Patent Application
20230095858
Most user-based ecommerce applications (apps) are already interfaced to product recommendation engines (services) to produce and present recommendations to user. However, using the recommended products effectively within the ecommerce apps to achieve an optimal conversion rate (purchase rate of recommended products) is challenging.
Furthermore, many retailers provide checkout terminal interfaces that use recommendation engines in efforts to recommend additional products in the store as the customer checkouts at the terminal.
Current product recommendation engines are missing out on an important aspect associated with successful conversions. This aspect is a customers “Purpose of Purchase (POP).” Current solutions rely on the “lowest common denominator” of customers that are shopping to influence recommendations without analyzing what actually brought the shoppers to the store/website to make a purchase in the first place. Often this lowest common denominator is associated with a given customers transaction history relative to other customers with similar transaction histories that were successfully converted during a purchase.
Personalization is one of the fastest growing areas in retail. Achieving high conversion rates on product recommendations provided through recommendation engines are critical for retailers who wish to sustain the growing competition faced in the e-commerce and in-store industry segments.
Unfortunately, some existing recommendation engines do not even attempt to factor in a customer's POP, which intuitively would be a most significant factor in the customer's reason for visiting an e-commerce site or visiting a store in person to make a purchase. Some in the industry are under a false impression that resolving a customer's POP cannot be achieved while others in the industry generically attempt to guess at the customer's POP, which can be more detrimental to a successful conversion than ignoring the POP altogether because POPs can vary widely.
Overall, existing recommendation engines are suboptimal because they fail to account for or they overly generalize their customers' POPs, resulting in lost opportunities for the retailers.
In various embodiments, methods and a system for assigning customers to Purposes of Purchase (POPs) and to boost a recommendation engine's conversion rate on recommendations made to the customers are presented.
According to an aspect, a method for providing a customer's POP as a factor in a recommendation provided by a recommendation engine/service, is presented. A Purpose of Purchase (POP) question is integrated into a transaction session of a customer. A POP profile is assigned to the customer of the transaction session when an answer is provided by the customer to the POP question during the transaction session. The POP profile is predicted for the customer when the customer fails to provide the answer to the POP question based on transaction data for the transaction session and customer data for the customer. The POP profile for the customer is provided to a recommendation service for use as a factor in providing the customer with a recommendation during the transaction session.
Furthermore, the various components (that are identified in
As will be discussed in the various embodiments that follow, the teachings provide techniques by which customers' POPs are collected directly from the customers following purchases made by the customers. A small survey question is integrated into e-commerce applications or in-store transaction manager interfaces. The POPs are then associated with customer transaction histories and other data surrounding the purchases (such as customer loyalty data) and used to train a machine-learning model (MLM) for purposes of predicting a subsequent customer's POP at the time a customer is engaged with an e-commerce application or performing a transaction in-store at a transaction terminal. The trained MLM classifies a given customer in a given situation/context (e-commerce session or in-store purchase) into one of the known POPs (a POP profile). When the customer answers the survey question, the MLM uses the answer to assign the customer to a POP profile. When the customer fails to answer the survey question, the MLM uses the customer's historical data to predict the most likely POP profile for the customer. A customer's assigned POP profile is provided in real-time to the recommendation engine associated with the e-commerce site or associated with the in-store transaction terminal to use as a factor of import in making a product recommendation to the customer. Following the customer transaction if a conversion was made or was not made on the recommended product, the assigned customer POP profile, an indication as to whether the recommended product was or was not purchased (was or was not converted), and the original customer purchase transaction details are saved for subsequent re-training of the MLM to improve the MLM's accuracy in assigning POP profiled to subsequent customers engaged with an e-commerce site or engaged in a transaction with an in-store terminal.
System 100 comprises a cloud/server 110, a plurality of recommendation servers 120, a plurality of ecommerce servers 130, and a plurality of in-store transaction terminals 140.
Cloud/Server 110 comprises at least one processor 111 and a non-transitory computer-readable storage medium 112. Medium 112 comprises executable instructions for one or more machine-learning models (algorithms) 113, a POP profiler 114, and a surveyor 115. The executable instructions when executed by processor 111 from the medium 112 cause processor 111 to perform operations discussed herein and below with model(s) 113, profiler 114, and surveyor 115.
Each recommendation server 120 comprises a processor 121 and a non-transitory computer-readable storage medium 122. Medium 122 comprises executable instructions for a recommendation service (engine) 123 and an Application Programming Interface (API) 124. The executable instructions when executed by processor 121 from medium 122 cause processor 121 to perform operations discussed herein and below with respect to engine 123 and API 124.
Each ecommerce server 130 comprises a processor 131 and a non-transitory computer-readable storage medium 132. Medium 132 comprises executable instructions for an ecommerce app 133 and an API 134. The executable instructions when executed by processor 131 from medium 132 cause processor 131 to perform operations discussed herein and below with respect to ecommerce app 133 and API 134.
Each transaction terminal 140 comprises a processor 141 and a non-transitory computer-readable storage medium 142. Medium 142 comprises executable instructions for a transaction manager 143 and an API 144. The executable instructions when executed by processor 141 from medium 142 cause processor 141 to perform operations discussed herein and below with respect to transaction manager 143 and API 144.
Initially, workflows associated with ecommerce app 133 and transaction interfaces of transaction manager 144 are modified to make a call via API 134 and API 144, respectively to receive and present customers with a question regarding the POP for an item/product purchased by each customer via the ecommerce app 133 and/or the transaction manager 143. The surveyor 115 asks one or a few simple questions of the customers to determine the POPs based on the transaction details for the customers' purchases. For example, depending upon the item code purchased by a customer, the retailer identifier, the store identifier, purchase location, etc. a multiple-choice question is asked, “did you purchase X today because: A) weekly purchase was time to purchase; B) party or event and was in need of X; C) stocking up on X; D) impulse buy; E) gift for someone else; F) vehicle needed fuel (when X is fuel); G) based on an advertisement online, on tv, online; H) recommended by someone; etc. A second question may simply be an input field that says in as few words as possible what brought you to the e-commerce site or the store today.
The options to present in the multiple-choice question can be contextually selected by surveyor 115 based on the retailer associated with the e-commerce app 133 or terminal 140, an item code associated with one or more items that were purchased by the customer, a location of the store, a time of day, a day of week, a calendar date, etc.
System 100 uses surveyor 115 to collect a predefined number of transactions with answers provided back by the customers through ecommerce app 133 and transaction manager 143. The answers, corresponding transaction details (including transaction location (online or instore and if instore specific geographical location and item information for item(s) purchased), and customer transaction histories for the customers, customer loyalty data for the customers. MLM 113 is then trained on the transaction details, transaction histories, and loyalty data to produce the corresponding answers as output associated with a POP profile.
In an embodiment, before the MLM 113 is trained, the answers are mapped to a predefined set of POP profiles, such that there is not a direct 1-1 mapping between each answer and each POP profile. The MLM 113 is then trained to produce the POP profile based on the answers.
When a customer fails to provide an answer to a survey from surveyor 115, MLM 113 is trained to predict a POP profile based on the transaction details, customer transaction history, and customer loyalty data. So, when a customer provides an answer, MLM 113 uses or maps the answer to a POP profile and when the customer does not provide an answer, MLM 113 predicts a POP profile based on the information provided for the transaction.
The POP profiles may further be mapped to simple selections associated with the survey choices provided to customers to answer following transactions and fed back to the MLM 113 during training to correct predicted POP profiles of the MLM 113.
Once MLM 113 is trained to a desired level of accuracy in predicting POPs of customers, a customer that answers the survey question by surveyor 115 is assigned the corresponding POP profile by profiler 114. A customer that provides no answer to the survey question results in profiler 114 providing the customer transaction history, loyalty data, and transaction details to MLM 113 as input and receiving as output a predicted POP profile classification for the customer and the transaction.
Profiler 114 provides the POP profile of the customer (ether provided by the customer or predicted by MLM 113) to recommendation service 123 through API 124. Recommendation service 134 uses the customer-assigned POP profile as a factor in making a product recommendation to the customer via ecommerce app 133 of via transaction interface of transaction manager 143. A result of the recommendation (converted/purchased or not purchased) is provided back to profiler 114 to re-train MLM 113 for purposes of achieving a higher conversion rate based on an assigned POP profile to a given customer in a given transaction.
Existing recommendation engines do not consider as a factor in product recommendation, the reason the customer is in fact making a purchase with a retailer on a retailer e-commerce site or at a retailer store. This reason is probably the most significant reason the retailer sold an item and yet it is not utilized in conventional recommendation services. System 100 corrects that deficiency by obtaining actual POPs of customers and training a model 113 to predict a POP profile for customers who do not identify their POPs. The assigned POP profile is then provided to an enhanced recommendation service 123 that uses the POP profile classification of a customer as an additional and a substantial factor in product recommendations to the customer. Conversions or non-conversions are noted, and the corresponding transaction information and customer information used to retrain the MLM 113 to achieve more accurate POP profile predictions and increased conversion rates by the recommendation service 123.
In an embodiment, the recommendation service 123 and API 124 may be subsumed into either cloud/server 110 and/or ecommerce server 130.
In an embodiment, MLM 113, surveyor 115, and profiler 114 are subsumed into either ecommerce server 130 and/or recommendation server 120.
In an embodiment, transaction terminals 140 can be Point-Of-Sale (POS) terminals operated by a cashier that checks a customer out at a store or a Self-Service Terminal (SST) that a customer operates to perform a self-checkout. Any SST can be an Automated Teller Machine (ATM), a Self-Checkout (SCO) terminal, and/or a kiosk.
The above-referenced embodiments and other embodiments are now discussed with reference to
In an embodiment, the device that executes the POP classifier is cloud 110. In an embodiment, the device that executes POP classifier is server 110.
In an embodiment, the POP classifier is all of, or some combination of model(s) 113, profiler 114, surveyor 115, API 124, and/or optimizer API 144.
At 210, POP classifier integrates a POP question into a transaction session of a customer.
In an embodiment, at 211, the POP classifier receives a request initiated through an interface of the transaction session for the POP question.
In an embodiment of 211 and at 212, the POP classifier receives the request from the interface via an API initiated within a workflow of the interface during the transaction session.
In an embodiment of 212 and at 213, the POP classifier receives the request via a device associated with an on-line transaction of the customer with an e-commerce service.
In an embodiment of 212 and at 214, the POP classifier receives the request via a transaction terminal associated with an in-store transaction of the customer with a retail store.
At 220, the POP classifier assigns a POP profile to the customer of the transaction session when an answer is provided by the customer to the POP question during the transaction session.
In an embodiment, at 221, the POP classifier maps the answer to one of a plurality of POP profiles based on the answer.
In an embodiment of 221 and at 222, the POP classifier trains a machine-learning model 113 based on the answer, the transaction data for the transaction session, and the customer data for the customer to subsequently predict a new POP profile without an answer being provided by a subsequent customer during a subsequent transaction session.
At 230, the POP classifier predicts the POP profile for the customer when the customer fails to provide the answer to the POP question based on transaction data for the transaction session and customer data for the customer.
In an embodiment, at 231, the POP classifier obtains the customer data as transaction history data associated with the customer and loyalty data associated with the customer.
In an embodiment of 231 and at 232, the POP classifier provides the transaction data, the transaction history, and the loyalty data to a trained machine-learning model 113 (MLM 113) and receives as output from the MLM 113 the POP profile.
At 240, the POP classifier provides the POP profile for the customer to a recommendation service/engine 123 for use as a factor in providing the customer with a product recommendation during the transaction session.
In an embodiment, at 250, the POP classifier is processed as an intermediary between a transaction interface associated with the transaction session and the recommendation service/engine 123.
In an embodiment of 250 and at 251, the POP classifier uses or processes an API provided to the transaction interface and the recommendation service/engine 123 for processing the intermediary interface.
In an embodiment, at 260, the POP classifier is provided as and processed as a Software-as-a-Service (SaaS) to a retailer's transaction system (ecommerce server 130 and/or transaction terminals 140) and the recommendation service/engine 123.
In an embodiment, the device that executes the real-time recommendation service POP profile predictor is cloud 110. In an embodiment, the device that executes the real-time recommendation service POP profile predictor is server 110.
In an embodiment, the real-time recommendation service POP profile predictor is all of, or some combination of model(s) 113, profiler 114, surveyor 115, API 124, optimizer API 144, and/or method 200.
The real-time recommendation service POP profile predictor presents another, and in some ways, enhanced processing perspective from that which was discussed above for method 200.
At 310, the real-time recommendation service POP profile predictor receives requests for POP profiles through transaction interfaces (133 and/or 143) associated with customers engaged in transaction sessions via the transaction interfaces.
At 320, the real-time recommendation service POP profile predictor provides POP questions back to the customers for presentation within the transaction interfaces to the customers.
In an embodiment, at 321, the real-time recommendation service POP profile predictor provides available answers as multiple-choice selections in each of the POP questions.
In an embodiment of 321 and at 322, the real-time recommendation service POP profile predictor maps each of or a unique combination of each multiple-choice selection to a particular POP profile.
At 330, the real-time recommendation service POP profile predictor receives answers back from first customers.
At 340, the real-time recommendation service POP profile predictor assigns the POP profiles to the first customers based on the answers.
At 350, the real-time recommendation service POP profile predictor trains a MLM 113 on transaction data and customer data associated with the first customers to predict the POP profiles for second customers that fail to provide any answers back to the POP questions.
At 360, the real-time recommendation service POP profile predictor obtains the POP profiles for the second customers based on second transaction data and second customer data provided as input to the MLM 113.
At 370, the real-time recommendation service POP profile predictor provides the POP profiles for the first customers and the second customers to a recommendation engine 123 associated with the transaction sessions for determining recommendations provided to the first customers and the second customers during the transaction sessions.
In an embodiment, at 380, the real-time recommendation service POP profile predictor is processed as an intermediary between the transaction interfaces and the recommendation engine 123 during the transaction sessions.
In an embodiment of 380 and at 381, the real-time recommendation service POP profile predictor processes an API for interaction with the transaction interfaces and the recommendation engine 123 during the transaction sessions.
In an embodiment, at 390, the real-time recommendation service POP profile predictor is provided and processed as a SaaS to the transaction interfaces and the recommendation engine 123.
It should be appreciated that where software is described in a particular form (such as a component or module) this is merely to aid understanding and is not intended to limit how software that implements those functions may be architected or structured. For example, modules are illustrated as separate modules, but may be implemented as homogenous code, as individual components, some, but not all of these modules may be combined, or the functions may be implemented in software structured in any other convenient manner.
Furthermore, although the software modules are illustrated as executing on one piece of hardware, the software may be distributed over multiple processors or in any other convenient manner.
The above description is illustrative, and not restrictive. Many other embodiments will be apparent to those of skill in the art upon reviewing the above description. The scope of embodiments should therefore be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled.
In the foregoing description of the embodiments, various features are grouped together in a single embodiment for the purpose of streamlining the disclosure. This method of disclosure is not to be interpreted as reflecting that the claimed embodiments have more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive subject matter lies in less than all features of a single disclosed embodiment. Thus, the following claims are hereby incorporated into the Description of the Embodiments, with each claim standing on its own as a separate exemplary embodiment.
