US 12,737,387 B2Granted

Intelligent transaction analytics

AssigneeKPMG International Services Ltd.

InventorsSantosh Kumar Kulkarni, Michael Woodford, Kallol K Ghosh, Scott Asher

Life of this patent

Estimating the term…

Abstract

Systems and methods disclosed herein provide for classification of business transactions. A client device runs an application that receives identification of a dataset for use in a classification analysis, where the dataset including a plurality of transactions. The application then receives user inputs to create a new analysis and runs a classification analysis of the transactions using one or more of manual classification, rules-based classification, and a machine learning model.

Claims7

Dependent claims

A computer-implemented method for classification of business transactions, executed by a transaction classification application operating on a client device communicatively coupled to a server via a network, the method comprising:

receiving, by the application, identification of a dataset including a plurality of transactions including one or more tax-related data types;
validating and mapping, by the application, the dataset to one or more fields using intelligent tagging techniques or templates stored in a database;
creating an analysis by receiving, via the application, an identification of an existing template stored in the database for use in the analysis and/or receiving, via the application, details of a new template for use in the analysis, the new template being stored in the database for subsequent analyses, and wherein the template specifies a field mapping schema and one or more rule priority levels, and wherein validating and mapping comprises normalizing the dataset into a canonical transaction table according to the field mapping schema;
receiving, via the application, a user selection of an AI model from among a plurality of pre-trained models stored in the database;
loading parameters of the selected AI model and generating, for each transaction, an AI-based classification label and confidence value;
classifying the transactions by applying:
a rules engine configured to apply classification rules and resolve rule conflicts;
manual user inputs provided via an interactive interface of the application;
and
the selected AI model trained on anonymized and synthetic transaction data, the model being updated based on changes in tax laws or business structures;
managing contras by:
identifying transaction pairs that offset each other as contras and presenting the identified contras to a user via the client device for review;
receiving, via the application, a user selection of a first transaction within a contra pair to retain for the analysis and discarding a second transaction within the contra pair;
and
storing a contra signature for each reviewed contra pair and, for subsequently received datasets, matching transactions to stored contra signatures and automatically flagging matched transactions for exclusion from at least one aggregate calculation;
and
exporting the classified transactions in a machine-readable format for storage in the database or for downstream tax analysis, wherein the selected AI model is trained to distinguish between different data types and classification types based on anonymized and synthetic transaction data.

14 more elements

A distributed computing system for intelligent classification of business transactions, comprising:

one or more servers configured with a transaction classification application and a database storing a plurality of artificial intelligence (AI) models, each AI model being purpose-built for a type of analysis and trained using anonymized and synthetic transaction data;
and
one or more client devices communicatively coupled to the servers via a secure network, the client devices configured to provide user inputs and display classification results through an interactive graphical interface, wherein the one or more servers are further configured to:
validate and map imported transaction datasets using intelligent tagging techniques and predefined templates stored in the database, wherein the intelligent tagging techniques recognize region-specific number formats, date formats, and column headers relative to pre-tagged columns for a selected data type;
receive, via the interactive graphical interface, user inputs to create an analysis by:
extracting keywords from selected fields of the dataset and generating one or more rules-based classifications based on the extracted keywords using classification buckets;
generating one or more advanced rules using a query builder interface that generates database query statements;
validating the query statements against a database schema corresponding to the predefined fields;
and
storing the created rules and compiled queries in the database with rule priority levels;
receive, via the interactive graphical interface, a user selection of an AI model from among the plurality of AI models;
execute concurrent classification of transactions using:
a rules engine configured to apply the stored classification rules and resolve rule conflicts;
manual user overrides received via the one or more client devices;
and
the selected AI model;
determine, for each transaction, an ultimate classified value by:
applying a priority order in which a manual user entry or override supersedes the rules engine based on the rule priority levels and the selected AI model;
and
tracking in the database the AI-based determination, the rules-based determination, and the manual determination for the transaction;
manage contras by:
detecting transaction pairs that offset each other, presenting identified contras to a user via the client device for review, and using the selected AI model to determine where contras are misidentified;
and
storing details of contra pairs such that subsequent analyses automatically recognize and fix corresponding contra pairs in subsequently received datasets;
and
export classification results in a structured machine-readable format by downloading results to the client device and/or pushing results to the database, wherein exported results include a breakdown indicating which transactions were classified according to which rules and wherein transaction data transmitted between the client devices and the servers is protected using secure network protocols and encryption.

23 more elements

A computer-implemented method executed by one or more server processors for classification of business transactions, the method comprising:

receiving, at the server, a dataset including a plurality of transactions of different tax-related data types from a client device via a secure network connection;
validating and mapping, by the server, data within the dataset to predefined fields using intelligent tagging algorithms and templates stored in a database accessible to the server, wherein the intelligent tagging algorithms recognize region-specific number formats, date formats, and column headers relative to pre-tagged columns for a selected data type;
receiving, at the server via an application operating on the client device, user inputs to create an analysis by:
extracting keywords from selected fields of the dataset and generating rules by assigning extracted keywords to classification buckets via the application;
generating advanced rules using a query builder interface that generates database query statements validated against the mapped predefined fields;
and
storing the rules and compiled queries in the database with rule priority levels;
receiving, at the server via the application, a user selection of an artificial intelligence (AI) model from among a predetermined set of AI models stored in the database, each AI model being purpose-built for a type of analysis and trained using anonymized and synthetic transaction data;
classifying, by the server, transactions in the dataset by:
applying the selected AI model hosted on the server, the model being trained on anonymized and synthetic transaction data and dynamically updated based on changes in tax laws retrieved through automated network queries to generate an AI-based classification determination and an associated confidence value for a transaction;
applying one or more stored rules via a rules engine to generate a rules-based classification determination for the transaction;
receiving, via the application, a manual user classification determination for the transaction;
and
resolving conflicts among the AI-based, rules-based, and manual determinations according to a priority order in which a manual user entry or override supersedes the rules-based determination based on the rule priority levels and the AI-based determination;
tracking, in the database, the AI-based determination, the rules-based determination, and the manual determination for each transaction and storing a final classification determination;
managing by:
detecting transaction pairs in which a debit and a credit affect the same parent account and result in a net-zero effect;
presenting identified contra pairs to a user via the client device for review;
using the selected AI model to analyze the contra pairs to determine where contras are misidentified and to correct uploaded or newly created data;
and
storing details of contra types and relationships in the database such that subsequent analyses automatically recognize and fix corresponding contra pairs in subsequently received datasets;
exporting classification results by downloading results to the client device via an interactive graphical interface and/or pushing results to the database in a machine-readable format, wherein exported results include a breakdown indicating which transactions were classified according to which rules;
and
pushing the classified dataset to the database for aggregation and further unsupervised training of the selected AI model, wherein transaction data transmitted between the client device and the server is protected using secure network protocols and encryption.

21 more elements

Dependencies are inferred from the claim text.

patentgrep

Loading patent…

Abstract

receiving, by the application, identification of a dataset including a plurality of transactions including one or more tax-related data types;
validating and mapping, by the application, the dataset to one or more fields using intelligent tagging techniques or templates stored in a database;
creating an analysis by receiving, via the application, an identification of an existing template stored in the database for use in the analysis and/or receiving, via the application, details of a new template for use in the analysis, the new template being stored in the database for subsequent analyses, and wherein the template specifies a field mapping schema and one or more rule priority levels, and wherein validating and mapping comprises normalizing the dataset into a canonical transaction table according to the field mapping schema;
receiving, via the application, a user selection of an AI model from among a plurality of pre-trained models stored in the database;
loading parameters of the selected AI model and generating, for each transaction, an AI-based classification label and confidence value;
classifying the transactions by applying:
a rules engine configured to apply classification rules and resolve rule conflicts;
manual user inputs provided via an interactive interface of the application;
and
the selected AI model trained on anonymized and synthetic transaction data, the model being updated based on changes in tax laws or business structures;
managing contras by:
identifying transaction pairs that offset each other as contras and presenting the identified contras to a user via the client device for review;
receiving, via the application, a user selection of a first transaction within a contra pair to retain for the analysis and discarding a second transaction within the contra pair;
and
storing a contra signature for each reviewed contra pair and, for subsequently received datasets, matching transactions to stored contra signatures and automatically flagging matched transactions for exclusion from at least one aggregate calculation;
and
exporting the classified transactions in a machine-readable format for storage in the database or for downstream tax analysis, wherein the selected AI model is trained to distinguish between different data types and classification types based on anonymized and synthetic transaction data.

14 more elements