Resistant.AI Transaction Detection Engine - Representation Learning Module

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F07825439%3A_____%2F22%3AN0000002" target="_blank" >RIV/07825439:_____/22:N0000002 - isvavai.cz</a>

Výsledek na webu

<a href="https://resistant.ai/use-cases/bnpl/" target="_blank" >https://resistant.ai/use-cases/bnpl/</a>

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Resistant.AI Transaction Detection Engine - Representation Learning Module

Popis výsledku v původním jazyce

Representation learning module for Resistant.AI Transaction detection engine that focuses on detection of fraudulent attempts to create e-shop orders using Buy Now Pay Later (BNPL) service. The representation learning module is focusing on universal representation of basket items of e-shop consumer's order that is easily transferable between individual countries. The representation is able to capture the properties necessary for the subsequent fraud detection algorithms as invariants of the national language allowing for easy deployment and maintenance of the whole system to new countries supporting different national languages. The anomaly detection methods that are using the universal representation are focusing on similarities among the transactions, building a relation graph to be able to detect large scale operation of single or a small group of fraudsters. The above described representation allows to reduce the amount of false positives of these detectors.

Název v anglickém jazyce

Resistant.AI Transaction Detection Engine - Representation Learning Module

Popis výsledku anglicky

Representation learning module for Resistant.AI Transaction detection engine that focuses on detection of fraudulent attempts to create e-shop orders using Buy Now Pay Later (BNPL) service. The representation learning module is focusing on universal representation of basket items of e-shop consumer's order that is easily transferable between individual countries. The representation is able to capture the properties necessary for the subsequent fraud detection algorithms as invariants of the national language allowing for easy deployment and maintenance of the whole system to new countries supporting different national languages. The anomaly detection methods that are using the universal representation are focusing on similarities among the transactions, building a relation graph to be able to detect large scale operation of single or a small group of fraudsters. The above described representation allows to reduce the amount of false positives of these detectors.

Druh

R - Software

CEP obor

—

OECD FORD obor

20202 - Communication engineering and systems

Projekt

<a href="/cs/project/FW02020147" target="_blank" >FW02020147: Transfer modelů umělé inteligence pro detekci podvodů podporujících expanzi na zahraničních trzích</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2022

Kód důvěrnosti údajů

C - Předmět řešení projektu podléhá obchodnímu tajemství (§ 504 Občanského zákoníku), ale název projektu, cíle projektu a u ukončeného nebo zastaveného projektu zhodnocení výsledku řešení projektu (údaje P03, P04, P15, P19, P29, PN8) dodané do CEP, jsou upraveny tak, aby byly zveřejnitelné.

Interní identifikační kód produktu

Resistant.AI Transaction - RL

Technické parametry

The result - representation learning module for a fraud detection system - has been fully integrated with the Resistant AI production environment and has been in production use since the H2 2021. The measurable technical parameters are twofold: precision and recall. On top of these parameters, we are improving key non-functional properties of the system: the amount of time it takes to deploy the new system in a previously non-covered geography or in a new context. The precision and recall parameters are essential for the economics of deployment, as we will detail below. The ability to expand to new territories is more difficult to measure directly, but is a significant factor in strategic expansion of Resistant AI services.

Ekonomické parametry

Precision increase: * Increases the acceptance rate for new customers as we remove false denials. It directly contributes to growth ratio and increases the efficiency of marketing spending, a key parameter for BNPL provider growth on a new market. * Contributes to faster growth in situations where the merchant selects the BNPL provider in a live dynamic auction. Being able to bid on candidate customers falsely rejected by other BNPL providers provides a competitive edge in the auction. * Achieves important societal goal, as the previously falsely rejected customers often came from disadvantaged backgrounds (refugees, itinerant populations). * Reduces the workload of fraud prevention. Recall increase: * We catch and stop more fraudsters. * We concentrate on professional criminals and target their ability to scale the fraud "business" they operate. * These fraudsters then reduce or stop their activities.

IČO vlastníka výsledku

07825439

Název vlastníka

Resistant AI s.r.o.