Defence Innovation | Force Centric Capability Management (FCCM) Initiative | D3IP Case Study
Can AI Take the Burden Out of Army Capability Planning?
The British Army asked whether AI could reduce the resource burden of producing and maintaining Force Centric Capability Management artefacts. D3IP ran a structured process to find out.
What if AI could handle some of the most resource-intensive work in defence capability planning? That was the challenge the British Army sought to solve using D3IP’s Innovation as a Service model and collaboration network.
Can AI & ML Reliably Automate Capability Planning?
Context: What is Force Centric Capability Management?
Force Centric Capability Management (FCCM) is the process through which the British Army maintains a coherent, structured picture of its force elements; the units, systems, and capabilities that constitute its fighting force, and how they relate to one another. That picture is expressed through a structured set of documents, templates, and outputs: FCCM artefacts.
Producing and maintaining FCCM artefacts is time-consuming and demands significant analytical resource. The Army’s challenge: could AI and machine learning (ML) reduce that burden and could it do so reliably, securely, and in a way that produced outputs a planner could actually trust, without compromising security, traceability, or operational relevance?
The initiative set out to answer three questions…
Can this be done within MOD-appropriate secure environments, and within clear IP arrangements?
What does a credible path from proof of concept to operational service actually look like?
A Collaboration Between the Army and Industry
The Discovery phase was designed to surface novel insights from the industry respresentatives (a mix SMEs and existing large scale Defence suppliers) within the D3IP Community and give the Army a first-hand view of the options available before looking to develop a proof of concept.
Phase One: Challenge Statement & Industry Response
D3IP issued a detailed challenge statement supported by a live briefing and webinar Q&A, giving all participants a consistent understanding of the problem. Industry was invited to respond, assessed against five criteria: alignment with FCCM objectives, technical and security feasibility, usability, evidence of capability, and commercial considerations.
Thirteen expressions of interest were received from industry and three hurdles quickly emerged:
- Data access and classification: How AI & ML tools can securely access and process Official data within MOD-appropriate environments without creating unacceptable aggregation risk or blurring classification boundaries.
- Tooling and hosting: The challenge of finding platforms that meet Ministry of Defence (MOD) security standards while still supporting agile development. Commercial AI tools often assume open, cloud-connected environments that are incompatible with defence requirements.
- Induction and onboarding: The practical challenge of establishing the protocols, governance arrangements, and knowledge transfer needed to bring industry partners into sensitive work effectively.
Phase Two: Review, Shortlisting & Presentations
D3IP’s technical specialists reviewed all 13 responses before passing them to the Army stakeholder panel for assessment. Seven organisations were invited to a presentation day at the Defence BattleLab, Dorset Innovation Park, each given 45 minutes to present their approach, demonstrate existing solutions, and take questions from a combined panel of 10 Army representatives, including colleagues from the Army AI Centre.
Following the presentations and a stakeholder discussion, two participants were selected to proceed to a funded Incubate phase: Hpad, Inc. and Mission Decisions.
Testing the Art of the Possible
Both selected suppliers were commissioned to independently generate the same real FCCM artefact, a Level 1 Concept of Operations (CONOP), end to end, with defensible traceability back to source documentation, within a bounded proof of concept window. Each used a distinct architectural approach.
-
Approach A (Knowledge-First): Building a structured, persistent model of the domain first — entities, relationships, and expert-defined reasoning — then generating artefacts from that model rather than directly from raw documents. Well suited to durable, reusable knowledge infrastructure.
-
Approach B (Data-First): Chaining a sequence of automated processing steps directly from source documents through to structured, traceable outputs, with each step’s output feeding the next. Well suited to fast, automated generation at scale.
Both suppliers operated within secure, MOD-appropriate environments, and both systems were explicitly designed to refuse to generate content outside their source documentation — a key safeguard against hallucinations.
What the Incubate Phas Revealed
Both suppliers successfully generated an indicative Level 1 CONOP with end-to-end traceability from source documentation through to structured output. Here is what they found along the way:
-
A key constraint is the accessibility of the source documents. Both suppliers independently identified mixed terminology, incomplete examples, and inaccessible documents as the primary barriers to high-quality AI output. Investment in making source documentation more consistently accessible would have a directly proportionate effect on AI output quality.
-
The two approaches are complementary, not competing. The knowledge-first approach builds durable, reusable knowledge infrastructure; the data-first approach enables fast, automated generation at scale. A combined approach was identified as the strongest path forward.
-
Subject matter expert input is the critical path. Without embedded defence expertise reviewing and calibrating outputs, results risk being coherent and traceable, but incorrect in operationally significant ways. Expert involvement is not optional.
-
There is value beyond the artefact itself. Both approaches surfaced terminology inconsistencies, information gaps, and conflicting policy direction across source material faster and more comprehensively than manual review could. This secondary benefit emerged as significant in its own right.
-
Security and IP arrangements are workable. Both suppliers demonstrated they could operate within MOD-appropriate secure environments with clear intellectual property arrangements, reducing a common barrier to further adoption of AI-enabled tools in Defence.
A Model Built for this Kind of Problem
The FCCM Initiative demonstrates how D3IP’s Innovation as a Service model works in practice, moving at pace from a complex challenge to working prototypes, without the cost and lead time of traditional procurement…
For SMEs and specialist organisations often put off by the complexity of traditional procurement, D3IP’s pre-procured environment removes those barriers, giving the D3IP Community a clear, low-friction route to engaging with Defence challenges and showcasing your capabilities.
Get Involved in Future Initiatives
The FCCM Initiative is one example of how D3IP is helping Defence explore the art of the possible with AI. Future initiatives are your opportunity to engage directly with Defence challenges and showcase your capabilities.
Discover more from D3IP…
Synthetic Data in Defence: Faster Innovation Without the Security Risk
What if Defence could experiment, train AI models, and test new systems without ever risking classified data? D3IP’s Synthetic Data Production Initiative, delivered in partnership with the Defence BattleLab, set out to answer exactly that question with the help of our SME community. Here’s what we found, and how you can get involved in future initiatives.
Assured C2: Building Communications That Survive The Fight
When jamming, cyber-attack, and area-denial can sever a signal in an instant, ensuring you have command and control (C2) capabilities that can survive the fight is essential. But that doesn’t boil down to developing one clever technology – rather than a silver bullet, you need a resilient and adaptable ecosystem.