Jamie Avi-Dan

‘Due-Diligence AI transforms investment due diligence from a 2–3-month manual process into a same-day, fully auditable workflow’

Part of the AI Leadership Index – a curated list of the most inspirational figures shaping the AI revolution.

Jamie Avi-Dan joined LPPI in 2019. He is responsible for all Information Technology, Artificial Intelligence & Cybersecurity. His responsibilities additionally extend to supporting LPFA and GLIL technologies.

Jamie possesses over 25 years of experience in managing the delivery of information technology across organisations including Merrill Lynch, Metro Bank, Financial Times, ATOS, BBC, Unilever and Ernst & Young, as well as within UK Government (Home Office & Foreign Office). Jamie read Physics at Kings College London.

Please share a case study of how you have you innovated with AI over the last year?

Over the past year, I developed Due-Diligence AI, transforming investment due diligence from a 2–3-month manual process into a same-day, fully auditable workflow. Traditionally, teams reviewed 100+ documents, cross-referenced thousands of DDQ questions and managed responses in spreadsheets - leading to inconsistent coverage, weak provenance (“where did this come from?”) and repeated rework when new materials arrived.

The platform addresses scale, reliability and confidentiality. It ingests large, multi-format document sets and distributes question batches across parallel AI “workers” for consistent, high-throughput analysis. It validates whether findings meet industry standards and, where gaps exist, proposes follow-up questions for external parties.

Security is embedded by design: files are cached for processing and deleted once validated.

The outcome has been dramatically faster turnaround, reduced manual effort and error, stronger audit trails, and significantly lower delivery costs.

What is the biggest challenge you have faced with AI innovation and how you have overcome it?

The greatest challenge was moving from “AI in a GUI” to trusted, repeatable automation for high-stakes decisions. Early prototypes could answer questions but struggled with what matters in due diligence: consistent coverage across 100+ documents, clear provenance, predictable cost and latency at scale, and confidentiality when using third-party models.

I addressed this by designing an auditable workflow first and layering AI into it. In Due-Diligence AI, documents are normalised and cached once, then analysed in parallel by structured “workers” handling DDQ batches. Outputs follow strict schemas, and every answer is linked to source evidence—removing black-box ambiguity.

To strengthen reliability, I added a quality gate that checks whether responses meet DDQ standards and, if not, generates targeted follow-up questions to reduce rework. The architecture is model-agnostic, routing sensitive tasks to internal models and others externally without altering governance controls.

The result is same-day turnaround, lower manual effort, reduced error risk and a far more defensible due diligence process.

How do you balance innovation with ethical safeguards?

It is essential to ensure human oversight, as well as a traceable and explainable set of process sets which AI undertakes. Ensuring every aspect of the process is subject to logging and reporting, ensures that any issue is quickly identified, explainable and resolvable promptly.

Other safeguards including ensuring prompts utilised are subject to ethical review (by both Human and AI parties), thereby placing ethical standards front and centre.

How do you believe the artificial intelligence space will change over the next 12-24 months?

Artificial intelligence will revolutionise each organisation in its own unique way, dependent on sector, regulation and requirement. Models will keep improving, but the bigger shift will be how they're used: more AI running inside products, workflows, and operating processes rather than sitting in a chat window.

We'll see stronger multi-modal capability (text, voice, image, video) become standard, and more "agentic" systems that can execute tasks across apps with guardrails, approvals, and audit trails. Cost and latency will drop as inference gets optimised and smaller, specialised models outperform general ones for specific jobs.

Enterprises will standardise patterns like retrieval, tool-use, orchestration, and evaluation, and treat prompts like code: versioned, tested, monitored. Regulation and governance will tighten, pushing better provenance, watermarking, data controls, and model risk management. Security will become more practical: defending against prompt injection, data leakage, and model supply-chain issues.

Net effect: fewer one-off AI experiments, more measurable automation, and a fight over trust, compliance, and integration rather than raw IQ.