---

Abstract #

The Heuristic Prediction Framework for Pharma (HPF-P) defines decision readiness through two complementary constructs: the Decision Readiness Index (DRI), which quantifies information sufficiency for a given decision context, and the Decision Readiness Level (DRL), which measures organizational maturity in applying AI-augmented decision processes. While previous work formalized DRI as a continuous metric^[2], DRL has remained conceptually defined but not operationally specified. This article bridges that gap by proposing a five-level DRL maturity model adapted from established frameworks — NASA’s Technology Readiness Levels (TRL) and the Capability Maturity Model Integration (CMMI) — tailored specifically for pharmaceutical portfolio decision-making. We define assessment criteria, transition gates, and measurement instruments for each DRL level, and demonstrate how DRL complements DRI to provide a complete picture of decision readiness within the HPF-P framework.

graph TD
    A["HPF-P Framework"] --> B["DRI: Decision Readiness Index"]
    A --> C["DRL: Decision Readiness Level"]
    B --> D["Information Sufficiency
Continuous metric [0,1]"]
    C --> E["Organizational Maturity
Discrete levels 1-5"]
    D --> F["Complete Decision
Readiness Assessment"]
    E --> F
    style A fill:#2d3748,stroke:#4a5568,color:#e2e8f0
    style F fill:#2b6cb0,stroke:#2c5282,color:#e2e8f0

Introduction: The Missing Half of Decision Readiness #

The HPF-P framework, as introduced in its foundational formulation^[3], posits that decision readiness in pharmaceutical portfolio management is a function of two orthogonal dimensions. The first — the Decision Readiness Index — captures whether sufficient information exists to make a particular decision. A DRI approaching 1.0 indicates that the data landscape is rich enough, the uncertainty bounds narrow enough, and the analytical outputs reliable enough to support a specific portfolio action.

But information sufficiency alone does not guarantee good decisions. An organization may possess all necessary data yet lack the processes, governance structures, or analytical capabilities to translate that information into effective portfolio actions. This is the domain of the second dimension: the Decision Readiness Level.

The concept of staged maturity assessment has deep roots in engineering and management science. NASA’s [4], first developed in the 1970s, provided a systematic method for evaluating the maturity of technologies before integration into operational systems. The software engineering community adapted this thinking through the AI-specific readiness frameworks^[5] that account for the unique challenges of machine learning systems — data dependency, model drift, and explainability requirements.

In the pharmaceutical domain specifically, clinical development trends in 2026^[6] reveal growing emphasis on readiness-based frameworks for integrating AI into drug development decisions. The challenge is not merely technical but organizational: teams must build capacity for protocol design optimization, adaptive trial strategies, and AI-informed portfolio rebalancing.

DRL synthesizes these traditions into a pharmaceutical-specific maturity model that captures an organization’s capacity to operationalize AI-augmented decision processes in portfolio management contexts.

Theoretical Foundations of DRL #

Adapting Maturity Models for Decision Contexts #

Traditional maturity models assess process capability: can an organization reliably execute a defined process? DRL extends this by assessing decision capability: can an organization reliably translate analytical outputs into portfolio actions under uncertainty?

This distinction matters because pharmaceutical portfolio decisions operate under conditions fundamentally different from typical software engineering projects. Drug development portfolios involve:

• Irreversible resource commitments — clinical trial investments cannot be recovered if a compound fails
• Extreme temporal asymmetry — decisions made today produce outcomes years or decades later
• Regulatory coupling — portfolio decisions interact with regulatory pathways that constrain future options
• Information cascades — a single Phase II failure can trigger reassessment of an entire therapeutic area

These characteristics mean that decision maturity in pharma requires not just process discipline but epistemic discipline — the organizational capacity to reason correctly about uncertainty, update beliefs appropriately, and act decisively when DRI thresholds are met^[2].

graph LR
    subgraph "Traditional Maturity (CMMI)"
        T1["Process
Capability"] --> T2["Repeatable
Execution"]
    end
    subgraph "DRL Extension"
        D1["Decision
Capability"] --> D2["Epistemic
Discipline"]
        D2 --> D3["Portfolio
Action Quality"]
    end
    T2 -.->|"informs"| D1
    style D3 fill:#2b6cb0,stroke:#2c5282,color:#e2e8f0

The DRI-DRL Complementarity #

DRI and DRL operate on orthogonal axes. Consider a 2×2 matrix:

• High DRI, High DRL: Ideal state — sufficient information and mature processes. Portfolio decisions are well-informed and systematically executed.
• High DRI, Low DRL: Data-rich but process-poor. The organization has the information but lacks the governance, analytical infrastructure, or decision protocols to use it effectively.
• Low DRI, High DRL: Process-mature but information-starved. Excellent decision infrastructure exists, but the underlying data and models are insufficient. This occurs in emerging therapeutic areas where organizational capability outpaces available evidence.
• Low DRI, Low DRL: Both dimensions deficient. Common in early-stage AI adoption where neither the analytical outputs nor the organizational processes have matured.

The experimental validation of HPF-P^[7] demonstrated that portfolio optimization quality correlates with both DRI and organizational adoption factors, supporting the theoretical need for a dual-metric approach.

The Five DRL Levels #

Drawing on CMMI’s five-level structure and adapting it for pharmaceutical decision contexts, we define DRL as follows:

DRL 1: Ad Hoc Decision-Making #

At this level, portfolio decisions are made through informal processes. Individual judgment drives resource allocation, with limited systematic use of analytical tools. AI models may exist in isolated research environments but are not integrated into decision workflows.

Characteristics:

• Decisions depend on individual expertise and intuition
• No standardized decision protocols for portfolio rebalancing
• AI outputs, if generated, are advisory footnotes rather than integral inputs
• Decision quality varies dramatically across teams and therapeutic areas

Assessment criteria: Decision processes are undocumented; no formal link between analytical outputs and portfolio actions.

DRL 2: Structured Decision Processes #

The organization has established repeatable processes for key portfolio decisions. Decision criteria are documented, and analytical inputs are formally required at defined decision gates.

Characteristics:

• Documented decision frameworks for go/no-go decisions
• AI models are consulted at defined checkpoints
• Standard templates for decision rationale documentation
• Basic metrics track decision outcomes retrospectively

Transition gate from DRL 1→2: Documented decision protocols exist for ≥80% of portfolio-level decisions; at least one AI-generated input is mandatory at each stage gate.

DRL 3: Quantitative Decision Integration #

AI-generated analytics are quantitatively integrated into decision processes. DRI scores are computed and used as formal inputs. Decision-makers can articulate confidence intervals and uncertainty bounds on portfolio projections.

Characteristics:

• DRI computed for all major portfolio decisions
• Environmental entropy metrics^[8] inform risk assessment
• Quantitative thresholds trigger automatic escalation or review
• Decision audit trails include model versions, data snapshots, and parameter configurations

Transition gate from DRL 2→3: DRI is computed for ≥90% of decisions; decision audit trails capture model provenance; quantitative thresholds are defined for key decision types.

DRL 4: Predictive Decision Management #

The organization uses AI not just to inform individual decisions but to predict decision outcomes and optimize decision timing. Platform architecture supports^[9] real-time integration of market signals, clinical data, and regulatory intelligence.

Characteristics:

• Predictive models estimate decision outcome distributions
• Decision timing optimization — the system recommends when to decide, not just what to decide
• Cross-portfolio optimization considers interaction effects between compounds
• Platform infrastructure^[10] enables continuous model retraining and deployment

Transition gate from DRL 3→4: Predictive accuracy on portfolio outcomes exceeds retrospective human judgment by ≥15% (measured on held-out decisions); decision timing recommendations are generated and tracked.

DRL 5: Self-Improving Decision Systems #

The decision system exhibits autonomous improvement. Models detect their own degradation, request additional data collection, and propose process modifications. This level draws inspiration from the multi-strategy optimization approach validated in HPF-P experiments^[7], where the system itself selects and adapts analytical strategies.

Characteristics:

• Autonomous model monitoring with drift detection and self-correction
• The system proposes process improvements based on decision outcome analysis
• Continuous integration of new data sources without manual pipeline reconfiguration
• Decision processes evolve based on measured outcome quality

Transition gate from DRL 4→5: System demonstrates measurable self-improvement over ≥4 quarterly review cycles; autonomous model updates maintain or improve prediction accuracy without manual intervention.

graph TB
    L1["DRL 1: Ad Hoc"] -->|"Document processes"| L2["DRL 2: Structured"]
    L2 -->|"Integrate DRI"| L3["DRL 3: Quantitative"]
    L3 -->|"Predict outcomes"| L4["DRL 4: Predictive"]
    L4 -->|"Self-improve"| L5["DRL 5: Self-Improving"]
    
    L1 --- C1["Individual judgment
No AI integration"]
    L2 --- C2["Documented protocols
AI at checkpoints"]
    L3 --- C3["DRI-driven decisions
Quantitative thresholds"]
    L4 --- C4["Outcome prediction
Timing optimization"]
    L5 --- C5["Autonomous improvement
Drift detection"]
    
    style L5 fill:#2b6cb0,stroke:#2c5282,color:#e2e8f0
    style L4 fill:#2c5282,stroke:#2a4365,color:#e2e8f0
    style L3 fill:#2a4365,stroke:#1a365d,color:#e2e8f0

Measurement Instruments #

The DRL Assessment Matrix #

Operationalizing DRL requires concrete measurement. We propose a multi-dimensional assessment matrix with six capability domains:

1. Decision Process Formalization — Are decision protocols documented, versioned, and enforced?
2. AI Integration Depth — How deeply are AI outputs embedded in decision workflows?
3. Uncertainty Quantification — Can the organization articulate and act on uncertainty bounds?
4. Decision Audit Capability — Can past decisions be reconstructed with full provenance?
5. Feedback Loop Maturity — How effectively do decision outcomes inform future processes?
6. Adaptive Capacity — Can the organization modify its decision processes in response to environmental changes?

Each domain is scored on a 1–5 scale aligned with the DRL levels. The overall DRL is the minimum across all six domains (following CMMI’s weakest-link principle), ensuring that advancement requires balanced capability development.

Scoring Protocol #

Assessment involves structured interviews with portfolio decision-makers, review of decision documentation artifacts, and analysis of historical decision outcome data. The scoring protocol requires:

• Evidence-based scoring — each level claim must be supported by documentary evidence
• Cross-functional validation — assessments include perspectives from R&D, commercial, regulatory, and data science teams
• Temporal stability — a level is only awarded if the capability has been demonstrated consistently over ≥2 decision cycles

This methodology draws on established assessment practices from contemporary AI maturity frameworks^[11] while adapting them for the specific epistemological requirements of pharmaceutical portfolio management.

Implementation Considerations #

The Pharmaceutical Industry’s Current Position #

Based on the emerging landscape of pharmaceutical AI adoption in 2026, most large pharmaceutical organizations currently operate at DRL 2 or early DRL 3. They have structured decision processes and are beginning to integrate AI analytics, but few have achieved the quantitative decision integration that defines mature DRL 3 operation.

The biotech sector’s experience with AI maturity^[12] reinforces this assessment: organizations are learning that AI maturity looks less like model performance benchmarks and more like sustained operational capability — the ability to keep systems stable, auditable, and useful across programs.

Mid-size pharmaceutical companies often find themselves at DRL 1–2, with pockets of DRL 3 capability in specific therapeutic areas where data science investment has been concentrated. The challenge for these organizations is achieving consistent capability across the portfolio rather than isolated excellence.

Advancement Strategies #

Moving between DRL levels requires different investment priorities:

DRL 1→2 requires primarily organizational change: documenting existing decision processes, establishing stage gates, and mandating analytical inputs. The investment is largely in governance and process design rather than technology.

DRL 2→3 requires technical infrastructure: implementing DRI computation, building decision audit systems, and deploying quantitative threshold frameworks. This is where the HPF-P platform architecture^[9] provides its primary value, offering a reference implementation for quantitative decision integration.

DRL 3→4 requires predictive modeling capability and the organizational willingness to act on predictions about decision outcomes — a significant cultural shift that demands executive sponsorship and demonstrated value from DRL 3 operations.

DRL 4→5 requires autonomous system capability and, critically, organizational trust in autonomous processes. This level may not be achievable or desirable for all organizations; the risk profile of pharmaceutical decisions may warrant permanent human oversight of certain decision categories.

The Role of AI Portfolio Management Platforms #

Modern AI-powered strategic portfolio management platforms^[13] provide some of the technical infrastructure needed for DRL advancement. However, platform capability alone is insufficient — DRL assessment explicitly captures organizational and process maturity alongside technical capability.

The distinction matters: a DRL 2 organization with a DRL 4-capable platform will still make DRL 2-quality decisions. The platform enables advancement but does not cause it. This finding aligns with the broader medical software readiness literature^[14], which consistently finds that technology readiness and organizational readiness must advance together.

Integrating DRL with the HPF-P Lifecycle #

Within the complete HPF-P framework, DRL serves as the organizational readiness complement to DRI’s informational readiness. The framework’s deployment architecture^[10] should include DRL assessment as a standard component of system implementation, ensuring that organizational capability is evaluated alongside technical deployment.

The practical integration follows a feedback cycle:

1. Assess current DRL using the six-domain matrix
2. Identify the binding constraint (lowest-scoring domain)
3. Design targeted interventions for the binding domain
4. Implement and measure over ≥2 decision cycles
5. Reassess and identify the new binding constraint

This cycle mirrors the continuous improvement philosophy embedded in both CMMI and the self-improving nature of DRL 5, creating a path from wherever an organization currently stands toward increasingly mature decision capability.

Conclusion #

The Decision Readiness Level completes the dual-metric architecture of the HPF-P framework. Where DRI asks “do we have enough information to decide?”, DRL asks “are we organizationally capable of deciding well?” Together, they provide a comprehensive assessment of decision readiness that accounts for both the informational and organizational dimensions of pharmaceutical portfolio management.

The five-level DRL model, adapted from TRL and CMMI traditions for pharmaceutical decision contexts, offers a concrete assessment framework with defined transition gates and measurement instruments. Its weakest-link scoring methodology ensures balanced capability development, preventing organizations from claiming high maturity based on isolated pockets of excellence.

For HPF-P implementations, DRL assessment should be conducted at project initiation and reviewed quarterly, with advancement targets aligned to organizational strategic planning cycles. The framework’s value lies not in achieving the highest possible DRL, but in providing honest, evidence-based assessment of current capability and a structured path for improvement — because in pharmaceutical portfolio management, the cost of overestimating one’s decision readiness can be measured in failed drugs, wasted capital, and delayed therapies.