Video

Anonymising Unstructured Test Data That Contains Sensitive Customer Information

When building GenAI solutions that process unstructured customer content, emails, call centre interactions, and financial requests, the test data has to read like the real thing. Redaction destroys the contextual signals your model needs to function. Synthetic substitution is a step forward but carries its own limitations. This video walks through all three approaches to anonymising unstructured test data, redaction, synthetic substitution, and fully constructed synthetic data, and sets out what each one means in practice for organisations that need zero re-identification risk, automatic labelling, and datasets that scale on demand.

Testing AI Agents

A December 2025 paper by Melissa Z. Pan et al. Measuring Agents in Production surveyed GenAI projects across industries and reached two findings that should concern any organisation deploying AI in high-accountability environments: determining the reliability of AI agents remains unsolved, and generating the gold-standard unstructured test data needed to measure that reliability is described as “nearly infeasible.” This video sets out a direct response to both. Agent reliability must be solved - the operational and regulatory exposure of deploying agents whose reliability cannot be demonstrated is not a sustainable position. And while generating gold-standard unstructured test data is genuinely difficult, it is not impossible. The video explains what that looks like in practice. 

Why Your Test Data Is Holding Your GenAI Back

Every organisation building a GenAI system faces the same structural problem: the real customer data that would make testing meaningful is the data you cannot safely use. The alternatives teams turn to instead, such as anonymisation, manual creation, and synthetic substitution, each fail in the same fundamental way, producing test data that does not reflect how real people actually communicate. The result is a GenAI system that performs well in testing and fails in production, triggering regulatory scrutiny, misclassified complaints, and remediation cycles measured in months. This video explains why that gap exists, what it costs organisations who encounter it in production rather than before, and how Data Infusion constructs unstructured synthetic datasets built around your operational context, validated and documented before delivery, with no private real data accessed at any stage.