Self-Assessment

ICAD Maturity Assessment

Evaluate where your organization stands on the Integrate → Contextualize → Analyze → Decide sequence

Integrate

0 / 12

I1. How does instrument data enter your governed data pipeline?

Automatically captured at the point of creation — no scientist intervention required 3 pts

Semi-automated — instruments write to a watched folder, but scientists must trigger export 2 pts

Batch upload — scientists manually export and upload data periodically 1 pt

No governed pipeline — data stays in instrument PCs or local drives 0 pts

I2. How is original instrument data preserved?

Native files stored immutably with cryptographic hash, alongside converted queryable representations 3 pts

Native files archived but without hash verification or converted representations 2 pts

Data is converted to a standard format, but native files are not systematically preserved 1 pt

Only processed results are saved — original instrument files are routinely overwritten or lost 0 pts

I3. How are new instrument integrations built?

Factory model — reusable converters, adapters, and validation modules; each build compounds the asset library 3 pts

Template-based — some reuse across integrations, but significant custom work per build 2 pts

Project-based — each integration is a standalone effort with limited reuse 1 pt

No systematic integration — point-to-point connections or manual data transfer 0 pts

I4. How long does it take to onboard a new instrument type (with existing technique family)?

Days — configuration of existing infrastructure, not engineering 3 pts

Low single-digit weeks — mostly reuse with some customization 2 pts

1–3 months — significant development required 1 pt

3–6+ months or instruments operate outside the governed pipeline indefinitely 0 pts

Contextualize

0 / 12

C1. Can you trace any data point to its method, sample, study, program, and regulatory submission?

Yes — links navigable by scientists and machines from measurement through method, sample, study, program, to submission 3 pts

Partially — some links exist but require manual lookup for full chain 2 pts

Convention-based — context is embedded in file names or folder structures, not explicit links 1 pt

No systematic linking — context lives in scientists' knowledge or ELN narratives 0 pts

C2. If you query "all data for compound X," do results include every site regardless of local naming?

Yes — canonical identity layer maps all local synonyms; cross-site queries are seamless 3 pts

Partially — some entities reconciled, but gaps exist across sites or systems 2 pts

Manually maintained — spreadsheets or lookup tables map names, but not systematically enforced 1 pt

No reconciliation — each site/system uses its own naming; cross-site queries require manual effort 0 pts

C3. Can you trace any value in a regulatory submission back to its original data source without human explanation?

Yes — complete DAG from instrument measurement through every transformation to decision, bidirectional 3 pts

Partially — lineage exists for some steps but has gaps requiring human knowledge 2 pts

Documentation-based — transformations are described in SOPs but not tracked by the system 1 pt

No systematic lineage — tracing requires interviewing analysts who did the work 0 pts

C4. Can a system (no human) query: "All stability data for compound X, condition Y, method Z, across all sites, last 24 months"?

Yes — structured schemas, controlled vocabularies, and APIs serve context programmatically 3 pts

Partially — some parameters queryable, but method details or conditions require opening documents 2 pts

Semi-structured — data is in databases but scientific context is in unstructured fields 1 pt

No — answering this query requires a human to open PDFs, decode spreadsheets, and compile results 0 pts

Analyze

0 / 12

A1. Does your analytical layer enforce that all inputs have passed through contextualization?

Yes — architectural gate prevents analysis on un-contextualized data 3 pts

Policy-based — guidelines exist but the system does not enforce the gate 2 pts

Ad hoc — some analyses use contextualized data, others pull from raw sources 1 pt

No gate — analysts routinely query raw databases, CSV exports, or file shares directly 0 pts

A2. When you compare data across programs or sites, are governance rules explicit and auditable?

Yes — governed comparison framework with declared inclusion/exclusion criteria, method equivalence verification, and statistical model 3 pts

Partially — some governance exists but rules are embedded in analyst scripts, not declared 2 pts

Informal — senior scientists review cross-program analyses, but no formal framework 1 pt

No governance — cross-site data is pooled without method equivalence verification 0 pts

A3. For any analytical or AI model, can you retrieve the exact training dataset, algorithm, parameters, and reproduce the output?

Yes — model registry with versioned training data (full lineage), algorithm, parameters, performance metrics 3 pts

Partially — models are documented but training data is not frozen/versioned with full lineage 2 pts

Documentation only — model descriptions exist in reports but are not machine-traceable 1 pt

No systematic tracking — models are developed ad hoc without provenance infrastructure 0 pts

A4. If two analysts at different sites run the same analysis on the same governed dataset version, do they get identical results?

Yes — deterministic: same dataset version + same parameters = same output, every time 3 pts

Usually — minor differences due to software versions or replication lag, but results are close 2 pts

Sometimes — results vary because analysts apply different filters or data selections 1 pt

Rarely tested — reproducibility across sites is not systematically verified 0 pts

Decide

0 / 12

D1. Do your AI/decision-support systems operate exclusively on data that has passed through the full I→C→A sequence?

Yes — architectural enforcement prevents AI from accessing un-governed data 3 pts

Policy-based — guidelines require governed inputs, but no technical enforcement 2 pts

Mixed — some AI systems use governed data, others consume raw or semi-structured data 1 pt

No gate — AI/ML models are trained on whatever data is available, regardless of governance status 0 pts

D2. When your AI system produces a recommendation, can you trace it through the full I→C→A provenance chain?

Yes — system-recorded provenance chain: recommendation → analysis → contextualized data → instrument measurement 3 pts

Partially — some traceability, but requires reverse-engineering from output 2 pts

Explainability only — post-hoc methods (SHAP, attention maps) explain the output, but actual provenance is not recorded 1 pt

Black box — AI recommendations cannot be traced to source data 0 pts

D3. Are autonomy levels explicitly configured per decision type?

Yes — each decision type has a documented, governed autonomy level tied to regulatory classification and risk 3 pts

Partially — some decision types have defined autonomy, but many default to ad-hoc human review 2 pts

Binary — all AI outputs are either fully manual review or fully autonomous, with no spectrum 1 pt

No autonomy framework — AI outputs are treated uniformly without risk-based configuration 0 pts

D4. Do decision outcomes systematically feed back to improve integration, context, and analysis?

Yes — structured feedback channels: D→I (missing data sources), D→C (context gaps), D→A (model refinement) 3 pts

Partially — some feedback occurs but is not systematic or tracked 2 pts

Reactive — feedback happens only when major failures occur 1 pt

No feedback loop — decision outcomes are not used to improve upstream layers 0 pts

Your ICAD Maturity

Answer the questions to see your ICAD maturity score

0% Pre-ICAD

Integrate 0%

Contextualize 0%

Analyze 0%

Decide 0%

Pre-ICAD

Your organization has not yet established the foundational infrastructure for governed scientific data operations. Most data remains siloed in instrument PCs and local systems.

Industry Benchmark

Help Build the First ICAD Maturity Benchmark

We are building an anonymized industry benchmark report showing where pharmaceutical R&D organizations stand on the ICAD sequence. Your contribution helps the entire community understand maturity gaps and prioritize infrastructure investment.

Complete the assessment above to see your scores and contribute to the benchmark.

—

Company Name *

Company Website * Used to verify your organization — not published.

Your Role *

Organization Type *

Work Email *

Your individual scores are never published. Only aggregated, anonymized industry averages appear in the benchmark report. See our data privacy policy.

Ready to Advance Your ICAD Maturity?

Discuss your assessment results with a ZONTAL infrastructure expert and map a path from your current state to compounding scientific data operations.

Book a Strategy Session Explore the ICAD Principles