Blog

Next-Gen Financial Crime Fighting with AI and external data

Written by Rebecca Lindley | 04 November 2025

As financial crime rises, so does pressure to reduce resource costs. Thousands of anti-financial crime leaders around the globe are battling this paradox every day. Widening intelligence gaps and growing compliance expectations mean that finding a solution is becoming more urgent. 

In this article, we’ll explore how, where and why artificial intelligence can be part of this solution. We’ll focus on AI’s application to investigations requiring external, publicly available data (Open Source Intelligence or OSINT), considering how humans and technology can be combined to maximise impact without additional costs.  

The OSINT problem 

External data – or OSINT – refers to the publicly available or licensable data found in sources such as corporate registries, search engines or even social media. This data is used within numerous types of financial crime investigation where internal data doesn’t provide a full picture of a client and their activity. OSINT is particularly essential to complex investigations, where a case has been escalated because it could not be closed using other available data. 

The challenge is that OSINT use in investigations is arguably one of the most inefficient processes in a financial institution. As a dynamic intelligence technique drawn from disparate sources, it’s nearly impossible to ensure consistency, quality and, above all, efficiency. Where numerous tools exist for processes such as transaction monitoring, teams responsible for more complex investigations are often left with rudimentary, non-specialised options such as Excel and Google.  

The outcome is that 70-80% of time in each case is spent on collecting relevant information and carrying out administrative tasks. Only 20-30% is spent on the valuable investigation work these teams are skilled in. Finally, upon investigation, only 15-25% of these complex cases are considered true positives in terms of risk: a very low proportion considering the time and effort spent in this part of the process. There is a clear opportunity for financial institutions to achieve efficiency gains by remedying the lack of tooling in this area. 

In this blog, we’ll discuss the types of solution financial crime investigators need and the best tools for the job.

Where does AI fit in? 

From trustworthiness to security, there are numerous valid concerns about AI implementation in an enterprise context. Financial institutions need to implement AI in a nuanced, thoughtful manner to ensure that they reap the benefits while minimising these risks.  

Let’s return to our ultimate aim: we want to get the most accurate and reliable information to investigators as quickly as possible, whilst taking into account security, data regulations and other considerations. AI and automation both have roles to play here. 

Put simply, humans and AI have complementary skillsets. Generally, AI is very good at sorting, filtering and summarising large volumes of data. When humans do this, it’s time-consuming, repetitive, inaccurate and costly. We’re good at things that are less repetitive but require more nuance and understanding, such as contextual analysis of the data AI has returned.  

Below are some practical examples of how and where AI can be implemented as part of OSINT investigations in banks, to maximise effectiveness without increasing risks. 

Prompt Engineering 

 “You have to be very selective when strictly prohibiting things. …if you don’t offer them an alternative, it’s likely to fail from the start, as there is a high likelihood that people will just use it anyway.”

- Moody's KYC AI Study, 2024

Many investigators will be using tools with integrated AI, often not necessarily by choice; even Google searches provide an AI summary automatically. At first glance, these tools can appear impressively accurate, but ensuring quality, consistency, auditability and security remain challenging in an enterprise context. Without these assurances, financial crime investigations teams cannot prove compliance to the regulator. 

Simply asking an AI model “where is the risk?”, for example, is vague and can lead to bias or hallucinations. To ensure a minimum standard across teams, financial crime leaders should consider implementing prompt frameworks, such as the AUTOMAT framework. These frameworks encourage a structured approach to prompt engineering so that AI outputs are reliable, explainable, and compliant. Most importantly, prompt frameworks can allow banks to set their own risk parameters without making resource-intensive changes to the solutions or the tools available. 

Implementing prompt frameworks reduces the risk of generic AI tool use. Teams can use similar prompts for similar use cases, creating all-important consistency whilst enjoying the benefits of AI. 

GraphRAG 

GraphRAG (Retrieval-Augmented Generation) is an example of an AI technology that can be preconfigured to specific investigation types. By combining graph databases or systems designed for interconnected, relational data with large language models, it enables context-aware reasoning across complex datasets, including entities, relationships, and entire networks. 

This is particularly relevant in financial crime investigations involving money laundering, sanctions evasion, or tax fraud, where analysts depend heavily on understanding networks of relationships across entities, transactions, and jurisdictions. 

GraphRAG enhances investigative intelligence through three core capabilities:  

  • Connection discovery: GraphRAG helps connect the dots between key entities in information that usually sits in separate silos, such as internal reports, news articles and corporate filings. This lets analysts uncover hidden insights that may normally stay buried. 
  • Multi-layer analysis: By using graph reasoning, GraphRAG can uncover indirect or hidden links within a wider network. For example, it could flag a sanctioned individual connected hidden in several layers of corporate networks, which traditional tools might miss. 
  • Intelligent prioritisation: GraphRAG can analyse thousands of entities and automatically highlight the riskiest or most suspicious connections in an investigation. By suggesting areas of focus, it saves investigators time and allows them to focus on human decision-making. 

Generative AI 

As already discussed, Generative AI tools like ChatGPT, Gemini and Copilot can be useful in an anti-financial crime context when the right prompts are applied. For example, they can be used for initial research and data collection, or to draft reports and summaries.  

However, a range of challenges remain when it comes to widespread implementation in an enterprise anti-financial crime context: 

  • Lack of use case optimisation – Publicly available AI tools are general models that have been built for a wide range of use cases. These tools are not trained or optimised for the types of data involved in complex investigations.   
  • Explainability - These tools are fundamentally lacking in the audit capability required to meet regulatory standards. They don’t show reasoning or the source of each point they make in a compliant manner. 
  • Data sourcing – Similarly, there’s little or no control over the data publicly available AI tools source. They are likely to review everything they can find on the internet, without weighting reliability in a way that is appropriate in an anti-financial crime investigation. For example, many of these commercial models might look at things like Reddit or Wikipedia and consider them as valuable as structured, curated data like corporate records.  
  • Security - Feeding these tools potentially sensitive data is a major security risk for financial institutions. Data needs to be kept within an organisation’s jurisdiction, where there are controls over what data can be accessed and who can access it. 

To get the most accurate results in the right way, financial institutions should consider using purpose-built tools instead. 

Agentic AI 

Agentic AI is an AI technology that executes tasks, makes decisions and takes actions towards a defined goal within preset guidelines. Today, Agentic AI is amongst the most controversial AI technologies in an enterprise context. Yet, with the correct safeguards, it has the potential to make a real difference to outcomes in AFC teams.  

In many cases, complex investigations are too resource-heavy for the outcomes they produce, both in terms of time management and cost. OSINT remains key to these processes – otherwise investigators may miss key details, risking regulatory action.  

This is where Agentic AI could help. Many of the steps investigators take in complex investigations, especially initial ones, are relatively repetitive. Agentic AI can carry out these data-rich tasks for us, including data collection and analysis. It can then give us a report on its findings, or even a partially completed case that investigators can then continue manually. Most importantly, the right system will tell us what it’s doing and why at every stage, ensuring that a human is kept in the loop. 

This approach can be used to identify cases that merit further, human-led investigation, and those that can be closed. Not only can it reduce the burden on complex investigations teams, it can also provide better quality information to start with so that their skills are more effectively utilised. 

Many organisations feel unsure about agentic AI because of its decision-making nature. However, the key to effective agentic AI implementation in this context is to contain the data it can access, provide extremely specific and well-engineered prompts, and ensure completely transparent, explainable reasoning. Financial institutions can also ensure that humans make final decisions to comply with regulations around compliance and AI. 

What’s next? 

In this article, we’ve highlighted a range of AI technologies and approaches that FIs should consider applying to enhance the use of external data or OSINT in complex investigations teams. While there is understandable nervousness around security, quality and ethical considerations, the potential efficiency gains outweigh risk in many cases.  

At Blackdot, we are experts in supporting investigations by providing AI and automation to the enterprise. By focusing on reducing the time teams spend on manual investigations, we’ve reduced investigation overheads by up to 80%. 

If you’re looking for a partner to support AI implementation in this context, don’t hesitate to reach out