Ishu Anand Jaiswal’s digital identity study wins Best Paper award at ICDAM-2026

A study proposing a more explicit way to measure confidence in unified consumer identities has received the Best Conference Paper award at the 7th International Conference on Data Analytics and Management (ICDAM-2026).

The paper, “Unified Consumer Profiles Based on Real-World Identity and Trust Levels,” was written by Ishu Anand Jaiswal and presented at the conference held from June 12 to 14, 2026, at London Metropolitan University. Jaiswal, a senior engineering manager at Intuit, is the paper’s sole author.

The paper proposes a consumer-profile framework that separates three related aspects of digital identity: whether records belong to the same person, how strongly that identity has been verified, and how behavioural or fraud signals should influence confidence over time.

The work addresses a common challenge faced by organisations that combine customer records from multiple applications, devices or business units. A system may determine that two records probably belong to the same person, but that match alone does not establish how strongly the identity has been verified or whether recent activity has increased the associated risk.

As the paper puts it, “a profile being technically unified does not automatically make it trustworthy.”

Matching records is only the first step

Many customer-data systems seek to create a single consolidated profile, often referred to as a “golden record”. Jaiswal’s framework separates this process into three related decisions: whether records belong together, how much verified evidence supports the identity, and how behavioural or fraud signals should change confidence over time.

The proposed architecture first normalises information from multiple sources. It then uses privacy-preserving record linkage (PPRL) to compare encoded representations of identifying attributes rather than relying only on raw values.

Verified credentials and weighted attributes contribute to an identity-assurance score, while a separate trust score is updated as new behavioural, contextual and risk signals emerge.

Keeping these measures distinct is the central idea of the paper. In principle, a service could apply stronger verification to a high-risk action while avoiding the same level of friction for a routine interaction. It could also retain a confidence measure alongside a merged record instead of treating every match as equally reliable.

The paper presents the framework as a general architecture rather than a finished commercial system. Its components include data ingestion, identity matching, real-world identity verification, dynamic trust modelling and an update engine for resolving conflicting attributes.

What the experiment found

Jaiswal evaluated the framework using about 50,000 synthetic records generated across five simulated data sources. The test set included duplicate identities, incomplete or corrupted records, simulated fraudulent identities and clean records.

In the experiment, the proposed framework produced an identity-resolution F1 score of 0.91. The deterministic and probabilistic baselines reported in the paper scored 0.79 and 0.86, respectively.

The framework also reported stronger fraud-detection and privacy-risk results than the comparison methods, although it required more processing time.

The results, however, remain preliminary. The dataset was synthetic, the baselines and parameters were selected within the study, and the findings have not been independently replicated. The paper does not report deployment using live consumer data.

The results therefore demonstrate how the proposed architecture performed under controlled test conditions rather than establishing how it would perform at production scale.

A broader shift towards layered assurance

The research reflects a wider shift in digital identity design towards treating identity assurance as more than a simple yes-or-no determination.

The paper notes that current US National Institute of Standards and Technology guidelines distinguish among assurance for identity proofing, authentication and federation, while also calling for continuous evaluation as threats, technology and user needs change.

Jaiswal’s proposal operates at a different layer, after records from several systems have been linked. It focuses on how a unified profile can retain and update evidence about identity confidence. The paper makes a conceptual connection with the NIST approach rather than claiming to implement the NIST framework.

The approach also has relevance for India as organisations adapt to the Digital Personal Data Protection Rules, 2025. Privacy-preserving matching can reduce direct exposure of identifiers during record linkage, but it does not by itself address broader requirements around consent, purpose, retention, security and accountability.

Questions before deployment

A system that changes an individual’s trust score based on behaviour also raises governance questions.

Travel, shared devices, accessibility needs or an unusual transaction could appear anomalous without necessarily indicating fraudulent behaviour. A real-world deployment would therefore require explainable decisions, safeguards against false positives, testing for unequal error rates and mechanisms through which users can challenge consequential outcomes.

The security architecture would also require further specification and independent testing. Encoding an identifier alone does not guarantee protection against re-identification or linkage attacks. The strength of an implementation would depend on the techniques employed, the threat model and the way keys, tokens and source data are governed.

These limitations point to the next stage of research. The paper’s central contribution is to make uncertainty in unified digital identities more explicit.

A unified profile, the research argues, should not only indicate who a system believes a person to be. It should also retain why the system reached that conclusion, how strong the supporting evidence is and when that confidence should change.