Doing More with Less: Automating Human Intelligence Data Analysis with SOAR HI

A Way to Manage Data

In his January 2025 article for Military Review, “Data Literacy: How We Prepare for the Future,” GEN Gary Brito wrote, “With so much data available, it is a challenge to rapidly leverage our systems to get meaningful information that drives decisions.”¹ Every intelligence professional has felt this challenge. All published intelligence information reports (IIRs) contains structured metadata: serial numbers, priority intelligence requirements (PIRs), intelligence collection requirements, collection management codes, locations of acquisition, countries, preparers, sources, and more. When analyzed in aggregate, this metadata reveals collection trends, outstanding information gaps, and team effectiveness in addressing the commander’s PIRs. The challenge is not that this analysis is impossible; rather, intelligence professionals do it daily through a combination of experience, institutional knowledge, and manual effort. The challenge remains that manual methods are slow, difficult to scale, and hard to repeat consistently across large volumes of reporting.

To address this challenge, I developed a Python-based software program, the Statistical Operations, Analysis, and Reporting for Human Intelligence (SOAR HI) tool, to automate the extraction and analysis of IIR metadata and to generate an interactive dashboard. This article describes the SOAR HI program’s capabilities, demonstrates its effectiveness through screenshots of simulated data, and shares lessons learned from testing multiple iterations across classified and unclassified platforms. Ultimately, this article outlines a practical approach to applying data literacy within the human intelligence (HUMINT) discipline and encourages the intelligence community to think strategically about how we manage and present our own data.

The Challenge

Intelligence professionals at every echelon regularly work with IIR data. Collection managers track which requirements are being addressed, while team leaders monitor their collectors’ output. Operations management teams and G-2 staff assess collection effectiveness to influence planning. These tasks are routine and important, and professionals accomplish them through a variety of methods including spreadsheets, databases, and institutional knowledge built through years of experience.

The difficulty arises with volume and complexity. A single IIR may address multiple PIRs, collection requirements, and countries. For example, a dataset of 171 IIRs contained 810 data points in a single metadata field alone. Large datasets like these can overwhelm collection managers with the sheer volume of collection assets, named areas of interest, and PIRs, which easily turn into a web of cross-references. Performing this work manually is not impossible, but it is time-consuming. I have done it myself: a quantitative assessment of a single HUMINT collection team’s effectiveness answering PIRs over three months of IIRs took approximately 6 hours of manual work. That assessment covered one team. An operations management team overseeing multiple teams across several locations could easily face days of work.

The Functions of SOAR HI

Conveniently, SOAR HI users can place IIR files in a folder, run the program, and receive a complete interactive dashboard. The same analysis that once took hours by hand is immediately available and can be repeated, filtered, and explored from multiple angles without additional effort. SOAR HI can ingest a batch of IIRs and break down each report into its structured data fields by serial number, subject, date of acquisition, location of acquisition, collection management codes, country or nonstate entity, preparer, PIRs, intelligence collection requirements, National Intelligence Priorities Framework (NIPF) topics, intelligence function codes, and source identifiers. The program then organizes this data into an interactive dashboard.

For each field, SOAR HI generates a frequency table showing each unique value, the number of reports associated with that value, and the percentage of the dataset it represents. It then calculates a five-number summary (a standard statistical snapshot covering the minimum, first quartile, median, third quartile, maximum) for fields where reports may contain multiple values, such as the number of requirements addressed per report. As a result, a time-series chart will display reporting volume over time by day or by month. Figure 1 shows two screenshots from SOAR HI, side by side: the collection management code analysis and country analysis, with frequency tables displaying counts and percentages.

The dashboard is interactive and user-friendly. Users can click any value in any frequency table to apply filters:

• Clicking a specific PIR, for example, filters every other table to show only the reports that address that PIR.
• The user can then see which locations, preparers, collection management codes, and requirements are associated with that PIR.
• The serial numbers of the filtered reports are also visible, pointing the user directly to the specific reports worth reading in detail.
• Filters can be stacked to further narrow the analysis, removed individually, or cleared entirely.
• Users can also apply filters from dropdown menus at the top of the dashboard without needing to scroll through tables first.

Knowing where to look across a large volume of reporting is itself a valuable product.

SOAR HI also includes a label toggle feature for fields that use codes. Intelligence collection requirements, PIRs, NIPF topics, intelligence function codes, and collection management codes are often represented by alphanumeric identifiers. A requirement code like “T-AOR-5200-IADS-26” is not immediately meaningful without context. When the label toggle is activated, the requirement description appears next to each code, so that same requirement appears as “T-AOR-5200-IADS-26: Integrated Air Defense System (IADS) Network.” Similarly, the collection management code “DB” appears as “DB: Debriefing,” and the NIPF code “CTER” appears as “CTER: Counter-Terrorism.” This feature addresses a persistent challenge across intelligence systems: coded data that is technically present but practically inaccessible without a separate reference. Figure 2 is a screenshot from SOAR HI showing the requirement analysis frequency table with labels toggled on, displaying the full description next to each requirement code.

How SOAR HI Supports Intelligence

SOAR HI does not perform intelligence analysis, nor does it assess sources, interpret threat intent, or draw analytical conclusions. It performs data analysis on IIR metadata, including counting, grouping, filtering, sorting, and calculating statistics. The output supports intelligence professionals by providing quantitative context that informs their own judgment and decision-making. Field Manual 2-0, Intelligence, establishes the core tasks of the Intelligence Warfighting Function.² SOAR HI supports several of these tasks in practical ways.

Support to situational understanding. SOAR HI organizes IIR metadata by topic, location, time, and collection method, contributing to an intelligence running estimate. A staff officer can open the dashboard and see what has been reported, where, and how recently, without manually reviewing each report.

Conduct information collection. The dashboard shows which requirements are being addressed and which have collection gaps. A collection manager can then use this information to update collection plans based on observed reporting patterns rather than relying on memory alone. If a specific requirement shows minimal reporting for the current period, the gap is immediately visible.

Support to targeting. The filtering feature allows users to isolate all reports associated with a specific country, location, or topic and examine the associated metadata, including the serial numbers of relevant reports. As this application matures, its ability to rapidly identify and cross-reference relevant reporting could support the development of targeting packages.

Beyond doctrinal tasks, SOAR HI has practical applications for quality control and performance management. Consider a G-2 overseeing two HUMINT teams. SOAR HI can generate a five-number summary for the number of requirements addressed per IIR for each team.

Table. Team Effectiveness Summary

Metric	Team Alpha	Team Bravo
Minimum	1	2
First Quartile (Q1)	1	4
Median	2	6
Third Quartile (Q3)	3	8
Maximum	5	12

Using the table above as an example, a leader can identify that Team Bravo’s median reporting addresses three times as many requirements as Team Alpha’s. Team Alpha’s first quartile is one, meaning at most a quarter of their reports address only a single requirement. These numbers do not explain why the gap exists, but they ensure the right questions are asked: is it a training issue, an accessibility problem, or something else entirely? The five-number summary provides an objective starting point for conversations that may otherwise rely on general impressions.

SOAR HI dashboards may also complement or streamline routine products. Whereas a traditional HUMINT Summary provides a qualitative overview by listing subject lines of recent reports, a SOAR HI dashboard provides the same information with quantitative context such as how many reports address each PIR, how reporting is distributed across locations, and how current output compares to the broader dataset. This additional depth can enhance the utility of routine reporting products for commanders, analysts, and collection managers alike.

Lessons Learned

I developed multiple iterations of SOAR HI across classified and unclassified environments. Each iteration led to lessons that extend beyond this specific tool; they can be applied to any effort to build data capabilities for the intelligence community.

Adapt to the data formats users encounter. The classified versions of SOAR HI, delivered via the Secret Internet Protocol Router Network (SIPRNet), read IIRs exported solely from the CHROME system, the Department of War’s standard HUMINT reporting system. CHROME can then export individual IIRs or batch-export search results into a single text file. The unclassified training version uses Microsoft Word documents, which accommodate draft IIRs and training scenarios. Across these versions, the lesson was consistent: meet users where their data resides.

Labels matter. Through my beta testing, one SIPRNet version used the default glossary from the system of record to display labels for requirement codes. That glossary was incomplete, leaving many codes without descriptions. A subsequent version used a custom glossary that included all available labels. The difference in usability was significant. Coded data without labels is likely to be scrolled past. Any system that stores coded data should maintain a complete, current glossary. This is a lesson for system developers and program managers as much as it is for individual tool builders.

Artificial intelligence (AI) is effective for programming but unreliable for data analysis. I used large language models to assist with writing the Python code for SOAR HI. This experience was generally positive, particularly with the more capable models. The AI could generate functional code from detailed instructions, debug its own errors when given specific feedback, and iterate rapidly. Building a reusable prototype with AI assistance took less time than performing a single manual assessment. However, my experience using AI for data analysis was unsatisfactory. When asked to track progress toward a production goal of 188 simulated IIRs, the AI distorted the count of completed reports. The AI did not process any real intelligence data and posed no security risk, but its analytical output required independent verification. The lesson is specific: AI is a capable programming partner for a knowledgeable developer, but it is not a reliable substitute for a human performing data analysis.

Simulated data enables training. Through recommendations from within the 101st Airborne Division (Air Assault), I created an unclassified version of SOAR HI that includes approximately 188 simulated IIRs set in a fictional operational environment using ODIN, the Operational Environment Data Integration Network, as a guide. These reports use simulated requirements, PIRs, sources, and collection management codes. The simulated data allows any user to run SOAR HI and explore its capabilities without access to classified reporting. For units looking to build data literacy skills, the simulated dataset offers a ready-made opportunity to practice examining distributions, comparing subgroups, identifying patterns, and asking questions that the data can answer. Figure 3 is a screenshot from SOAR HI, showing the interactive filter bar with multiple filters applied and how the dashboard narrows the analysis to a specific subset of reports.