Understanding and Analyzing the AFWERX Portfolio by Company Type
SHELDON
August 13, 2025
This document reviews a powerful data analysis technique called Topological Data Analysis (TDA). Based on the principles outlined in prior work, the core idea is to move beyond simple spreadsheets and create a “shape-based” map of the AFWERX portfolio. This approach reveals the deep, underlying relationships between companies, showing us not just who they are, but how they fit together into a larger strategic landscape. It helps us see the natural clusters, the unique outliers, and the unexplored gaps in our innovation ecosystem.
The power of this method comes from a series of deliberate, strategic choices made before the analysis begins. The final map is a direct reflection of the questions we ask and the data we use. The key decisions involved are:
Ultimately, TDA is not an automated answer machine; it is a tool for enhanced strategic thinking. The “shape” that emerges from the analysis is a mirror, reflecting the intent and wisdom of the questions asked. It provides a framework for understanding our portfolio not as a list of assets, but as a dynamic, interconnected ecosystem.
A crucial distinction of the analysis reviewed here is its specific temporal focus: the data evaluates companies at the moment before they received their first AFWERX award. This is not a study of what companies became with our help, but rather a foundational map of what they were at the instant of their selection. This approach establishes a critical “Genesis Block”—a baseline against which all subsequent performance and impact can be measured.
The primary goal was to understand the raw material we started with. What did the innovation landscape from which we selected our initial awardees truly look like? Were we choosing from dense clusters of technologically similar companies, or were we identifying unique pioneers from the outset?
To answer these questions, we employed Topological Data Analysis (TDA), an unsupervised machine learning framework. TDA excels at creating a “shape-based” map of data, revealing the natural groupings and hidden relationships within a complex portfolio. By using an unsupervised method, we avoided imposing our own biases and instead allowed the data itself to reveal the fundamental structure of the pre-award innovation ecosystem.
A vital step in this process is data preparation and scaling. To create a fair and unbiased map, every characteristic of a company—from its patents and team size to its industry and technology type—was converted into a universal, numerical language. The data was then scaled, preventing any single metric (like total prior funding) from dominating the analysis. This ensures we are comparing companies based on their intrinsic nature and capabilities, not just their surface-level metrics. This process is the bedrock of any credible impact analysis, as it provides an honest, unvarnished measurement of what each company was before AFWERX intervention.
This analysis provides two distinct lenses for viewing the portfolio at that foundational moment:
The Unnormalized View (“The World That Was”): This is the direct historical record. It shows the landscape as it existed, highlighting the companies that were the largest, most established, and best-funded before receiving their first AFWERX dollar. It is a map of the existing reality we chose from.
The Normalized View (“The Potential We Saw”): This is the scout’s map. For this view, we adjusted the data to look at metrics like performance per employee or innovation per dollar of funding. This lens reveals latent potential, elevating the efficient, scrappy innovators who might otherwise be overshadowed by larger competitors.
By examining our past selections through these two lenses, we can analyze the nature of our initial decisions. Did we favor established players, or were we betting on companies with the highest latent potential? Understanding our historical selection strategy is the most powerful tool we have for refining our future investment decisions and maximizing the impact of the AFWERX program.
If TDA provides the philosophical framework for our analysis (understanding the “shape” of data), then the Growing Neural Gas (GNG) algorithm is the engine that actually builds the map. GNG is a type of artificial neural network that is particularly well-suited for learning and visualizing the topological structure of a dataset.
Think of it as a flexible net thrown over the data points (our companies). The net starts small and simple, then iteratively “grows” and adapts:
The end result is a network graph where the nodes are archetypal companies and the connections represent their similarity. This graph is the TDA map we use to analyze the portfolio, allowing us to see the clusters, outliers, and empty spaces in our innovation landscape.
To make this process more concrete, consider the following workflow diagram:
flowchart TD
A["Raw Company Data\n(e.g., funding, team size, industry)"] --> B["Data Preparation\n- Numeric Conversion\n- Scaling / Normalization"]
B --> C{"Growing Neural Gas Algorithm"}
C --> D["Topological Map\n(Network Graph)"]
D --> E["Analysis & Insights\n- Identify Clusters\n- Find Outliers\n- Discover Gaps"]
For a more foundational understanding of this methodology, readers are encouraged to consult the original source document. The link provides additional details and illustrative examples that are key to grasping the nuances of the TDA process. Specifically, the source document elaborates on:
The full text can be reviewed here: https://www.sheldon-insights.com/deliverables/nestt/clin003/afwerx/understanding_tda_gng/write_up
The collection of materials from the AFWERX presentation offers more than just data; it provides a masterclass in constructing a strategic narrative. To understand its full impact, one must view it not as a series of slides, but as a deliberate, multi-layered journey designed to lead leadership from high-level observation to granular, actionable insight. The full exhibit list can be reviewed here: https://www.sheldon-insights.com/deliverables/nestt/clin006/afwerx/afwerx_presentation/analysis_materials/
The narrative of the analysis unfolds in a logical, cascading sequence, with each layer building upon the last with concrete data:
Establishing the Framework (The “Who” and “How”): The analysis first establishes its core lexicon. This is the foundational layer that makes the rest of the discussion possible.
The 30,000-Foot View (The “What”): With the framework in place, the analysis begins with high-level summaries. The “SUMMARY TABLE” and “VISUAL BREAKDOWN” exhibits provide the essential executive overview. For instance, they establish the baseline reality that Mature DAF/DoD Suppliers enter the program with a median of $1.3M in prior obligations, while New DoD Participants start at nearly $0. This immediately quantifies the vast difference in experience between the quadrants.
Introducing the Time Dimension (The “When”): The analysis then deepens by introducing the element of time. The “SUBSEQUENT VS PRIOR GROUP ANALYSIS” moves beyond a static snapshot to reveal the portfolio’s dynamics. It shows that in the two years following their AFWERX award, the New DoD Participants quadrant demonstrates explosive growth, achieving a median of $450k in new obligations. This directly contrasts with the Mature DAF/DoD Suppliers, whose subsequent obligations are a fraction of their prior work, illustrating a fundamentally different engagement model.
The Diagnostic Deep Dive (The “Why”): The narrative then zooms in to diagnose the reasons behind these trends. The “KEY METRICS FOR…” exhibits reveal the underlying mechanics. We see that the growth for New DoD Participants is not just from small awards; their median subsequent award size is $150k, indicating they are successfully landing significant, meaningful contracts. This is a critical diagnostic insight when compared to the MAWs, whose median subsequent award size might be smaller, suggesting a different transition challenge.
Grounding in Reality (The “Who, Specifically?”): Finally, after establishing the archetypes and their quantified dynamics, the analysis grounds itself in concrete reality with the “DEEPER DIVE, FPDS PHASE IIIS BY COMPANY” exhibit. It moves from the abstract median of “$450k in new obligations” to showing the specific companies, like “Company X with a $2.5M Phase III,” that make up that number. This provides tangible examples of success, making the data real and relatable, and showing what a successful transition looks like in practice.
By structuring the analysis in this holistic way—from a qualitative framework to a high-level quantitative overview, and then to time-based dynamics, diagnostic metrics, and specific examples—the presentation creates a powerful and persuasive narrative. It equips leaders not just with data points, but with a comprehensive, evidence-based model for understanding their portfolio.
This methodology is not just an academic exercise; it is designed to provide answers to concrete strategic questions about the AFWERX portfolio. By creating a topological map of our pre-award companies, we can address the following:
This baseline analysis of the pre-award portfolio is the first step in a larger journey of data-driven portfolio management. The foundational map we have created enables several critical future analyses:
This document has reviewed a powerful analytical lens: Topological Data Analysis. We have moved from the abstract theory to the concrete application, showing how this methodology can deconstruct the AFWERX portfolio into a landscape of archetypes, pathways, and quantifiable insights. The final question is not “Was this analysis interesting?” but “What is the strategic imperative of this capability?”
Our primary military competitors build their industrial power through centralized, top-down, five-year plans—an approach that is rigid and predictable. The United States’ core asymmetric advantage is our dynamic, decentralized ecosystem of innovators. This is our greatest strength, but also our most profound challenge. A network we cannot see is a network we cannot strategically cultivate. The critical challenge for AFWERX is to harness the power of our network without crushing it under a bureaucracy that mimics our adversaries’ weakness.
TDA presents a path forward, offering more than just data science; it is a command-and-control interface for a network-based strategy.
The decision before AFWERX is therefore not a technical one, but a philosophical one. Do we wish to continue managing a list of individual contracts, or do we want to cultivate a strategic innovation ecosystem? If the answer is the latter, then the path forward is clear. We must continue to invest in the tools that allow us to see, understand, and shape the network that is our single greatest asymmetric advantage.
To further explore the concepts and applications of Topological Data Analysis, the following is a curated list of accessible articles, tutorials, and case studies.