In many defense circles, the phrase "the right data" gets thrown around like an objective commodity waiting to be collected.
If one spends enough time inside operational environments (e.g. unmanned maritime systems or border-security command centers), you’ll see a different reality emerge.
There is no single "right" dataset. There is only the right data for your mission, your vantage point, and your decisions.
This distinction matters more than ever. Defining the right data up front—computationally, not just verbally—pays massive dividends in speed, clarity, and relevance.
Consider these counter-drug examples:
Same counter-drug mission. Same general timing and focus. Completely different definitions of "right."
This paradox is what lead to the foundation of the Joint All-Domain Command and Control (JADC2) which has the goal of connecting sensors across the US Military to allow “automation, artificial intelligence (AI), predictive analytics, and machine learning to deliver informed solutions via a resilient and robust network environment.”
The struggle, according to analysts within the military, is that "actors at the strategic, operational, and tactical levels” are dealing with only elements of the massive amount of data. The result is that each “shooter” (i.e. person or system meant to target an enemy) and staff function gets only the subset relevant to its mission.
This is why so many data-fusion programs fail: they assume one set of feeds can satisfy every stakeholder. Pushing all the data into a single UI increases cognitive load. Thus, data needs define mission structure and goals versus the other way around.
Anyone who’s worked for any time in a large organization has experienced this scenario: a Tactical Operations Center (TOC) brimming with unfiltered info, video feeds, sensor logs, chat rooms, and spreadsheets. Screens everywhere. Data pouring in from every sensor and system available.
And yet when the commander asks something direct—"Where is the target now?" or "Can we intercept before they cross the border?"
The room goes quiet. Nobody can answer the actual question.
CSIS (the Center for Strategic and International Studies) argues that pushing "all sensors, all shooters, all the time" into one system builds an "iron mountain" of data that dwarfs human ability to make use of it.
JADC2 aims to connect even more systems than are already operational yet practitioners with detailed knowledge of the project note that if "all actors get all information," the force ends up "drowning in data" instead of gaining decision advantage.
The quick solution, especially in a post-ChatGPT world, is to feed all that information into an AI.
IT hygiene is only the tip of the iceberg. Disconnected or overloaded data flows undermine operational outcomes. GAO found that poor interoperability among intelligence systems made it harder to hit time-critical targets and distinguish friend from foe. AI alone can’t compete against the age-old principle of garbage-in/garbage-out.
Defining the right data forces tough but necessary clarity:
Once you articulate those answers, everything downstream moves faster and becomes more interpretable. That includes sensor tasking, fusion logic, agent behavior, alerting, and more.
Defense analysts have used structured analytic techniques for decades to overcome cognitive bias and ambiguity. They take time upfront, but they save more hassle downstream.
Defining what matters before collection begins means fewer wasted sensor cycles, less noise in the fusion pipeline, and faster time-to-answer when the mission is live. Teams that skip this step often find themselves playing cleanup or starting from scratch, only now it’s under operational pressure.
Therefore, they should be considered an investment in the future.
The DoD has codified this principle in determining information up front in its AI ethics guidance, which requires that AI capabilities be "traceable," with transparent methodologies and data sources. In practice, this pushes teams toward structured techniques that make assumptions and hypotheses explicit before sensors ever collect data.
Three techniques matter here in determining what the “right” data is:
When a maritime interdiction team decomposes "illicit transport vessel" into entity attributes, they know what "the right data" looks like before a single sensor collects anything.
These techniques become a bridge between operators, analysts, and technologists.
Having a shared understanding of what the right data is among team members is good. Encoding it computationally changes everything.
That's where ontologies come in. An ontology lets you formally define the entities, relationships, and attributes that matter for your mission. This translates understanding from a person-to-person level into machine-readable structure.
After all, unleashing an AI on all your data without a definition of what you care about is going to fill in the blanks versus interpret what is already present in the data.
A DoD Inspector General report stressed the issues that can arise from this as "properly exchanging data between systems and maintaining semantic understanding are critical for successful decision-making and joint military operations"—exactly what mission-specific ontologies enforce.
Mission-specific examples:
Meanwhile, CSIS argues that for ad-hoc joint command networks, "having the same data language is critical" so units can plug into each other under combat time pressure. Mission ontologies play precisely this role for humans and AI agents alike.
When you give an AI agent this structure, you're giving it a worldview to query versus data to process.
Defining the right data also has direct implications for how quickly and effectively sensor data can be combined into a coherent picture (aka sensor fusion).
One reason sensor fusion is traditionally slow: we try to fuse data without first defining what "good fusion" even looks like.
The "Creating Data Advantage" memo tells DoD leaders to make data "visible, accessible, understandable, linked, trustworthy, interoperable, and secure." That only becomes real when missions model what entities exist, what attributes they require, and where the gaps are.
When you model the mission explicitly, you can suddenly see:
Example: A drone detects a potential GPS jammer via RF signatures. If the mission ontology defines "jammer" as an entity with attributes like geolocation, frequency characterization, and possible visual confirmation, then an agentic system immediately knows:
Without the ontology, the system produces an alert. With the ontology, it produces actionable decisions.
Work on Bayesian multi-sensor fusion for target identification notes that operational users expect standardized identity categories (per NATO STANAGs), which forces programs to define entities and attributes consistently before fusing AIS, GMTI, and other feeds.
Organizations that are accelerating the fastest aren't collecting as much data as possible. Rather, they’re defining what they need more clearly.
When you define the right data,
DoD's AI principles insist that capabilities be "traceable" and "reliable" within well-defined uses. That's only achievable when missions give agents an explicit ontology. Otherwise, the system is guessing over opaque correlations.
The right data is not a data lake. It's a map of your mission for the AI agent to reason against when querying data on your behalf.
When you define that map clearly, you can finally let AI accelerate the mission instead of overwhelming it.
IBIS™ is built around this principle: mission-defined ontologies, structured analytic techniques, and agentic AI working together to accelerate fusion across sensors and domains.
Our 8-week Pilot Partnership Program lets you test IBIS™ with:
See how chat-based queries + mission-derived context + AI governance eliminates the tradeoff between speed and accuracy with IBIS™.
Schedule a demo
Send us a message and we’ll get in touch shortly.
