Tradecraft at Scale: How Insights Investigator Brings Expert Methodology to AI

As artificial intelligence has made information more accessible than ever, the challenge facing investigators is not simply finding data. Finding relevant, trusted, and mission-appropriate information across fragmented sources remains difficult. The larger challenge is finding it, connecting it, and determining what matters.

Experienced analysts, investigators, and intelligence professionals don't simply search for information. They know how to frame questions, identify relevant sources, test assumptions, follow leads, validate findings, and determine when a conclusion is supported — or when more evidence is needed. That expertise is often built through years of operational experience.

As Babel Street developed Insights Investigator, one of our goals was to ensure that the system reflected proven investigative principles. Rather than approaching AI as a standalone answer engine, we focused on combining advanced language models with Babel Street's unique data holdings, investigative workflows, and practitioner expertise.

The result is a system designed to go beyond information retrieval and help users conduct more effective investigations.

In part 1 of our Talking Tradecraft series, we explored our leaders’ views of tradecraft and what it means in the agentic AI era.

Here in part 2, we’ll discuss how Babel Street incorporates tradecraft into Insights Investigator workflows.

Finally, in part 3, we’ll speak with our practitioners to learn the rules they follow for sound investigative tradecraft.

Moving beyond generic language models for investigative AI

Large language models (LLMs) are extraordinarily capable tools, but like any technology, their effectiveness depends on how they are applied.

Modern AI systems can be guided through structured instructions, operational constraints, and decision frameworks that shape how they approach a task. These mechanisms help ensure the model follows consistent processes, evaluates information appropriately, and behaves in ways that support the user's objectives in the most accurate way possible.

Andrew Alach, Product Manager, explains the objective was never to create a system that immediately jumps to conclusions, rather “We compel the system to never try to draw immediate conclusions based on the sources it finds.”

That philosophy is important because investigations rarely produce perfect information. Effective analysis requires gathering evidence, evaluating competing information, and understanding where uncertainty remains.

Why retrieval-augmented generation matters for investigations

A key component of modern investigative AI is retrieval-augmented generation (RAG). In simple terms, RAG allows a language model to retrieve relevant information before generating a response. Rather than relying solely on what it learned during training, the system can access current data collections, identify relevant content, and ground its responses in those materials.

That grounding is necessary, but it is not the same as tradecraft. RAG helps ensure agents work from current, relevant source material. Tradecraft comes from how expert investigative methods are encoded into the agent’s instructions, planning frameworks, tools, decision logic, investigative pivots, validation steps, and corroboration practices.

John Larson, President and Chief AI Officer, described this challenge through the broader lens of expertise. Whether the task involves identity resolution, geopolitical analysis, or technical intelligence, investigators often need access to a large body of accumulated knowledge to make informed decisions. RAG helps make that expertise available at the moment it is needed.

Babel Street’s agents retrieve better information and are grounded by Babel Street’s data and imbued with the tradecraft of experienced practitioners.

Investigative AI powered by Babel Street Data

As a natural conclusion, the effectiveness of any investigative system ultimately depends on the quality and breadth of the information available to it. This is where Babel Street's data strategy becomes a significant differentiator.

Insights Investigator works across Babel Street's vast collection of publicly available, commercially available, and mission-relevant data sources. Instead of being limited to a narrow repository of documents or a single class of information, the system leverages the breadth of Babel Street's data ecosystem to identify relationships, uncover context, and surface relevant intelligence.

For investigators, this means starting with a question rather than a dataset.

Whether the investigation begins with a person, organization, username, phone number, vessel, event, location, or threat indicator, Investigator can draw upon Babel Street's extensive data collections to help identify potential pathways for further analysis.

The result is a more complete investigative starting point and greater visibility into the information landscape surrounding a target of interest.

Building investigative discipline into AI workflows

While access to data is an important part of the equation, a central design principle behind Insights Investigator is that technology should reflect the discipline of experienced investigators.

According to Jesse Bennetter, Director of Product Management, the development team worked closely with experienced practitioners across the company. The goal was to understand how skilled investigators use available tools, what questions they ask, how they refine searches, and how they pursue leads.

He said, “We've taken best practices from the folks in this company, and we've tried to understand how they utilize our capabilities. That effort enabled the team to capture the investigative methodologies that experienced practitioners apply every day and incorporate them into the way Investigator approaches investigative tasks.”

In effect, Babel Street has operationalized proven investigative habits within Insights Investigator.

Capturing investigative tradecraft, not just search queries

Our tradecraft experts confirmed that investigations are rarely linear. Usually, experienced investigators begin with a question, develop a hypothesis, pursue multiple avenues of inquiry, evaluate alternative explanations, and continually refine their understanding as new information emerges.

John described this process through identity-resolution examples, where investigators move from one fragment of information to another — linking identities, accounts, devices, locations, and records through a sequence of investigative pivots.

The challenge was determining how a system could support that style of thinking at scale. Rather than just functioning according to prompts, Investigator is designed to decompose a task, develop and execute a research plan, use available tools, follow multiple lines of inquiry, challenge assumptions, and surface the evidence behind its findings.

This approach extends beyond simple instructions. It reflects an effort to capture investigative methodology itself — the ways experienced analysts break down problems, pursue evidence, test alternative explanations, and evaluate information as an investigation evolves.

Why human judgment remains central to AI investigations

While AI can dramatically accelerate investigative workflows, judgment remains a human responsibility. No AI system can guarantee truth. Insights Investigator is designed to seek corroboration, identify conflicting information and uncertainty, and provide sourced evidence for the analyst to validate.

John also highlighted the danger of blindly accepting AI-generated conclusions — confirmation bias. This is where “you just assume that because AI gave it to you, it’s right. And this is a huge fallacy.”

For that reason, Investigator was designed to support analysts rather than replace them. John further noted, “At the end of the day, the human needs to be at the center of it ... to look at, to evaluate, to assess, to corroborate.”

The future of investigative AI and agentic workflows

The future of investigations will not be defined solely by larger language models or faster processing speeds.

It will be defined by how effectively organizations combine trusted data, investigative expertise, and AI-driven workflows.

By bringing together Babel Street Data, retrieval-driven intelligence workflows, expert-informed investigative methodologies, and concrete operational guardrails, Insights Investigator helps organizations move beyond simple information retrieval toward a more sophisticated model of investigative support. Those guardrails include controlled access to data and tools, transparent plans and actions, human approval where appropriate, citations and provenance, and an auditable record of how the system reached its findings.

The result is an AI system built not just to answer questions, but to help practitioners investigate more effectively, work more efficiently, and make more informed decisions.