Patent Issued for Dynamic ranking of recommendation pairings (USPTO 11715144): Salesforce Inc.

2023 AUG 21 (NewsRx) -- By a News Reporter-Staff News Editor at Insurance Daily News -- From Alexandria, Virginia, NewsRx journalists report that a patent by the inventors Casalaina, Marco (San Francisco, CA, US), Sanghavi, Tejas (San Francisco, CA, US), filed on May 24, 2019, was published online on August 1, 2023.

The patent’s assignee for patent number 11715144 is Salesforce Inc. (San Francisco, California, United States).

News editors obtained the following quote from the background information supplied by the inventors: “A challenge for businesses is figuring out which product, service, or actions are appropriate fits for new and existing customers. Consumers are bombarded with advertisements, offers and promotions and they are getting immune to these, especially because most of it used to be based on what the company wants to promote, rather than what the individual wants or needs. Broad brush rules, such as heuristic or intuitive rules, or customer segments, are not able to achieve the same performance as a personalized recommendation that is unique to each individual customer. Providing such irrelevant offers may frustrate customers leading to dissatisfaction or even attrition.

“Thus, there is a need for an improved systems and methods for determining appropriate products and services that are relevant for particular customers to drive sales of products or services.”

As a supplement to the background information on this patent, NewsRx correspondents also obtained the inventors’ summary information for this patent: “Some implementations of the disclosed systems, apparatus, methods and computer program products are configured for integrating a machine learning prediction platform with business rules to sort recommendations in near real-time based on propensity of the target object to accept. The predictive model is continuously updated with recommendation responses received by the system (closed-loop feedback). The described systems and methods also provide a solution to the “cold start” problem to provide recommendations for building a prediction model when data is lacking.

“As described in further detail below, such prediction platform may be seamlessly integrated with any type of application or service such as a Customer Relationship Management (CRM) Platform, a social networking system, or any other consumer or business software. While CRM platforms are discussed herein as an example of an application or service, the examples of applications or services described herein may be substituted for any suitable application or service such as those described above.

“Existing methods use heuristic and intuitive rules to established product fit recommendations. However, these broad-brush rules do not achieve the same performance as a personalized recommendation that is unique to each individual customer. Machine learning approaches for personalized recommendations work when substantial amounts of data are available. However, businesses with large catalogues of products increase the chances that they will offer customers irrelevant services/products because the machine learning (ML) models are poorly trained.

“By way of illustration, Raymond works as Inside Sales for a Sunrise Insurance, a major insurance provider that serves high-touch B2C (business-to-consumer) and smaller B2B (business-to-business) clients. He primarily focuses on existing accounts and increasing their share of wallet across their entire product line, for instance by offering Home or Renter’s insurance products to existing Auto insurance customers. However, simply offering additional insurance products to existing customers, who do not need such products or already have purchased similar products, may frustrate customers who may feel pressured by unwanted solicitations or annoyed by receiving irrelevant information.

“Using conventional techniques, offers may be provided in accordance with static rules. For example, the in-house marketing team at Sunrise Insurance may define account outreach strategies each quarter based on ad-hoc analyses about which types of customers are most likely to purchase specific products. These strategies also include various state specific rules based on which products can be sold in each state. However, the application of static rules ignores customer-specific characteristics.”

The claims supplied by the inventors are:

“1. A method comprising: determining via a processor whether a prediction model for an object definition is associated with sufficient training data to produce predictions at a designated accuracy rate, the prediction model predicting for a given object instance of the object definition a respective probability of acceptance for each of a plurality of recommendations; when it is determined that the prediction model is not associated with sufficient training data, transmitting via a communication interface for each of a first set of object instances a respective first message that includes a respective first one of the recommendations, the respective first one of the recommendations determined based on a static ranking rule, the static ranking rule applying one or more criteria to one or more object fields associated with the respective object instance; and updating the prediction model to include additional training data based on a plurality of responses, each response corresponding to a respective first one of the recommendations.

“2. The method as recited in claim 1, wherein the probability of acceptance is modeled as a propensity score.

“3. The method as recited in claim 1, the method further comprising: for each of a second set of object instances, transmitting via the communications interface a respective second message that includes a respective second one of the recommendations selected at random from a respective subset of the recommendations.

“4. The method as recited in claim 3, further comprising: updating the prediction model to include additional training data based on a plurality of responses, each response corresponding to a respective second one of the recommendations.

“5. The method as recited in claim 3, wherein each of the respective second one of the recommendations meets the one or more criteria associated with the respective object instance with which it is associated.

“6. The method as recited in claim 3, wherein a designated proportion specifies a relative size of the first set of object instances as compared to the second set of object instances.

“7. The method as recited in claim 1, the method further comprising: for each of a second set of object instances, transmitting via the communications interface a respective second message that includes a respective second one of the recommendations selected from a respective subset of the recommendations, the respective second one of the recommendations meeting a designated popularity threshold.

“8. The method as recited claim 1, wherein the respective object instance is a contact associated with an account, and wherein the respective first message is transmitted to a client device corresponding to the contact for display within an application executed on the client device.

“9. The method as recited in claim 1, wherein a notification message is transmitted indicating that the prediction model is available when the prediction model has been updated to include sufficient training data, wherein the notification message includes prediction accuracy metrics of the prediction model.

“10. The method as recited in claim 1, wherein each response of the plurality of responses includes at least one object field associated with the respective recommendation.

“11. A database system implemented using a server system, the database system configurable to cause: determining via a processor whether a prediction model for an object definition is associated with sufficient training data to produce predictions at a designated accuracy rate, the prediction model predicting for a given object instance of the object definition a respective probability of acceptance for each of a plurality of recommendations; when it is determined that the prediction model is not associated with sufficient training data, transmitting via a communication interface for each of a first set of object instances a respective first message that includes a respective first one of the recommendations, the respective first one of the recommendations determined based on a static ranking rule, the static ranking rule applying one or more criteria to one or more object fields associated with the respective object instance; and updating the prediction model to include additional training data based on a plurality of responses, each response corresponding to a respective first one of the recommendations.

“12. The database system as recited in claim 11, wherein the probability of acceptance is modeled as a propensity score.

“13. The database system as recited in claim 11, wherein the database system is further configurable to cause: for each of a second set of object instances, transmitting via the communications interface a respective second message that includes a respective second one of the recommendations selected at random from a respective subset of the recommendations.

“14. The database system as recited in claim 11, wherein the database system is further configurable to cause: updating the prediction model to include additional training data based on a plurality of responses, each response corresponding to a respective second one of the recommendations.

“15. The database system as recited in claim 11, wherein the respective object instance is a contact associated with an account, and wherein the respective first message is transmitted to a client device corresponding to the contact for display within an application executed on the client device.

“16. The database system as recited in claim 11, wherein a notification message is transmitted indicating that the prediction model is available when the prediction model has been updated to include sufficient training data, wherein the notification message includes prediction accuracy metrics of the prediction model.

“17. A computer program product comprising computer-readable program code capable of being executed by one or more processors when retrieved from a non-transitory computer-readable medium, the program code comprising instructions configurable to cause: determining via a processor whether a prediction model for an object definition is associated with sufficient training data to produce predictions at a designated accuracy rate, the prediction model predicting for a given object instance of the object definition a respective probability of acceptance for each of a plurality of recommendations; when it is determined that the prediction model is not associated with sufficient training data, transmitting via a communication interface for each of a first set of object instances a respective first message that includes a respective first one of the recommendations, the respective first one of the recommendations determined based on a static ranking rule, the static ranking rule applying one or more criteria to one or more object fields associated with the respective object instance; and updating the prediction model to include additional training data based on a plurality of responses, each response corresponding to a respective first one of the recommendations.

“18. The computer program product as recited in claim 17, the program code further comprising instructions configurable to cause: for each of a second set of object instances, transmitting via the communications interface a respective second message that includes a respective second one of the recommendations selected from a respective subset of the recommendations, the respective second one of the recommendations meeting a designated popularity threshold.

“19. The computer program product as recited in claim 17, wherein a notification message is transmitted indicating that the prediction model is available when the prediction model has been updated to include sufficient training data, wherein the notification message includes prediction accuracy metrics of the prediction model.

“20. The computer program product as recited in claim 17, wherein each response of the plurality of responses includes at least one object field associated with the respective recommendation.”

For additional information on this patent, see: Casalaina, Marco. Dynamic ranking of recommendation pairings. U.S. Patent Number 11715144, filed May 24, 2019, and published online on August 1, 2023. Patent URL (for desktop use only): https://ppubs.uspto.gov/pubwebapp/external.html?q=(11715144)&db=USPAT&type=ids

(Our reports deliver fact-based news of research and discoveries from around the world.)