Back to results
US 12,743,707 B2Granted
AI-enabled fraud analysis
AssigneeKPMG International Services Ltd.
InventorsJan Gelke, Jon Gray, Sebastian Stockle, Dennis Triepke
Life of this patent
Estimating the term…
Abstract
Methods, apparatus, and computer program products described herein provide for calculation of a score representative of the risk of fraud by a target entity. A set of financial ratios, lens model scores, risk area scores, and academic scores are all calculated in relation to the target entity and a group of peer entities. These calculated scores are all indicative of the risk of fraud by the target entity. The calculated scores and other market data related to the target entity are displayed to a user to evaluate the risk of fraud by the target entity.
Claims16
A computer-implemented method for calculating a score representative of the risk of fraud by a target entity, the method comprising, by one or more processors executing stored instructions:
- ingesting, from one or more data sources, financial statement data and regulatory reporting data associated with the target entity and with a plurality of peer entities;
- selecting the plurality of peer entities for comparison to the target entity by algorithmically identifying entities sharing at least one common industry classification and exceeding a minimum peer-count threshold, wherein selecting comprises:
- determining whether a sub-industry of the target entity has at least the minimum peer-count threshold of peer entities and, if so, selecting the peer entities from the sub-industry;
- otherwise, determining whether an industry of the target industry has at least the minimum peer-count threshold of peer entities and, if so, selecting the peer entities from the industry;
- and
- otherwise selecting the peer entities from an industry-group of the target entity;
- performing a missing data treatment on gaps in financial statement data and regulatory reporting data related to the target entity by imputing values based on peer-group decile position;
- performing an outlier data treatment on the financial statement data or regulatory reporting data related to the target entity by excluding ratio values outside a predefined interquartile range threshold;
- calculating a plurality of financial ratios by transforming the financial statement data into normalized ratio values for the target entity and the peer entities;
- calculating a decile rank of the target entity for each financial ratio compared to the peer entities to generate the peer-relative representations of the normalized ratio values;
- calculating a plurality of lens model scores indicative of fraud risk for the target entity by applying a trained machine learning model to peer-relative representations, wherein the peer-relative representations include the decile ranks of the target entity for the financial ratios;
- calculating a plurality of risk area scores indicative of fraud risk for the target entity based on weighted combinations of the decile ranks of the normalized ratio values grouped into predefined financial risk categories;
- calculating one or more academic scores indicative of fraud risk for the target entity;
- storing, in a non-transitory memory, the lens model scores, risk area scores, and academic scores as structured fraud risk indicators for subsequent analysis;
- and
- displaying the lens model scores, risk area scores, and academic scores to a user to evaluate the overall fraud risk for the target entity.
13 more elements
An apparatus for determining the risk of fraud by a target entity, comprising:
- one or more processors;
- and
- memory storing instructions that, when executed by the one or more processors, cause the apparatus to:
- ingest financial statement data and regulatory reporting data associated with the target entity and with a plurality of peer entities;
- select the plurality of peer entities for comparison to the target entity by algorithmically identifying entities sharing at least one common industry classification and exceeding a minimum peer-count threshold, wherein selecting comprises:
- determining whether a sub-industry of the target entity has at least the minimum peer-count threshold of peer entities and, if so, selecting the peer entities from the sub-industry;
- otherwise, determining whether an industry of the target entity has at least the minimum peer-count threshold of peer entities and, if so, selecting the peer entities from the industry;
- and
- otherwise selecting the peer entities from an industry-group of the target entity;
- perform a missing data treatment on gaps in financial statement data and regulatory reporting data related to the target entity by imputing values based on peer-group decile positions;
- perform an outlier data treatment on the financial data or regulatory reporting data related to the target entity by excluding ratio values outside a predefined interquartile range threshold;
- calculate a plurality of financial ratios by transforming the financial statement data into normalized ratio values for the target entity and the peer entities;
- calculate a decile rank of the target entity for each financial ratio compared to the peer entities to generate the peer-relative representations of the normalized ratio values;
- calculate a plurality of lens model scores indicative of fraud risk by applying a trained machine learning model to peer-relative representations, wherein the peer-relative representations include the decile ranks of the target entity for the financial ratios;
- calculate a plurality of risk area scores indicative of fraud risk based on weighted combinations of the decile ranks of the normalized ratio values grouped into predefined financial risk categories;
- calculate one or more academic scores indicative of fraud risk for the target entity;
- store the lens model scores, risk area scores, and academic scores in a non-transitory memory as structured fraud risk indicators;
- and
- display the lens model scores, risk area scores, and academic scores to a user.
16 more elements
A non-transitory machine-readable medium storing instructions executable by one or more processors, the instructions causing the one or more processors to:
- ingest financial statement data and regulatory reporting data associated with a target entity and with a plurality of peer entities;
- select the plurality of peer entities for comparison to the target entity by algorithmically identifying entities sharing at least one common industry classification and exceeding a minimum peer-count threshold, wherein the selection includes:
- determining whether a sub-industry of the target entity has at least the minimum peer-count threshold of peer entities and, if so, selecting the peer entities from the sub-industry;
- otherwise, determining whether an industry of the target entity has at least the minimum peer-count threshold of peer entities and, if so, selecting the peer entities from the industry;
- and
- otherwise selecting the peer entities from an industry-group of the target entity;
- perform a missing data treatment on gaps in financial statement data and regulatory reporting data related to the target entity by imputing values based on peer-group decile positions;
- perform an outlier data treatment on the financial data or regulatory reporting data related to the target entity by excluding ratio values outside a predefined interquartile range threshold;
- calculate a plurality of financial ratios by transforming the financial statement data into normalized ratio values for the target entity and the peer entities;
- calculate a decile rank of the target entity for each financial ratio compared to the peer entities to generate peer-relative representations of the normalized ratio values;
- calculate a plurality of lens model scores indicative of fraud risk by applying a trained machine learning model to peer-relative representations, wherein the peer-relative representations comprise the decile ranks of the target entity for the financial ratios;
- calculate a plurality of risk area scores indicative of fraud risk based on weighted combinations of the decile ranks of the normalized ratio values grouped into predefined financial risk categories;
- calculate one or more academic scores indicative of fraud risk for the target entity;
- store the lens model scores, risk area scores, and academic scores as structured fraud risk indicators;
- and
- display the lens model scores, risk area scores, and academic scores to a user.
13 more elements
Dependencies are inferred from the claim text.