The portfolio average conceals the portfolio
A study can be tracking to plan while a third of its sites underperform, because the strong sites absorb the shortfall. The aggregate hides exactly the sites an operations team could still help.
Clinical Trial Supply Chain IntelligenceLife Sciences & Clinical Operations
Clinical Trial Site Performance Intelligence
How sites are executing now
A portfolio enrollment number is an average, and averages are where operational problems go to hide. Underneath one healthy-looking figure sit sites that activated late, sites carrying a retention problem, sites generating queries faster than a monitor can close them — and sites doing none of that, quietly carrying the study.
This experience is about the present tense. It ranks and profiles investigator sites on what they have actually done, indicates where dropout, protocol-deviation and data-quality exposure is accumulating, compares countries and regions, and runs operational scenarios — so a clinical-operations team can decide which sites to look at this week. Where enrollment is heading is a different question, and it belongs to a different experience.
Current Site ExecutionSite Ranking SupportOperational Risk IndicationData Quality & MonitoringSynthetic Data
Synthetic clinical-trial data onlyNo real patient dataSite-level aggregates, not patient recordsOperational, not medical or regulatoryHuman review required
The Premise
Why current site execution deserves its own view
Trial performance is uneven by nature. Sites differ in activation pace, staffing, patient access, retention and data discipline — and those differences are operational facts a team can act on this week, not projections to be debated.
Operational risk shows up before the number does
Rising query age, slipping data quality, an accumulating deviation pattern — these move before enrollment does. Read late, they become the explanation rather than the warning.
A ranked list is not a judgement
Ranking sites is useful precisely because it is provisional. It orders a review queue; it does not qualify an investigator, approve an activation, or decide anything about a site.
Geography concentrates exposure
Startup pace, monitoring burden and retention behave differently by country and region. Seeing that concentration is what turns a list of weak sites into a structural conversation.
Scope Within the Family
Current execution here, forward outlook next door
This is the family’s single owner of how sites are executing now. The question of where enrollment is heading — and how far to trust that answer — is a separate discipline with a separate owner.
This experience owns
- Site ranking and selection support
- Site profiles and site-level operational review
- Observed recruitment performance
- Current site execution signals
- Site-level dropout-risk indication
- Site-level protocol-deviation-risk indication
- Data quality and monitoring intelligence
- Country and region operational performance
- Clinical-operations scenario analysis
- Operational recommendations for human review
This experience does not own
- Forward enrollment forecasting methodology
- Rolling-origin forecast evaluation
- Baseline-versus-model forecast comparison
- Future enrollment trajectory as a primary capability
- Kit-demand translation and depot or site supply inventory
- Resupply, expiry and cold-chain supply planning
Enrollment appears on this page only as observed recruitment performance — a current execution signal used to explain how a site is doing.
Forward outlook — sibling experience
Clinical Enrollment Forecasting & Site Risk Intelligence
Future enrollment trajectory, forecast methodology, rolling-origin evaluation and forecast-risk interpretation belong there. The underlying prototype behind this page does contain an enrollment-prediction view, but within the portfolio family it is supporting context only — forecast ownership and methodological rigour sit with the dedicated sibling.
Downstream — where this eventually leads
Clinical Supply Planning & Resupply Intelligence
Observed site execution shapes the enrollment outlook, and that outlook eventually becomes a clinical-supply requirement. Kit demand, depot and site inventory, expiry, shipment and resupply are that experience’s domain and appear nowhere on this page.
Reading this page correctly
Everything shown is generated synthetic data. Study, site, investigator, country and region identifiers are synthetic labels. The dataset holds site-level aggregates — the underlying event records are rolled up before analysis, and the patient-population table contains site-level estimates such as eligible-population size and monthly patient flow rather than any patient record. Dropout and deviation risk are site-level operational indications, never statements about an individual. Nothing here is a medical, protocol, safety or regulatory judgement.
Decision Framework
Site performance decision journey
Nine stages, each tagged with what kind of thing it is. Almost all of them are present tense — that is the point of this experience.
- Stage 01Site portfolioThe whole synthetic site estate in one view — how many sites are active, how performance is distributed, and where operational exposure is concentrated.Current
- Stage 02Site ranking & selection signalSites ordered by a weighted composite of their observed execution, filterable by study, geography, site type, therapeutic area and risk band.Current
- Stage 03Site profileA single site opened up — type, status, prior trial experience, investigator context and staffing capacity, assembled from the underlying synthetic tables.Current
- Stage 04Observed recruitment performanceWhat the site has actually screened and randomised over its history, alongside its screen-failure trend. This is measured behaviour, not a projection.Observed
- Stage 05Dropout & deviation riskSite-level indication of where retention and protocol-deviation exposure is accumulating, with the contributing factors named.Indication
- Stage 06Data quality & monitoringQuery rate and age, missing data, source-data-verification issues and monitoring findings — the operational quality signal a monitoring team works from.Indication
- Stage 07Country & region contextWhere operational strength and weakness concentrates geographically, so a site-level pattern can be read as a structural one.Current
- Stage 08Operational scenarioA defined operational assumption applied to the current baseline, with the projected consequence held against that baseline rather than replacing it.Modelled
- Stage 09Intervention reviewRule-derived recommendations, prioritised and reason-carrying, handed to a qualified clinical-operations reviewer who decides what actually happens.Human
Functional Coverage
Twelve working screens, grouped by decision purpose
The prototype contains twelve screens running against one seeded synthetic estate. Six are presented as motion showcases below; all twelve are listed here, including the enrollment-prediction screen, which is shown in its correct place rather than hidden.
- AI Executive Command Center
- Site Ranking and Selection
- Site Profile Search
- Dropout Risk Intelligence
- Protocol Deviation Risk Intelligence
- Data Quality and Monitoring
- Country and Region Performance
- Clinical Operations Scenario Simulator
- AI Recommendations
- Enrollment Prediction (supporting context only)
- Synthetic Data Explorer
- AI Terminology Guide
Showcase 01 · Portfolio position
Clinical Site Executive Overview
Current estatePerformance distributionExposure concentration
Current / observed / modelled context
Current state across the whole synthetic site estate. Every figure describes where sites stand today, not where recruitment is heading.
Business question
Where is current operational site exposure concentrated, and where should the clinical-operations team look first?
Recorded from the running prototype on synthetic data. Site, investigator, country and study identifiers are generated labels; no real site, investigator or sponsor is represented.
Key signals on the screen
- Study and active-site counts alongside high- and low-performer splits across the estate
- Portfolio-level averages for dropout and deviation risk on a common 0–100 scale
- Site performance and site ranking distribution, so the shape of the estate is visible rather than a single mean
- Dropout risk broken out by region and deviation risk by site type, plus a top operational alerts list
Decision supported
Which slice of the estate — a region, a site type, or a specific risk band — deserves the first operational review.
Captured synthetic state
12 synthetic studies, 275 active sites of 300, 104 high performers against 99 low performers, 4 critical-risk sites, average dropout risk 47.8 and average deviation risk 33.5 on a 0–100 scale, 227 sites carrying critical or high recommendations, and a portfolio health score of 75.
Interpretation & human-review boundary
A triage view over synthetic data at a point in time. It carries no operational, clinical or regulatory standing, and it qualifies no site.
Showcase 02 · Ordering the review queue
Site Ranking & Selection Support
Weighted compositeNine declared componentsDecision support
Current / observed / modelled context
Current-state ranking built from observed execution. This orders a review queue — it is not a selection, qualification or activation decision.
Business question
Which sites appear operationally stronger or weaker under the modelled criteria?
Recorded from the running prototype on synthetic data. Ranking positions are modelled operational signals, not endorsements of any site.
Key signals on the screen
- Six independent filters — study, region, country, site type, therapeutic area and risk level — applied over the full ranked estate
- A single ranked list of all sites in scope, exportable for offline review
- Best-performing and bottom-risk cohorts surfaced separately so both ends of the distribution are visible
- A coarse overall site risk level carried alongside the composite ranking score
Decision supported
Which sites to prioritise for operational review, and which cohort a given site sits in relative to comparable sites.
Captured synthetic state
300 ranked synthetic sites scored on a weighted 0–100 composite, with top-ten and bottom-ten cohorts isolated and the full ranking available as an export.
Interpretation & human-review boundary
Decision support only. The ranking performs no site selection, no site activation, no investigator qualification and no regulatory or clinical suitability determination. Where the prototype labels a site as recommended, that means a modelled review candidate.
Showcase 03 · One site, assembled
Site Profile & Observed Recruitment Performance
Site profileObserved historyNamed drivers
Current / observed / modelled context
Observed history for a single site. The recruitment series shown here is what the site has actually done; the forward trajectory question belongs to the Enrollment Forecasting sibling.
Business question
What current operational signals explain how a specific site is performing?
Recorded from the running prototype on synthetic data. The site, investigator and staffing details shown are generated for demonstration.
Key signals on the screen
- Search across site identifier, site name, investigator, country and study to open one site from the estate
- Site metadata with investigator context and staff capacity, including coordinator and research-nurse counts and a turnover-risk indicator
- Observed screening and randomisation history alongside dropout, deviation, data-quality and screen-failure trends over the site's history
- A performance snapshot naming the top performance drivers and top risk factors behind the site's classification
Decision supported
Whether a site's position is explained by recruitment, screening efficiency, staffing, startup or data quality — and therefore what to raise with it.
Captured synthetic state
300 searchable synthetic sites. The captured example resolves to an academic medical centre in Africa with a moderate-performer classification, a site performance score of 38.7, strong historical enrollment and staff capacity as top drivers, and high screen failure and a long startup delay as top risk factors.
Interpretation & human-review boundary
Site-level operational analysis on synthetic aggregates. No patient record is shown or inferred, and no clinical, medical or investigator judgement is made.
Showcase 04 · Operational risk indication
Dropout & Protocol-Deviation Risk
Site-level indicationNot patient-levelNamed factors
Current / observed / modelled context
Indication, not prediction of any individual. Both models score sites — each row is a site within a study, never a person — and both are trained on synthetic operational history.
Business question
Which sites show elevated operational-risk signals that deserve review?
Recorded from the running prototype on synthetic data. Both screens operate at site level; the prototype's own headings describe them as site-level intelligence.
Key signals on the screen
- Average dropout and deviation risk carried on both a 0–1 probability scale and a 0–100 score scale
- Critical and high site counts isolated for each risk type so the actionable cohort is explicit
- Dropout risk distributed by region and deviation risk by site type, revealing where exposure clusters structurally
- Synthetic dropout reasons and deviation categories split by severity, with per-site risk factors and a suggested operational focus
Decision supported
Which sites warrant a retention conversation or a monitoring and training focus, and which specific factor to open that conversation with.
Captured synthetic state
Average dropout probability 0.30 and risk score 47.8, with 30 critical and 58 high dropout sites. Average deviation probability 0.30 and risk score 33.5, with 30 critical and 60 high deviation sites across eight deviation categories.
Interpretation & human-review boundary
These are site-level operational risk indications on synthetic aggregates. They do not predict that any individual will withdraw, make no patient-safety or medical determination, and perform no protocol interpretation, deviation approval, compliance certification or regulatory judgement.
Showcase 05 · Monitoring burden and geography
Data Quality, Monitoring & Geographic Performance
Operational monitoringNot regulatory validationSite operations by country
Current / observed / modelled context
Current operational quality signals and their geographic distribution. This is monitoring workload intelligence, not data validation or compliance assessment.
Business question
Where are data-quality and monitoring concerns concentrated, and how does operational performance differ by country and region?
Recorded from the running prototype on synthetic data. Country and region names are used as synthetic operational groupings only.
Key signals on the screen
- Query rate and average query age alongside missing-data and source-data-verification issue rates across the estate
- A list of sites needing monitoring attention, derived from the same operational quality signals
- Monitoring findings split by visit type — initiation, routine, triggered and close-out — plus the distribution of critical findings
- Average site ranking by country and a region-level performance comparison across the full geographic spread
Decision supported
Where to direct monitoring capacity, and whether a quality or performance pattern is site-specific or geographically structural.
Captured synthetic state
Average query rate 0.15, average query age 11.5 days, missing data rate 5.02% and source-data-verification issue rate 7.08%, with monitoring findings across four visit types, and ranking comparisons spanning 35 synthetic countries and 8 regions.
Interpretation & human-review boundary
Operational monitoring intelligence only. It is not source-data verification, not a data-validation or database-lock readiness assessment, and it makes no GCP compliance, audit-readiness or regulatory-inspection determination. Country performance here means site operations, not supply position — country inventory belongs to the Supply Planning sibling.
Showcase 06 · Assumption and response
Operational Scenario & Recommendation Review
Baseline retainedDeclared assumptionAdvisory only
Current / observed / modelled context
Baseline is the current modelled state; the scenario is a declared operational assumption; the result is a projected consequence held against that baseline, never replacing it.
Business question
How does a defined operational assumption change site exposure, and which interventions deserve human review?
Recorded from the running prototype on synthetic data. Scenario outcomes are modelled illustrations, not predictions about a real study.
Key signals on the screen
- Each scenario metric shown with its baseline printed alongside it, so the delta is explicit rather than inferred
- Movement across site ranking, monthly enrollment, dropout risk, deviation risk, overall risk and critical-site count under one assumption
- A modelled timeline delta in days, and a named operational action associated with the selected scenario
- Recommendations banded by priority with filters across priority, recommendation type and region, plus an export for review
Decision supported
Whether a proposed operational change is worth the resource it would consume, and which recommendations to work first.
Captured synthetic state
Under the captured scenario, average site ranking moves to 50.5 from a 46.5 baseline, average monthly enrollment to 8.3 from 7.8, average dropout risk down to 45.5 from 47.8, deviation risk to 32.5 from 33.5 and overall risk to 16.7 from 24.7, with critical-risk sites unchanged at 4 and a modelled timeline delta of −10 days. The recommendation set holds 4 critical, 223 high and 25 medium items.
Interpretation & human-review boundary
Scenario outputs are modelled consequences of a declared assumption, not forecasts. Recommendations are advisory: nothing here directs an investigator, changes a protocol, activates or deactivates a site, contacts a patient, or executes any recruitment or operational action.
Method, Stated Plainly
Which layers are learned, and which are arithmetic
The screens are titled with the word AI throughout. That is the prototype’s own labelling, and it is not a reliable guide to what each layer actually does — so each one is named accurately below.
- Site ranking method
- Deterministic weighted composite scored 0–100 over nine normalised components, led by enrollment performance and patient-population context, with trial experience, investigator performance, staff capacity and data quality next, and startup speed, dropout risk and deviation risk carrying the smallest weights. Not a learned model.
- Dropout risk method
- Supervised multi-class classification. Risk bands are derived from quantiles of the underlying synthetic retention signals, three candidate classifiers are trained, and the best by weighted F1 is retained. Scored per site within a study.
- Protocol deviation risk method
- The same pattern as dropout risk — quantile-derived bands over synthetic deviation history, three candidate classifiers, best retained by weighted F1. Scored per site within a study.
- Site performance classification
- Supervised classification into performer classes, again over quantile-derived labels, with a simple rule-based classifier retained internally as a reference comparison.
- Data quality method
- Deterministic aggregation of synthetic query, monitoring-visit and source-data-verification records into per-site operational quality metrics. No classifier is involved.
- Scenario method
- Deterministic deltas applied to baseline aggregates for a named operational assumption, with the baseline retained and displayed alongside every scenario figure. The prototype's own code describes its critical-risk impact estimate as approximate.
- Recommendation method
- A fixed inventory of rules keyed on observed site signals — overall risk, recruitment, retention, deviation, data quality, startup, staffing, monitoring focus, site selection candidacy and reallocation — each emitting a priority-tagged, reason-carrying recommendation.
- Model types used
- Logistic regression, random forest and gradient boosting classifiers for the site performance, dropout and deviation tasks. The enrollment-prediction module additionally uses regression models, but that capability is owned by the Enrollment Forecasting sibling and is not presented here.
- Analytical level
- Site, and site within study. Event-level synthetic records are aggregated before analysis; nothing is scored, predicted or reported for an individual person.
Why no model scores are published here
The prototype does publish per-model evaluation figures inside its own terminology screen, and they look strong. They are deliberately not reproduced on this page, because of how the labels are constructed: the performance, dropout and deviation classes are derived by banding quantiles of the same synthetic signals the classifiers then learn from. A high score therefore measures how faithfully a model reproduces a deterministic rule on generated data — not clinical or operational predictive power. Quoting those numbers as evidence of capability would be misleading, so the models are named and their scores are not. Nothing on this page should be read as validated prediction, clinical validation, real-world accuracy or generalisable performance.
Supporting Intelligence
The remaining screens, and one deliberate handoff
These run against the same seeded estate but are not presented as showcases — in one case because the capability is owned by a sibling experience rather than by this one.
Enrollment Prediction — supporting context only
The underlying prototype does include an enrollment-prediction view, with per-site forward projections and target-attainment probabilities. Within this family it is treated as supporting context rather than a capability this page owns. Forward enrollment trajectory, forecast methodology, baseline-versus-model comparison and rolling-origin evaluation are owned by Clinical Enrollment Forecasting & Site Risk Intelligence, and are deliberately not reproduced here.
Synthetic Data Explorer
Direct inspection of the underlying synthetic tables — site master, investigator master, screening and enrollment history, dropout and deviation history, monitoring visits, query management, staffing, startup and payments — so any figure elsewhere in the prototype can be traced to the rows behind it.
AI Terminology Guide
A plain-language glossary of the concepts the prototype uses, together with an in-application model registry that publishes each candidate model's evaluation on the synthetic dataset. Those figures are discussed in the method note below rather than reproduced as performance claims.
Capabilities
What each capability does, decides and uses
Site ranking composite
What it does
Scores every site on a weighted 0–100 composite built from nine normalised components and orders the estate accordingly.
Decision supported
Which sites to review first, relative to comparable sites.
Evidence used
Declared component weights led by enrollment performance and patient-population context, with data quality, staffing, experience, startup speed and risk contributing the remainder.
Site profile assembly
What it does
Joins site metadata, investigator context, staffing, recruitment history, risk trends and data quality into one site record.
Decision supported
What explains an individual site's current position.
Evidence used
A per-site join across the synthetic site, investigator, staffing, screening, enrollment, deviation and query tables.
Observed recruitment monitoring
What it does
Tracks what each site has actually screened and randomised over its history, with the screen-failure trend alongside.
Decision supported
Whether a site's recruitment problem is volume or conversion.
Evidence used
Twenty-four months of synthetic monthly screening and enrollment history per site.
Site-level dropout risk indication
What it does
Produces a per-site dropout probability, 0–100 score, risk band and named contributing factors.
Decision supported
Which sites warrant a retention focus.
Evidence used
Supervised classification over synthetic site retention history — scored per site within a study, never per patient.
Site-level deviation risk indication
What it does
Produces a per-site protocol-deviation probability, score, band and contributing factors, with categories split by severity.
Decision supported
Which sites warrant a monitoring or training focus.
Evidence used
Supervised classification over synthetic deviation history, again at site level.
Data quality & monitoring analysis
What it does
Aggregates query rate and age, missing data, source-data-verification issues and monitoring findings into an operational quality view.
Decision supported
Where to direct monitoring capacity.
Evidence used
Deterministic aggregation over synthetic query-management and monitoring-visit records.
Country & region operational comparison
What it does
Compares average site ranking and performance across countries and regions.
Decision supported
Whether a weakness is site-specific or geographically structural.
Evidence used
Geographic aggregation of the site ranking composite across the synthetic country and region master.
Operational scenario analysis
What it does
Applies a declared operational assumption to the current baseline and reports the projected movement across each headline metric.
Decision supported
Whether an operational change is worth its cost.
Evidence used
Deterministic scenario deltas applied to baseline aggregates, with the baseline retained and displayed alongside.
Explainable recommendations
What it does
Applies rules keyed on observed site signals to produce prioritised, reason-carrying recommendations.
Decision supported
Which sites to act on and what to raise.
Evidence used
A fixed rule inventory spanning overall risk, recruitment, retention, deviation, data quality, startup, staffing, monitoring focus and reallocation.
System Design
Architecture and data basis
A seeded pipeline from synthetic study and site generation through operational event history, aggregation and scoring, into a review surface — with the deterministic layers and the learned layers kept separate.
- Layer 01Synthetic study, site & investigator generationA seeded generator produces 12 studies, 300 sites, 500 investigators across 35 countries and 8 regions, with therapeutic areas, phases and site types drawn from reference taxonomies.
- Layer 02Synthetic operational event historyTwenty-four months of screening, enrollment, dropout, protocol-deviation, monitoring-visit and query records, generated at event level and consumed as site-level aggregates.
- Layer 03Relational store & loadersThe generated tables are persisted to CSV and a local SQLite database, so every screen reads from one consistent set of records.
- Layer 04Feature assembly per siteEach analytical module joins the relevant tables into a per-site frame covering recruitment, retention, deviation, quality, staffing and startup signals.
- Layer 05Deterministic ranking & risk compositesSite ranking and overall operational risk are weighted composites over normalised components — explicit arithmetic, not learned models.
- Layer 06Supervised classification layerSite performance, dropout risk and deviation risk each train several candidate classifiers and retain the best by weighted F1, together with per-site contributing factors.
- Layer 07Scenario layerDeclared operational assumptions are applied deterministically to baseline aggregates and stored alongside the baseline for comparison.
- Layer 08Rule-based recommendation layerA fixed rule inventory converts scored exposures into prioritised recommendations, each carrying its triggering signal.
- Layer 09Application & evidence surfaceA twelve-screen Streamlit application with filters, exports, a synthetic data explorer and an in-app model registry, carrying its demo disclaimer on every screen.
- Layer 10Human-review boundaryEvery output ends at a qualified clinical-operations reviewer. The prototype ranks, indicates and proposes; it never acts.
Built in Python with Streamlit, pandas, NumPy, scikit-learn, Plotly, SQLite and ReportLab, running locally against a deterministically generated synthetic estate. There is no connection to any clinical trial management, electronic data capture, or randomisation and trial supply management system.
Lab Standards
Explainability, governance and guardrails
Site-performance language sits close to decisions that carry real professional consequences for investigators and real regulatory weight for sponsors. The boundaries below are therefore stated explicitly rather than assumed, and the prototype carries its own demo disclaimer on every screen.
- Synthetic site-level clinical-trial operational data only
- No real patient data — no patient-level records anywhere in the dataset
- No real trial, sponsor, site or investigator data
- Operational clinical-trial execution support only
- No treatment or medical decisions
- No efficacy prediction
- No patient-safety determination
- No patient-level clinical prediction
- No protocol interpretation, amendment or deviation approval
- No regulatory decision or compliance determination
- No GCP or GxP validation claim
- No source-data verification or database-lock readiness assessment
- No site selection, qualification, approval or activation execution
- No investigator qualification determination
- No autonomous recruitment or operational intervention
- Operational scenario outputs are modelled illustrations, not forecasts
- No live CTMS, EDC or IRT integration
- Independent professional prototype
- Qualified human review required for every recommendation
Honest Status
Current prototype maturity
The prototype runs locally against a seeded synthetic estate. It is not deployed, not published as an application, and not available to run.
Working today
- Twelve working screens over one seeded synthetic site estate
- Portfolio overview across studies, sites, performer splits and risk averages
- Weighted site ranking with six filters and exportable results
- Searchable site profiles assembling metadata, investigator, staffing and history
- Observed recruitment, screening, retention, deviation and quality trends per site
- Site-level dropout and protocol-deviation risk indication with named contributing factors
- Data-quality and monitoring analysis across queries, missing data and monitoring findings
- Country and region operational performance comparison
- Operational scenario analysis with the baseline retained alongside each result
- Rule-derived, prioritised recommendations with exports and an evidence explorer
Intentionally not claimed
- No autonomous site selection, qualification, approval or activation
- No investigator qualification determination
- No patient-level clinical prediction of any kind
- No medical, treatment, efficacy or patient-safety judgement
- No protocol decision, deviation approval or regulatory determination
- No GCP or GxP validation, and no audit-readiness certification
- No validated clinical model and no real-world predictive performance
- No live CTMS, EDC or IRT integration
- No autonomous recruitment or operational intervention
- No realized enrollment improvement in any real study
- Not deployed, not published as an application, and not available to run
These are deliberate boundaries of a synthetic-data prototype, not pending features. The site-level framing of dropout and deviation risk in particular is a decision about what the data can honestly support, not a limitation waiting to be lifted.
What This Demonstrates
Professional capability demonstrated
Building this experience exercised the following, end to end.
Operational framing
Turning a portfolio average into a per-site operational question, and keeping the present tense — current execution — distinct from the forward outlook a sibling experience owns.
Explainable prioritisation
Ranking sites on declared, weighted components and carrying the contributing factors through to every risk indication, so an output can be argued with rather than merely accepted.
Naming the analytical level
Holding dropout and deviation risk explicitly at site level, and refusing the much more marketable — and much less honest — patient-level reading of the same output.
Boundary discipline
Separating operational monitoring from regulatory validation, ranking support from site selection, and modelled scenario consequence from forecast, with clinical-operations authority left with a person.
Continue through the family
This experience opens the Clinical Trial Supply Chain Intelligence family. The forward enrollment outlook built on the execution shown here belongs to the forecasting sibling, and the clinical-supply response that follows belongs to Supply Planning & Resupply Intelligence.