Control boundary
Proprietary research and institutional context can cross organizational boundaries as AI usage expands.
Institution-controlled AI for asset managers and asset owners
Control the data.
Control the models.
Control the outcome.
ZTA Labs helps asset managers and asset owners deploy governed AI using open-weight models, institution-controlled infrastructure, and workflows designed around investment research, Human Judgment, and measurable outcomes.
Client Data + Research
ZTA AI Control Plane
Apis + Model Council
Human Judgment
Recommendation + Outcome
Built for institutional investors
Asset Managers · Hedge Funds · Pension Funds · Insurance · Reinsurance
The model is replaceable. The institution’s intelligence is not.
Why the architecture has to change
AI adoption is moving faster than institutional control.
Asset managers are adopting AI across research and investment workflows, but most implementations still depend on external model providers, disconnected governance, and systems that do not preserve how Human Judgment was reached.
Auditability
Model output is often difficult to reproduce, challenge, and trace back to the evidence and controls that shaped it.
Institutional memory
AI-generated knowledge can remain trapped inside tools instead of becoming durable institutional intelligence.
Model dependency
Performance and economics remain tied to external providers even as better models appear at an accelerating pace.
The problem is not access to AI. It is building an operating environment the institution can govern.
A governed AI operating environment
The model is replaceable. The institution’s intelligence is not.
ZTA separates the institution’s data, governance, workflows, and accumulated intelligence from any single model provider. Models can change as better ones emerge without requiring the institution to rebuild its AI operating environment.
Control
Sensitive institutional data and research remain inside the institution’s chosen control boundary.
Governance
Models, evidence, policies, Human Judgment, and decisions remain traceable.
Economics
AI usage does not have to create perpetual third-party token dependency.
Model Independence
Better models can be introduced without redesigning the surrounding institutional architecture.
Institution-controlled architecture
One governed structure. Multiple AI capabilities.
Data, models, compute, agents, governance, and Human Judgment should not operate as disconnected technology layers. ZTA brings them into one institution-controlled architecture designed to share capability without creating gaps in control.
Shared structure.
Replaceable components.
Institution-owned intelligence.
Institutional
IntelligenceThe durable asset.
IntelligenceThe durable asset.
Shared-wall architectureHover, focus, or tap a capability to explore how it contributes to Institutional Intelligence.
Maximum capability. Minimum structural waste. No gaps in control.
From research to decision
Turn model output into governed institutional judgment.
ZTA connects research, model analysis, Human Judgment, recommendation management, and outcomes into one governed workflow.
ResearchApis / Model CouncilInsightsThesisRecommendationClearancePM DecisionPortfolio ActionOutcome
Insights are provisional.Model output remains separate from the analyst’s thesis until intentionally promoted.
Human Judgment is authoritative.Apis guides. Model Council challenges. The investment professional decides.
Lineage survives the workflow.Evidence, models, policy, rationale, PM disposition, action, and outcome remain connected.
Research Management · Credit deterioration assessment
Assessment
Does recent financial and operating evidence warrant changing the internal credit view from Stable to Deteriorating?
Evidence set: 14 approved sources.
Model Council
Response A · 4.10
Evidence supports a cautious view but stops short of a rating change.
Response B · 4.42
Greater weight to deteriorating liquidity and covenant headroom, with a clearer downside-first framing.
Response C · 4.55
Evaluator recommendation. Strong on evidence coverage, less aligned to the firm’s risk posture.
Response D · 3.91
Underweights the change in operating flexibility and overstates the stabilizing factors.
Governed model diversity
One model should not become the institution’s point of failure.
For elevated-risk research questions, ZTA can route analysis through a governed Model Council. Multiple approved open-weight models analyze the same authorized evidence independently, an evaluator compares the responses, and the investment professional makes the final judgment.
Model AModel BModel CModel D
Independent analysis
same approved evidence
same approved evidence
EvaluatorCompares evidence, reasoning, strengths & disagreement
Recommendation
not decision
not decision
Human JudgmentSelect · edit · combine · reject
Independent model analysisCommon approved evidenceDisagreement surfaced explicitlyFull decision lineage
The evaluator recommends. The investment professional decides.
For investment leaders
Investment leaders should be able to answer five questions.
Traditional research systems rarely connect analyst judgment, portfolio implementation, research quality, and realized value in one view. ZTA does.
Who is adding value?
Analyst Ranking · Active Return · Net Value Add · Analysts Outperforming
Why?
Methodology-specific attribution · top and bottom contributors · Apis explanation
Is the research reaching the portfolio?
PM Alignment · adoption and modification rates · Implementation Shortfall
Is the research process itself rigorous?
Research Quality Score · Evidence Support · Source Quality · Completeness · Disconfirming Evidence
What value is the organization ultimately capturing?
CIO Portfolio Value Creation · aggregate Implementation Shortfall · asset-class contribution
Who → Why → Adoption → Quality → Captured Value
Measure what the research actually contributed
Separate analyst insight from portfolio implementation.
ZTA measures analyst paper-portfolio performance against approved benchmarks and compares it with actual portfolio outcomes. Common performance metrics can be aggregated while methodology-specific attribution effects remain separate.
Paper Portfolio Return8.7%
Benchmark Return6.9%
Active Return+1.8%
Implementation Shortfall-0.4%
Net Value Add+1.4%
Equities
Brinson-Fachler
Allocation · Equity Selection · Interaction
Fixed Income
Campisi
Income · Treasury · Spread · Fixed Income Selection
Aggregate the performance. Preserve the methodology. Let Apis explain the story.
Make model selection evidence-led
Generic benchmarks do not tell you which models are fit for your workflows.
ZTA Labs Model Assurance evaluates open-weight models against your data, your document classes, your question types, and your investment workflows—not generic public leaderboards.
Baseline Model Assurance2 weeks
Establish the evidence before production.
Build a client-specific calibration corpus, benchmark candidate open-weight models, test reliability and evaluator integrity, assess Model Council contribution, and establish the baseline for future comparison.
Starting at $15,000when purchased standalone
- 20+ page Model Assurance Baseline
- Model-level strengths and failure modes
- Recommended Model Council configuration
- Production and governance recommendations
Quarterly Model AssuranceEvery quarter
Keep model selection current.
Re-benchmark current and candidate models, compare against the baseline, evaluate new versions, review Model Council composition, and incorporate production Human Judgment and outcomes where available.
- Quarterly re-benchmarking of open-weight models
- New model and version review
- Reliability, evaluator, and drift analysis
- Recommended model and governance changes
Sample Model Assurance Baseline
See the methodology in practice.
A governed model assessment with model-level findings, failure analysis, source-level auditability, governance requirements, and deployment recommendations.
| Candidate | Overall | Hard fails | Model Council |
|---|---|---|---|
| gpt-oss 120B | 96.4 | 1 | Admitted |
| Qwen3 30B A3B | 89.7 | 3 | Admitted |
| Mistral Small 3.1 24B | 87.2 | 4 | Admitted |
| DeepSeek-R1-Distill-Qwen 32B | 84.1 | 6 | Conditional |
Illustrative engagement artifact. Admission is based on governed qualification criteria, not a public leaderboard.
Question
What was gross margin on a GAAP basis in the quarter?
Model answered
“50.4 percent” with a citation that resolves correctly to the earnings release.
Correct answer
50.3 percent. 50.4 percent is the non-GAAP figure three lines below on the same page.
Finding
Correct source. Wrong accounting basis. The error survives casual review precisely because the citation is correct.
The benchmark stays with the institution.
A practical path to production
Start small. Prove the architecture. Scale with evidence.
ZTA engagements are designed to move from a bounded first workflow to a production environment with clear scope, timing, and commercial expectations.
Path 1
ZTA Architecture
Scope30–60 minutes
Define the workflow, data, governance requirements, and success criteria.
1-Week ZTA Architecture Trial1 week
Prove the architecture on a defined workflow using representative client data.
Trial + Baseline Model AssuranceStarting at $15,000*
Package the 1-week Architecture Trial with the two-week Baseline Model Assurance assessment.
Production ArchitectureTypical: 2–4 weeks
Design data, model, governance, identity, and integration architecture.
Build & ValidateTypical: 6–12 weeks
Implement and validate the first governed production workflow.
ProductionOngoing
Operate inside the institution’s chosen control boundary.
Quarterly Model AssuranceEvery quarter
Re-benchmark models, review production evidence, and evolve the environment.
Path 2
Standalone Model Assurance
Scope30–60 minutes
Define the workflows, document classes, candidate models, and evaluation objectives.
Baseline Model Assurance2 weeks · Starting at $15,000*
Benchmark open-weight models against the client’s data and workflows and deliver the Model Assurance Baseline.
Quarterly Model AssuranceEvery quarter
Re-benchmark the model pool and keep the evidence current.
*Starting pricing reflects a defined scope. Final pricing depends on the number of workflows, data sources, models, integrations, and governance requirements. ZTA Labs provides a fixed-price proposal before the engagement begins.
Built on measurable controls
Evidence before claims.
ZTA Labs is built around documented, versioned methodologies rather than vendor assertions or opaque AI scoring.
Client-specific Model Assurance
Models are tested against institutional workflows, not only generic benchmarks.
Evaluator assurance
ZTA tests the instrument used to judge models, not just the models themselves.
Versioned methodology
Model Assurance, attribution, and governance methodologies are designed to remain documented and auditable.
Human Judgment captured
What investment professionals choose—and why—becomes part of the institutional record.
Decision lineage
Evidence, model outputs, policies, Human Judgment, PM disposition, and outcomes remain traceable.
Built to work with the institution’s existing environment
Open architecture. Open ecosystem.
ZTA Labs is designed to integrate with existing institutional data platforms, identity systems, infrastructure, research repositories, and investment systems.
ZTA AI
Control Plane
Enterprise Data ManagementInfrastructure & HardwareOpen-Weight ModelsSystems IntegratorsResearch RepositoriesInvestment Systems
Control Plane
Open-source first. Vendor-flexible by design. Institution-controlled by default.
The Buzz · ZTA Labs Research
Thinking about what comes next in institutional AI.
Research and perspectives on models, architecture, governance, AI economics, Investment Research, and institutional intelligence.
Open-Weight AI Has Crossed the Enterprise Threshold
The capability gap has not disappeared. The deployment decision has changed.
Public evidence shows that open-weight AI has become a credible enterprise architecture option. But leaderboard performance cannot establish whether a model is fit for an institution’s work.
Earlier: The Seven Steps to AI Independence →
Start with one workflow
See whether ZTA fits your environment.
Define a specific investment workflow, data boundary, and governance requirement. ZTA Labs will help determine the right first step—Architecture Trial, Model Assurance, or both.