Back to case study
Business case · Decision Platforms

Business case: a standing decision assessment platform

Reference engagement: Andrew W. Marshall Foundation

Executive summary

For roughly the cost of one consulting engagement of comparable scope, Syntheos embeds Deepfield and we keep the standing system our team operates from then on. Each subsequent assessment runs against the graph the first one built, at the marginal cost of compute. Every recommendation ships with the source passage that produced it. The platform was developed with the Andrew W. Marshall Foundation. Marshall's analytic disciplines from the Office of Net Assessment encode directly as runnable pipeline stages, and the platform has been generalized to handle strategic assessment ontologies beyond defense.

The problem

Every consulting engagement starts at zero. The first month is the analyst learning our domain, our sources, our adversaries, our prior decisions. We pay for that month every time. The deliverable is a frozen snapshot that goes stale the day a key input changes, and we can't interrogate any conclusion down to the document that produced it. When a reviewer challenges a number on page four, we're calling the consultant back. When the world shifts, we start over.

Proposed engagement

Twelve to sixteen weeks. Syntheos configures Deepfield for our domain, ingests our sources, sets credibility priors with our subject matter experts, tunes the reasoning weights, and runs the first real assessment alongside our team. Our analysts are in the loop at every stage. By week sixteen the system runs end-to-end on a strategic question we choose, the knowledge graph is built, our team has run the pipeline at least once with Syntheos out of the room, and we own the deployed code, the graph, and the data.

What we get

  • Decisions made in days after the first deployment. Subsequent assessments run against the existing graph for the marginal cost of compute, with our team operating the pipeline.
  • Provenance ready before the challenge arrives. Every recommendation ships with the source passage behind it, so a reviewer's question is answered by opening the entry.
  • An asset that compounds. The graph built during the first assessment is the foundation for every subsequent one. New sources extend it. New questions interrogate it from new angles.
  • Independence from the vendor. Code, pipeline, graph, and data run on our infrastructure. If Syntheos is unavailable for any reason, our team continues to operate the system.
  • An assessment our team understands. The analytic disciplines (Heuer, Marshall, Mason and Mitroff, RAND directed-collection) run as inspectable pipeline stages, and any analyst on the team can trace what each one does and why.

Risks and mitigations

  • What if Syntheos disappears. The deployed code, the pipeline, the knowledge graph, and the data live on our infrastructure under our license. The system runs without Syntheos personnel.
  • What if a recommendation is wrong. Every recommendation traces back to the source passage that produced it, and the chain shows the contradicting evidence the reasoner considered before issuing a verdict. Our team's defense is the chain itself.
  • What if our team can't run it without Syntheos in the room. Training is part of scope, and the engagement does not close until our team has run a complete assessment with Syntheos out of the loop.
  • What if the data classification changes mid-engagement. Deepfield supports air-gapped deployment, including environments where source material cannot leave a controlled boundary.
  • What if we want to leave. License terms allow our team to fork and operate independently, and we own the graph and the data outright.

Success metrics

  • First strategic assessment delivered in week sixteen, on a question that matters to our team.
  • Our team runs the second assessment with Syntheos out of the room and ships the output independently.
  • A reviewer challenges a number. Our team opens the entry behind it and reproduces the answer in real time.
  • Twelve months after deployment, the knowledge graph supports cross-assessment questions our team did not anticipate at the start.

Investment and timeline

Twelve to sixteen weeks for the first deployment. Cost varies with the depth of the question, the size of the source corpus, and the security environment, so a single public range would be misleading. A scoping conversation produces a fixed first-year number before any contract is signed. For most engagements that first-year investment lands close to the cost of one consulting engagement of similar scope. The second assessment runs against the graph the first one built, at the marginal cost of compute, which is when the system pays for itself. Annual hosting and support is a small percentage of the first-year cost. We can self-host or use Syntheos-managed infrastructure depending on classification and operational preferences.

Recommended next step

Send Syntheos a strategic question from our domain. Within two weeks, they'll run it through Deepfield's first three stages (charter, intake, and research) and walk us through the output. There is no fee. By the end of the two-hour walkthrough, we'll know whether the system produces material we trust.

Syntheossyntheos.io