 , wherein the at least one server computer is configured to select the first product using the conversation features.

15. One or more non-transitory computer-readable media comprising computer executable instructions that, when executed, cause at least one processor to perform actions comprising:

receiving information about a first customer from a data store of customer information;

computing customer features from the information about the first customer;

receiving information about a message between the first customer and a customer service representative;

computing conversation features from the information about the message;

determining to suggest an upsell to the customer service representative by processing the conversation features with a first classifier;

computing, in response to determining to suggest the upsell, a first plurality of scores by processing the customer features with a second classifier, wherein each score of the first plurality of scores corresponds to a respective product of a plurality of products;

determining customer data indicating products provided to other customers;

performing collaborative filtering using the customer features and the customer data to generate a second plurality of scores, wherein each score of the second plurality of scores corresponds to a respective product of the plurality of products;

computing a third plurality of scores by interpolating the first plurality of scores and the second plurality of scores, wherein each score of the third plurality of scores corresponds to a respective product of the plurality of products;

using the third plurality of scores, selecting a first product from the plurality of products based at least in part on a score corresponding to the first product; and

causing presentation, to the customer service representative, of a suggestion to upsell the first product to the first customer.

16. The one or more non-transitory computer-readable media of claim 15 , wherein the conversation features comprise topic features.

17. The one or more non-transitory computer-readable media of claim 15 , wherein the actions comprise:

selecting a second product from the plurality of products by processing the customer features with the second classifier; and

causing presentation, to the customer service representative, of a suggestion to upsell the second product to the first customer.

18. The one or more non-transitory computer-readable media of claim 15 , wherein the first classifier generates a boolean value indicating whether to suggest an upsell to the customer service representative.

19. The one or more non-transitory computer-readable media of claim 15 , wherein the first classifier comprises a support vector machine and the second classifier comprises a multinomial logistic regression classifier.