Digital Twins in Poultry Farming: Deconstructing the Evidence Gap Between Promise and Performance

Abstract

Digital twins, understood as computational replicas of poultry production systems updated in real time by sensor data, are increasingly invoked as transformative tools for precision livestock farming and sustainable agriculture. They are credited with enhancing feed efficiency, reducing greenhouse gas emissions, enabling disease detection earlier and improving animal welfare. Yet close examination of the published evidence reveals that these promises rest on a surprisingly narrow empirical foundation. Across the available literature, no peer reviewed study has quantified the full lifecycle carbon footprint of digital twin infrastructure in poultry production. Only one field validated investigation reports a measurable improvement in feed conversion ratio attributable to digital optimization, and that study’s design constrains its general applicability. A standardized performance assessment framework specific to poultry has not been established. Quantitative evaluations of reliability are scarce, limited to a small number of studies reporting data loss, sensor degradation and cloud system downtime, and no work has documented abandonment timelines or reasons for discontinuation. The result is a pronounced gap between technological aspiration and verified performance. Progress in this domain will depend on small-scale, deeply instrumented deployments capable of generating the longitudinal, multidimensional evidence required to substantiate the environmental and operational benefits attributed to digital twins.

1. Introduction

1.1. Digital Twins: Aspiration and Ambiguity

Digital twins occupy an increasingly central and largely unexamined position in contemporary discussions of precision livestock farming. Conceptually, they are intended to function as living computational mirrors of barns and animals, with virtual representations held in continuous synchronization with physical reality through high-frequency sensor streams. In this idealized form, digital twins enable prediction of future states, exploration of counterfactual scenarios and optimization of interventions before they are deployed in situ. In poultry production, this vision has attracted a constellation of technologies, from elementary networked environmental sensors and commodity microcontrollers to elaborate cloud-connected cyber-physical architectures that consolidate environmental monitoring, computer vision, acoustic analysis, gas detection, automated climate and feed control and algorithmic decision support [1,2,3].

The enthusiasm surrounding digital twins has grown rapidly in parallel with mounting pressures on livestock systems. These include climate imperatives, animal welfare expectations and labor scarcity. It is tempting to construct a narrative arc in which continuous monitoring, predictive analytics and decision support culminate in fewer mortalities, enhanced feed efficiency, reduced energy demand, superior welfare outcomes and more transparent supply chains. Conference presentations, trade publications and vendor communications often cite claims of 8–12 percent improvements in feed conversion ratio, double-digit energy reductions, disease detection sensitivity exceeding 70–90 percent and substantial mitigation of greenhouse gas emissions. However, when one poses a deceptively straightforward empirical question—namely, what has actually been quantified in field conditions under commercial protocols and documented in peer-reviewed literature—the available evidence contracts sharply.

1.2. The Disconnect: Claims Versus Evidence

On closer inspection, claims of improved feed efficiency rest on a single broiler study that reports a five percent improvement in feed conversion ratio, and that study was conducted without randomization or reported statistical testing [4]. Discussions of so-called green or climate-smart digital agriculture emphasize avoided emissions from more efficient production, yet no lifecycle carbon assessment has been published for digital infrastructure itself in any poultry context. Comparative evaluations of different digital twin architectures implicitly assume the existence of agreed measurement frameworks, but systematic examination shows that no such standards have been operationalized or validated in poultry production.

Technical deployments are described in detail in many papers, but quantitative characterization of long-term reliability, field data quality and the prevalence and causes of system abandonment by commercial operators remains conspicuously sparse. The result is a growing gap between rhetorical promise and empirical foundation. The question is not whether digital twins could enhance poultry production, as their technical potential is evident, but whether the current published evidence substantiates the specific claims now circulating and what remains unmeasured. Figure 1 summarizes the disparity between circulating claims and documented evidence across these four domains.

1.3. Research Scope and Methodology

This systematic review responds to that disconnect by synthesizing evidence across four thematic domains, each organized around a specific, empirically tractable question. By conducting comprehensive searches of leading academic databases and screening over two thousand candidate papers, we map the landscape of what has been measured, what remains unmeasured and where methodological pathways forward are clearest.

The four central research questions examined are as follows. First, carbon accounting: have lifecycle greenhouse gas footprints, including Scope 1, 2 and 3 emissions, of digital twin hardware, cloud infrastructure and communication networks deployed in poultry farming operations been quantified in peer-reviewed work? Second, feed efficiency: what field-validated evidence exists demonstrating feed conversion ratio improvements attributable to digital twins or related optimization systems in commercial poultry production when compared with conventional management? Third, performance standards: do standardized frameworks exist for systematically measuring and comparing digital twin performance across heterogeneous poultry production configurations? Fourth, operational reliability and discontinuation: what data characterize system failure rates, maintenance requirements, uptime percentages and abandonment timelines for precision livestock farming technologies in commercial poultry operations?

These four lines of inquiry directly probe the foundations of digital twin claims. By clarifying where knowledge is robust versus absent, we identify research priorities capable of advancing the field from aspiration toward evidence-grounded practice. We further argue that small-scale but comprehensively instrumented digital twin deployments, exemplified by a 2000-bird research flock monitored continuously across feed intake, water consumption, egg production, mortality, environmental conditions, equipment runtime, maintenance events, labor allocation, hardware energy use and cloud computing resource demand over two-year periods, are uniquely positioned to generate the multi-dimensional, longitudinal datasets that the field presently lacks. Such deployments can address simultaneously the carbon, feed efficiency, performance assessment and reliability gaps and generate the rigorous evidence base required to move digital twins from prototype promise to validated performance.

1.4. Integrated Analytical Framework: Interdependencies Among Evidence Gaps

The four research questions examined in this review, lifecycle carbon footprints, feed efficiency, performance assessment frameworks, and operational reliability, are often treated in the literature as distinct or only loosely connected topics. When viewed through a systems perspective, however, they constitute interdependent dimensions of a single underlying challenge: the absence of integrated empirical validation for digital twin systems deployed in poultry production. Considering these dimensions in isolation limits the interpretability of reported outcomes and obscures how constraints in one domain influence conclusions drawn in others.

A primary interdependency exists between carbon accounting and feed efficiency. Reported improvements in feed conversion ratio (FCR) are frequently presented as evidence that digital twins contribute to more sustainable poultry production. Such claims can only be evaluated meaningfully when reductions in feed-related emissions are assessed alongside the lifecycle greenhouse gas emissions associated with the digital infrastructure itself. Sensors, communication networks, edge devices, and cloud-based computation introduce embodied and operational emissions that may partially or fully offset gains achieved through improved feed efficiency. In the absence of Scope 1, 2, and 3 carbon accounting for digital twin systems, the net environmental impact of these technologies, defined as avoided production emissions minus infrastructure-related emissions, cannot be quantified. As a result, assertions of climate benefit remain empirically unresolved.

A second interdependency links performance standardization with operational reliability. The absence of standardized performance assessment frameworks tailored to poultry production constrains both the interpretation and comparability of reported reliability outcomes. Existing studies employ heterogeneous definitions of performance, adopt inconsistent system boundaries, and report reliability using non-uniform indicators such as accuracy, uptime, or qualitative user assessments. Without agreed metrics, thresholds, and reporting conventions, reliability characteristics including system uptime, data loss rates, sensor degradation, and maintenance requirements cannot be systematically compared across deployments. Consequently, the available evidence remains fragmented and dominated by isolated case studies rather than cumulative, comparable evaluations.

A third interdependency connects feed efficiency outcomes with system reliability over time. Production gains attributed to digital optimization depend on the sustained functionality of the sensing, communication, and control components that underpin digital twin operation. A system that demonstrates improved ventilation control or feeding strategies under short-term or favorable conditions may fail to deliver consistent benefits if performance degrades due to sensor drift, intermittent connectivity, software instability, or increasing maintenance demands. Short-duration trials therefore provide limited insight into whether observed efficiency gains are robust across production cycles, seasons, and flock variations. Without longitudinal reliability data explicitly linked to production outcomes, it remains unclear whether reported efficiency improvements reflect stable system behavior or context-specific successes.

Together, these interdependencies point to a broader challenge of cross-domain integration. Carbon assessments that do not incorporate realistic assumptions about reliability, maintenance, and replacement rates are likely to underestimate infrastructure-related emissions. Performance frameworks that omit environmental impact and operational robustness risk favoring technically sophisticated yet practically fragile systems. Similarly, field trials that report production improvements without documenting system uptime, maintenance burden, or generalization across flocks and seasons provide an incomplete basis for economic and environmental decision-making. As a result, key stakeholders, including producers, technology developers, and regulators, lack an integrated evidentiary foundation for determining whether digital twin systems can deliver sustained net benefits under commercial poultry production conditions.

2. Systematic Review Methodology

2.1. Search Strategy

A comprehensive systematic search was performed across three leading academic databases: Web of Science, Scopus and Google Scholar. Search strategies employed a combination of controlled terms and free-text keywords targeting digital twin technologies, precision livestock farming and poultry production.

The primary search query combined the expressions “digital twin” OR “cyber-physical system” OR “precision livestock farming” OR “smart poultry” OR “IoT monitoring” with “poultry” OR “broiler” OR “layer” OR “chicken” and with performance-related terms such as “performance”, “efficiency”, “carbon”, “failure”, “reliability”, “abandonment” or “sustainability”. Secondary queries were formulated to focus more narrowly on each domain. For carbon accounting, terms related to “digital twin”, “precision livestock farming” or “IoT” were coupled with “poultry” and with “carbon footprint”, “lifecycle assessment”, “greenhouse gas”, “energy consumption” or “embodied emissions”. For feed efficiency, search terms combined digitalisation expressions with “poultry” or “broiler” and with “feed conversion ratio”, “FCR”, “growth performance” or “production efficiency”. For performance frameworks, searches linked “digital twin” with “assessment”, “framework”, “standards”, “performance metrics” or “benchmarking” and with “poultry” or “livestock”. For reliability and failure, digital and precision farming terms were coupled with “poultry” and with “failure”, “reliability”, “downtime”, “abandonment”, “discontinuation” or “maintenance”.

Searches were conducted between June and November 2025, with no date limitations imposed beyond database coverage. Boolean operators and proximity searching were used to refine results and filter out obvious noise. Citation information and abstracts for all retrieved papers were exported for screening.

2.2. Inclusion and Exclusion Criteria

Studies were included if they were peer-reviewed and fell into one of three broad categories: original research, systematic reviews or quantitative syntheses. Eligible studies needed to be conducted in commercial, pilot or experimental poultry operations with a minimum of 1000 birds. They had to involve digital or precision technologies that addressed monitoring, control, optimization or decision support, including, but not limited to, sensors, IoT platforms, AI-based analytics or cyber-physical management systems. Studies were required to provide quantitative data regarding system performance, carbon footprints, failure modes or related outcomes. Only English-language publications appearing between 2000 and 2025 were considered.

Studies were excluded if they were purely theoretical models with no empirical component, non-poultry livestock studies without explicit comparative relevance, non-digital precision farming such as mechanical automation only, conference abstracts without a full-text article, opinion editorials, non-peer-reviewed gray literature or studies that lacked quantitative outcome measurement.

2.3. Screening and Study Selection

The initial set of searches across all databases returned 2147 records. After deduplication, 1854 unique records remained. Two independent reviewers screened titles and abstracts against the inclusion criteria, achieving substantial agreement as indicated by Cohen’s kappa coefficient of 0.78. Discrepancies were resolved through discussion or, where necessary, consultation with a third reviewer.

Full texts were obtained for 312 potentially eligible studies. Of these, 160 met all eligibility criteria and were retained for detailed analysis. For each thematic area—namely, carbon footprints, feed efficiency, performance frameworks and reliability—forty studies with the strongest methodological quality (Figure 2) and relevance were selected for in-depth extraction and synthesis. Methodological quality was appraised using a structured tool adapted from the Cochrane Risk of Bias framework, modified for engineering and agricultural contexts. This ensured that findings discussed in the results give greatest weight to the most reliable evidence.

2.4. Data Extraction

Structured data extraction was conducted by trained reviewers using standardized forms. For each study, variables such as publication year, country, study design, technology type, poultry production system, sample size and duration were recorded. For carbon-oriented studies, the focus of extraction was on embodied emissions from manufacturing, operational emissions from electricity and fuel, treatment of cloud infrastructure and networks, system boundaries, lifecycle stages assessed, the carbon intensity values used and validation methods.

For feed efficiency studies, extraction captured experimental design, including whether trials were randomized or observational, the number of replicates or production cycles, baseline and intervention feed conversion ratio values, statistical testing methods, potential confounding factors and contextual variables such as genetic line, feed formulation and climatic conditions.

For performance framework studies, extraction documented the framework name and type, core components or dimensions, validation status, whether the framework was adapted to poultry or livestock, and which performance dimensions—such as technical accuracy, economic viability, environmental impact or social aspects—were considered. Implementation challenges and barriers were also noted.

Across reviewed studies, environmental sensors typically operate at sampling intervals of 1–5 min, while vision and acoustic systems generate continuous or near-continuous data streams, creating synchronization and temporal alignment challenges that are rarely documented explicitly.

For reliability studies, extraction recorded types and frequencies of failures, duration of monitoring, methods used to measure reliability, environmental or infrastructural factors affecting system performance, maintenance requirements and any available information on system abandonment or discontinuation.

2.5. Synthesis Approach

Given substantial heterogeneity in study designs, outcome definitions and quality, we employed narrative synthesis structured by thematic domain rather than quantitative meta-analysis. Within each domain, evidence was organized according to strength and direction of findings. Quantitative results are reported using original units and effect measures where possible. Particular attention is paid to consistency, or lack of consistency, in effect directions across studies. Methodological limitations in the literature are discussed explicitly in order to provide realistic bounds on inference and highlight areas where future study designs should be strengthened.

Figures and tables in this review present analytical syntheses derived from the screened literature rather than reproductions of third-party datasets. Study inclusion, variable extraction, and interpretive synthesis were conducted according to the criteria described in Section 2.1, Section 2.2, Section 2.3 and Section 2.4, with primary sources cited directly within tables or captions where quantitative values are reported.

3. Results: Four Interlocking Evidence Gaps

Table 1. Evidence landscape and unresolved gaps across four thematic domains in poultry-oriented digital twin research.

Domain	Sub-Theme	Core Research Question	Evidence Availability	Strongest Empirical Finding	Study Context	Methodological Limitations	Generalisability	Key Unresolved Issue	Priority for Future Work
Carbon footprint	Lifecycle assessment	Have Scope 1–3 lifecycle carbon footprints of poultry digital twin systems been quantified?	None in poultry; one related study in crops	0.2–0.3 Mt CO2e per year from national deployment of crop-monitoring devices [5]	Digital agriculture devices in French crop systems	Non-poultry; partial boundaries; excludes cloud and networks	Conceptually relevant but numerically not transferable to poultry	Net carbon balance of poultry digital twin deployments is unknown	Conduct full Scope 1–3 LCAs of poultry digital infrastructures with carbon payback analysis
Carbon footprint	Farm-level emissions context	How carbon-intensive are conventional poultry houses without digital twins?	Moderate: several farm-level models	383–524 t CO2e per broiler house per year, dominated by LPG heating [6]	Conventional broiler operations without digital infrastructure	Region-specific energy mix; simplified emission sources	Provides emission scale but not marginal impact of digital tech	How much digital twins can realistically reduce whole-farm GHG emissions	Run paired-farm comparisons with and without digital twins using harmonized carbon accounting
Carbon footprint	Renewable integration	How do smart energy systems interact with poultry emissions?	Limited: some solar/energy studies	~1698 t CO2e over 30 years for solar-intelligent systems [7]	Solar-powered feeding infrastructure	Focus on solar hardware; excludes full twin stack	Shows green infrastructure has non-trivial embodied emissions	Interaction of twins, renewables, net emissions unclear	Integrate twin and energy system LCAs in joint optimization
Feed efficiency	Direct FCR improvement	Do digital twins improve FCR under commercial conditions?	Very limited: one primary study	FCR improved from 1.640 to 1.561 (~5%) [8]	Commercial broiler houses	Non-randomized; single cycle; limited stats	Suggestive but not robust	True distribution of effects unknown	Run randomized, multi-cycle trials across regions
Feed efficiency	Indirect performance metrics	Do digital systems improve metrics correlated with FCR?	Moderate	Growth, mortality, welfare improvements [9]	Brooding and layer systems	FCR absent or confounded	Benefits indirect, not quantified as FCR	How indirect gains map to FCR	Include FCR and cost metrics in trials
Performance frameworks	Generic DT frameworks	Are there established general DT frameworks?	Strong outside agriculture	Capability, cooperability, lifecycle frameworks [10]	Industrial DTs	Not applied to livestock	Untested in poultry	Adaptation needed	Translate and pilot frameworks in poultry
Performance frameworks	Poultry-specific proposals	Any poultry-specific indices?	Conceptual only	Domain Transfer Score, Data Reliability Index [11]	Conceptual welfare system	No field DTS/DRI values	Addresses generalisability in theory only	Behavior under real conditions unknown	Apply on multi-farm datasets
Performance frameworks	Measured dimensions	Which dimensions are measured?	Heterogeneous	Focus on accuracy; sparse economic, welfare metrics [12]	Poultry phenotyping	No holistic frameworks	Bias toward technical metrics	Need for integrated metrics	Require multi-domain reporting
Reliability	Data quality	How reliable are data streams?	Sparse	40–50% discarded early; 10–20% after fix [13]	Broiler barns	Single system; research-managed	Heavy post-processing needed	Typical discard rates unknown	Standardize data quality reporting
Reliability	Tracking accuracy	How accurate is tracking?	One detailed study	23% raw vs. 91–99% processed; strong zone effects [14]	Layer aviary	Short duration; one design	Highly layout-dependent	Performance across designs unknown	Cross-barn tracking studies
Reliability	System availability	How does architecture influence uptime?	Model-based	34% downtime no redundancy; 9% with redundancy [15]	Dependability model	No field validation	Suggests redundancy helps	Field uptime distributions unknown	Publish real uptime logs
Reliability	Environmental stressors	Which environmental factors matter?	Fragmented	Path loss from cages; dust, ammonia degrade sensors [16]	Wireless sensors	Short-term; specific setups	Likely generalisable	Failure probabilities unclear	Link environment to failure logs
Reliability	Abandonment	How often are systems abandoned?	None	No empirical studies	None	Unknown adoption trajectories	Completely unmeasured	Drivers of discontinuation	Run adoption—abandonment studies
Cross-cutting	Research-commercial divide	How does study setting shape evidence?	Strong pattern	~50% in research barns	Study classification	Ambiguous reporting	Research barns not representative	Translation to commercial	Prioritize commercial studies
Cross-cutting	Standardization	Are definitions consistent?	Weak	Terms used inconsistently	Terminology review	No ontology	Hampers synthesis	Agreement lacking	Develop reporting standards

synthesizes findings from studies meeting inclusion criteria, organized by domain, sub-theme and core research question. For each area, it summarizes the current availability of evidence, the strongest empirical insight, the study context, key methodological limitations, generalizability concerns, the principal unresolved issue and the priority direction for future research. Together, these entries highlight the sharp asymmetry between technical development and empirical validation, underscoring the need for longitudinal, multidimensional evidence generation in commercial poultry environments.

3.1. Carbon Footprints: A Complete Absence of Poultry-Specific Lifecycle Assessments

The first major finding of this review concerns lifecycle carbon footprints of digital twin systems in poultry. Among the 2147 records initially identified and forty studies examined in detail for carbon-related outcomes, no peer-reviewed work quantified Scope 1, 2 and 3 emissions for digital twins or digital infrastructures in poultry production.

The only study that offers a detailed lifecycle assessment of digital agriculture devices is that by La Rocca et al., and this study focuses on crop monitoring rather than poultry [5]. Their parametric bottom-up modeling approach estimates that deployment of digital devices such as RFID tags, collars, cameras and edge computing units across French crop farms would produce between 0.2 and 0.3 megatonnes of CO₂ equivalent per year. Their assessment considered embodied emissions based on device lifetimes and lifecycle databases, as well as operational emissions tied to device energy use and grid carbon intensity. The study emphasized that more complex devices had higher footprints and that improved performance did not always offset these costs. However, the work did not include cloud data centers, network infrastructure or data transport in its boundaries and does not address poultry houses, where conditions for hardware and networking are very different. Device-level assessments alone are insufficient because a substantial share of digital twin energy use and emissions arises from the computational load of cloud processing, data storage and continuous network transmission, none of which were included in that model. Additionally, poultry houses impose harsh environmental conditions, including dust, ammonia and sustained humidity and temperature fluctuations, which can shorten the operational lifetime of sensors and edge devices, thereby increasing embodied emissions per functional year.

In poultry-specific studies, researchers provide rich descriptions of technology components used in digital or precision systems. These include environmental sensors such as DHT temperature-humidity units, gas sensors for carbon dioxide, ammonia and methane, particulate matter sensors, oxygen sensors, microcontrollers such as ESP32 and ESP8266, RFID tags, cameras, wearables and edge computing devices. Cloud platforms such as Firebase, AWS IoT and ThingSpeak are frequently used, and communication networks include WiFi, LoRa, GSM and MQTT-based configurations. Yet no study attributes emissions to these devices or to the underlying infrastructure.

Important contextual work by Fan et al. uses a carbon footprint estimation tool to estimate emissions from broiler houses, finding that a typical house emits between 383 and 524 tonnes of CO₂ equivalent per year, largely due to LPG heating [6]. Yang et al. examine the cost and emissions associated with solar-powered poultry feeding systems, reporting almost 1700 tonnes of CO₂ equivalent over thirty years for the solar infrastructure [17]. These baselines underscore the carbon intensity of poultry operations but do not include contributions from digital components.

The net result is that digital infrastructures, including hardware, cloud services and networks, are effectively invisible in lifecycle carbon accounting for poultry. This is problematic in light of strong sustainability claims for digital twins. Without measured emissions associated with digital systems themselves, it is impossible to determine whether they yield net reductions at farm or sector scale or simply shift emissions from one domain to another. The absence of carbon payback analysis for digital twin deployment in poultry therefore constitutes a central and unresolved research gap.

Taken together, the available evidence establishes with confidence that digital twin infrastructures are currently excluded from lifecycle carbon accounting in poultry production, despite frequent sustainability claims associated with their deployment. What remains unmeasured is the net environmental balance between infrastructure-related emissions, including hardware manufacture, energy use, networking, and cloud computation, and any production-side emission reductions achieved through improved efficiency. This omission directly constrains real-world adoption by preventing producers, integrators, and policymakers from assessing whether digital twins deliver genuine climate benefit or simply redistribute emissions across system boundaries. Addressing this gap requires full Scope 1, 2, and 3 lifecycle assessments explicitly tailored to poultry digital twin deployments, with carbon payback analysis that links infrastructure emissions to empirically validated production outcomes.

To contextualize these carbon and performance gaps within the broader computational landscape of poultry precision systems, we examined the algorithmic methods reported across all included studies.

Table 2. Distribution of computational methods across 160 systematically reviewed studies of precision livestock farming and cyber-physical systems in poultry. Studies are classified by primary algorithmic approach and highest reported validation level (Conceptual, Research, Pilot, Commercial). Only 4.4% achieved commercial-scale validation, and none meet criteria for fully realized digital twins.

Algorithmic Category	Method/Architecture	N Studies	% of Total	Primary Application	Validation Level
Machine Learning (Classical)
	Random forests	18	11.3%	Behavior classification, mortality prediction	Lab/research only
	Support vector machines (SVM)	12	7.5%	Disease detection, weight estimation	Research barns
	Gradient boosting (XGBoost, LightGBM)	9	5.6%	Production outcome prediction	Research barns
	Decision trees	8	5.0%	Simple classification tasks	Research only
	k-Nearest neighbors (k-NN)	5	3.1%	Anomaly detection	Research only
Deep Learning
	Convolutional neural networks (CNN)	24	15.0%	Computer vision, behavior recognition	Lab/pilot scale
	Recurrent neural networks (RNN/LSTM)	14	8.8%	Time-series prediction, temperature forecasting	Research/pilot
	Long short-term memory (LSTM) specifically	8	5.0%	Sequential data (feed intake, growth curves)	Research barns
	Bidirectional LSTM (BiLSTM)	3	1.9%	Temperature prediction, spatial interpolation	Research onlyapplsci-16-01317-Jan-18-2026.docx
	Transformers/attention mechanisms	2	1.3%	Multimodal sensor fusion	Conceptual only
	Autoencoders	4	2.5%	Anomaly detection, data compression	Research only
Optimization Algorithms
	Genetic algorithms (GA)	6	3.8%	Feed schedule optimization, climate control	One commercial trialapplsci-16-01317-Jan-18-2026.docx
	Particle swarm optimization (PSO)	3	1.9%	Parameter tuning	Research only
	Simulated annealing	2	1.3%	Resource allocation	Conceptual
Fuzzy Systems
	Fuzzy logic controllers	7	4.4%	Climate control, brooding automation	Pilot scaleapplsci-16-01317-Jan-18-2026.docx
	Adaptive neuro-fuzzy inference (ANFIS)	3	1.9%	Decision support systems	Research only
Physics-Based/Mechanistic Models
	Computational fluid dynamics (CFD)	9	5.6%	Airflow, ventilation optimization	Simulation only
	Energy balance equations	11	6.9%	Temperature modeling, heating/cooling	Validated in some studies
	Growth curve models (Gompertz, logistic)	6	3.8%	Weight prediction	Commercial baseline
Hybrid Approaches
	Physics-informed neural networks (PINN)	2	1.3%	Temperature + biological modeling	Conceptual only
	ML + mechanistic constraints	8	5.0%	Combining data-driven and rule-based	Research barns
Statistical Methods
	Linear/multiple regression	15	9.4%	Baseline comparisons	Commercial use
	ARIMA, exponential smoothing	5	3.1%	Time-series forecasting	Research only
	Bayesian networks	4	2.5%	Uncertainty quantification	Research only
Agent-Based Models
	Individual-based bird simulation	4	2.5%	Flock behavior emergence	Simulation only
No Algorithmic Component
	Monitoring only (dashboards, alerts)	42	26.3%	Data visualization, threshold alerts	Commercial deployment
	Conceptual/theoretical frameworks	18	11.3%	No implementation	Not applicable

summarizes the distribution of computational approaches used in poultry precision livestock farming and cyber-physical systems reviewed in this evidence map, together with their highest reported validation level. Importantly, while a wide range of machine-learning, optimization, and mechanistic models is reported, none meet the criteria for fully realized digital twins operating under sustained commercial conditions.

3.2. Feed Efficiency: The Single Five Percent Study

The second domain concerns feed efficiency, particularly feed conversion ratio. Out of the 2147 records screened, only a single study met the criteria for a field-validated comparison of feed conversion ratio between a digital twin-like optimization system and conventional management in a commercial poultry setting.

In this study, Klotz and colleagues deployed a system that combined long short-term memory neural networks with genetic algorithms to optimize action plans in interconnected commercial broiler houses [8]. Under specialist-configured conventional management, the feed conversion ratio was 1.640. Under digital optimization, the feed conversion ratio decreased to 1.561. This corresponds to a relative improvement of about five percent, equivalent to 79 g less feed per kilogram of meat produced. At flock scale, this was estimated to save around 7500 kg of feed for a 34,000-bird flock over a forty-day production cycle.

Despite the apparent importance of this result, several methodological limitations constrain its generality. The design was not randomized and comprised a quasi-experimental comparison rather than a controlled trial. The reported results cover only a single cycle and do not account for seasonal or management variability. Statistical testing is not reported in detail, making it difficult to assess confidence intervals or effect robustness. The study also does not partition the contributions of different management changes, so that specific effects of feed-related optimization remain indistinct. Moreover, the trial did not assess cost–benefit outcomes, carbon implications or any welfare indicators, underscoring that its findings address only a narrow slice of system performance and reinforcing how siloed the existing evidence base remains.

Other studies address related questions but do not deliver robust FCR comparisons. Elwakeel et al. examined automated environmental control in broiler houses, reporting some performance changes but not clearly establishing a significant or replicable FCR improvement [9]. Park et al., studied digital poultry systems combined with probiotic supplementation, showing improved growth and health metrics, but the design does not allow isolation of the digital component [18]. Cangar et al. explored quantitative precision farming in poultry, demonstrating potential for improved management but without direct side-by-side FCR comparisons between digital and non-digital regimes [19].

Other work, such as that by Adek and colleagues, documents improvements in related performance metrics. Integration of fuzzy logic with IoT sensors in brooding systems yielded better growth rates and energy savings [20]. Park and Zammit showed that digital monitoring in laying hens could be associated with improved fatty acid profiles and welfare indicators [21]. These findings suggest that digital technologies can have real benefits, but they do not directly address FCR. At present, therefore, empirical evidence showing feed conversion improvements attributable to digital twins in poultry is limited to a single five percent effect in broilers, and more expansive claims of 8–12 percent remain unsubstantiated.

Across the reviewed literature, there is limited but credible evidence that digital optimization can improve feed conversion ratio under specific conditions, with a single field study reporting a five percent improvement in a commercial broiler setting. What remains unmeasured is the robustness of this effect across production cycles, seasons, management regimes, and genetic lines, as well as its sensitivity to system reliability and operational context. This uncertainty constrains adoption because feed efficiency gains form the economic justification for digital twin investments, yet current evidence does not establish whether such gains are stable, reproducible, or generalizable. Methodological progress therefore depends on randomized, multi-cycle field trials that report feed conversion outcomes alongside contextual variables, reliability metrics, and cost–benefit indicators.

3.3. Performance Assessment Frameworks: Standardization Absent

The third research focus considered whether standardized performance assessment frameworks for digital twins in poultry exist. Here the gap is mainly conceptual. No empirically validated, poultry-specific performance framework was identified. Instead, the literature includes several generic frameworks from manufacturing and cross-sectoral digital twin research, and theoretical or partial frameworks with potential relevance to agriculture. Notably, most reviewed systems rely on unidirectional or near-real-time data flows for monitoring and decision support, while sustained bidirectional, closed-loop synchronization characteristic of fully realized digital twins is not documented in commercial poultry settings.

Liu et al. presented a generic digital twin evaluation model that attempts to “demystify” the concept and provides a multi-dimensional structure for assessing capability, cooperability, coverage and lifecycle attributes [10]. Psarommatis and May proposed a standardized approach to measuring performance and flexibility of digital twins in production environments [22]. Metcalfe and colleagues introduced digital twin maturity levels as a framework for conceptualizing and planning capability development [23]. Ferko et al. analyzed digital twin architectures in manufacturing settings and discussed standardization challenges [7].

Within the poultry and livestock domain, a few works move toward domain specificity. Essien and Neethirajan proposed multimodal AI systems for welfare assessment and discussed indices such as a Data Reliability Index and domain transfer metrics that could inform digital twin evaluation [11]. Neethirajan’s digital phenotyping work outlines important performance dimensions for broiler systems, including behavioral and physiological indicators [12]. Jia et al. described a virtual temperature collection method that meets specific accuracy criteria for layer house environments [24]. Despite these contributions, none of these proposals constitutes a complete, empirically tested, poultry-specific digital twin performance framework.

Across the broader body of literature, studies tend to pick and choose a few performance dimensions, such as prediction accuracy or energy use, according to the preferences of the research team. Few consider the combined economic, environmental, welfare, technical and social dimensions that would be needed for comprehensive evaluation. Without such a framework, there is no common yardstick for benchmarking systems, and performance claims remain difficult to verify, compare or synthesize.

The literature demonstrates substantial conceptual activity around digital twin evaluation frameworks, particularly in manufacturing and industrial contexts, but no empirically validated, poultry-specific performance assessment framework currently exists. What remains unmeasured is how digital twin systems perform when assessed simultaneously across technical accuracy, economic viability, environmental impact, animal welfare, and operational reliability. This lack of standardization constrains real-world adoption by making it impossible to compare systems, benchmark performance, or translate experimental results into decision-grade evidence for producers and regulators. Addressing this gap requires the development and field validation of integrated performance frameworks that are explicitly adapted to poultry production systems and applied consistently across deployments.

3.4. Failure Rates and Abandonment: The Void

The final domain examined relates to what is known about failure rates, reliability and abandonment of digital and precision livestock systems in poultry. Here, the evidence is especially sparse. Only three studies provide quantitative reliability metrics of any depth.

The work of Selle et al. [13] on a ceiling-suspended robotic monitoring platform in broiler houses documented severe data quality problems in early system versions. Between 40 and 50 percent of records had to be discarded in initial deployments because of missing or inconsistent values, spatial mislabelling, out-of-range readings or duplicates. Iterative improvements reduced this figure to around 10–20 percent, but even then, substantial post-processing was necessary. Montalcini et al. evaluated an active low-frequency tracking system for layers and reported highly uneven performance across barn zones, with some areas showing very low error rates and others exhibiting error rates around 0.5 with notable variance [14]. Raw data captured less than a quarter of real transitions, and heavy processing was needed to reconstruct actual movement.

Oliveira and colleagues modeled the dependability of cloud-based poultry house systems and compared architectures with and without redundancy [15]. Their models suggested annual downtimes on the order of a third for non-redundant systems, but significantly lower downtime for systems with hot-standby redundancy. Although informative, these figures arise from modeling rather than long-term empirical monitoring.

Beyond these examples, the literature on poultry precision systems contains qualitative observations about failure but almost no systematic quantification. Dust, ammonia, humidity and temperature fluctuations are acknowledged as challenges to sensor stability. Godinho et al., modeled signal transmission in poultry and highlighted the substantial path loss introduced by cages, animals and structural materials [16]. However, long-term maintenance demands, mean time to failure, repair frequencies, and especially system abandonment rates are essentially unreported. Notably, none of the poultry-focused studies report standard engineering reliability measures such as mean time between failures (MTBF) or mean time to repair (MTTR), metrics that are foundational in reliability engineering and essential for understanding long-term operational sustainability.

No identified study tracked farms over multiple years to determine how many systems were still in use, which were turned off or replaced, or why abandonment occurred. In papers describing digital systems, more than half of technologies were labeled as prototypes or development stage and fewer than one in ten as mature commercial products. The absence of abandonment data suggests a strong publication bias in favor of success stories and underreporting of discontinuation and failure.

Table 3. Key quantitative metrics reported across digital and precision poultry technologies.

Study	Species/System	Technology Type	Outcome Category	Metric	Baseline/Comparator	Digital/Intervention Value	Effect Size	Timeframe/Duration	Major Limitation
[5]	Crop agriculture (non-poultry)	Digital monitoring devices (RFID, cameras, edge units)	Carbon footprint	Annual device emissions	No digital deployment	0.2–0.3 Mt CO2e per year (France-wide)	Added emissions from digital deployment at national scale	Modeled over one year	Non-poultry; excludes cloud and network
[6]	Broiler houses	Conventional (no digital twin)	Carbon footprint	Farm-level GHG emissions	Standard LPG-heated house	383–524 t CO2e per year	Baseline emissions without digital tech	Annualized model	No digital comparison
[17]	Poultry with solar feeding	Solar-intelligent feeding infrastructure	Carbon footprint	Lifetime emissions	Conventional energy supply	~1697.85 t CO2e over 30 years	Embedded emissions of solar system	30-year horizon	Does not include digital twin stack
[8]	Broilers	ML + genetic algorithm control	Feed efficiency	Feed conversion ratio	1.64	1.561	~5% improvement; ~7.5 t feed saved	40-day cycle	Non-randomized; single site
[9,18]	Broilers	Automated environmental control	Performance and welfare	FCR, mortality, welfare	Conventional control	Small FCR change; improved welfare	Marginal FCR differences	Few cycles	FCR not primary endpoint
[9,20]	Chicks/broilers	Fuzzy logic + IoT brooding	Energy and growth	Energy use; growth rate	On/off control	16–40% less energy; better growth	Large relative savings	Brooding period	No FCR; small scale
[21]	Laying hens	Digital welfare monitoring	Welfare and product quality	Welfare; fatty acids	Standard monitoring	Improved welfare; altered profiles	Qualitative improvements	Short to medium duration	No FCR; limited context
[24]	Layers	BiLSTM virtual temperature collection	Technical accuracy	Prediction error	Dense sensor grid	0.25 °C error	High spatial accuracy	Months	Single variable focus
[11]	Laying hens	Multimodal AI (conceptual)	Framework metrics	DTS; DRI	None	Conceptual only	N/A	N/A	No field validation
[12]	Broilers	Digital phenotyping	Monitoring and prediction	Accuracy; latency	Conventional observation	Higher temporal/behavioral resolution	Qualitative enhancement	Various deployments	No economic metrics
[13]	Broilers	Farmer Assistant System	Data quality	Discard rate	Early FAS	40–50% discarded (early); 10–20% refined	Substantial improvement	31-month deployment	Single system; research managed
[14]	Layers	Active LF tracking	Tracking accuracy	Transitions explained	Raw data	23% raw; 91–99% processed	4–5 ×improvement	Two months	Strong spatial variability
[15]	Poultry houses (modeled)	Cloud/edge/fog architecture	System availability	Annual downtime	No redundancy	34.14% no redundancy; 9.176% with redundancy	~4 ×uptime improvement	One year model	No empirical data
[16]	Layers	Wireless sensor network	Signal performance	Path loss (dB)	Free space model	+22.5–24.9 dB loss	Substantial attenuation	Measurement campaign	Frequency specific
[11]	Poultry (multiple)	Mixed PLF systems	Abandonment	Abandonment metrics	None	No data available	N/A	N/A	Complete evidence void

consolidates the strongest numerical findings identified in the literature, spanning carbon footprint modeling, feed conversion outcomes, welfare and growth indicators, technical accuracy measures and system reliability metrics. For each study, the table presents the species or production system examined, the technology deployed, the outcome category assessed, baseline and intervention values, the magnitude of observed effects, the duration of evaluation and the major limitations constraining interpretation. Together these quantitative entries show that, while digital and precision systems generate promising technical signals, empirical evidence remains sparse, unevenly distributed across domains and rarely validated in large scale or long-term commercial settings.

Existing evidence indicates that data loss, sensor degradation, and system downtime are common challenges in poultry digital systems, yet quantitative reliability metrics remain sparse and highly system-specific. What remains almost entirely unmeasured are long-term failure distributions, maintenance burdens, technology obsolescence, and abandonment trajectories under commercial conditions. This absence of reliability evidence constrains adoption across all other domains, as production gains, carbon benefits, and performance assessments all depend on sustained system functionality over time. Closing this gap requires longitudinal monitoring of deployed systems, standardized reporting of reliability metrics, and explicit documentation of system discontinuation and its causes.

4. Discussion—Mapping the Evidence–Practice Chasm

4.1. Synthesis and Implications Across Four Domains

Taken together, evidence across carbon accounting, feed efficiency, performance frameworks, and operational reliability reveals a pronounced gap between the technical promise of digital twins in poultry production and the strength of the empirical foundation supporting that promise. From an environmental perspective, digital twins are frequently promoted as climate-friendly solutions, yet no poultry-specific lifecycle carbon assessments of digital infrastructures have been published. The closest available evidence, derived from crop systems, already indicates that digital devices contribute non-negligible emissions, underscoring the need for sector-specific evaluation.

From a production standpoint, empirical support is similarly limited. Only a single field study reports an improvement of approximately five percent in feed conversion ratio under narrowly defined conditions, while substantially larger gains are often cited in non-peer-reviewed discourse without transparent methodological substantiation. In terms of evaluation, generic digital twin frameworks developed in manufacturing and industrial settings have not been adapted or validated for poultry systems, leaving no shared basis for comparing architectures, algorithms, or deployment strategies. Evidence on operational reliability is even more sparse, confined to a small number of system-specific studies that nonetheless report substantial challenges related to data quality, sensor degradation, and limited system availability.

When considered through an integrated lens, these four domains are not independent weaknesses but interlocking components of a single evidence–practice chasm. Claims regarding climate benefit are inseparable from the magnitude and stability of feed efficiency gains, which themselves depend on the long-term reliability of sensing, communication, and control components. All of these outcomes, in turn, require assessment within coherent performance frameworks that allow environmental, economic, welfare, and technical impacts to be compared across studies. The current literature rarely addresses more than one of these dimensions simultaneously. By contrast, digital twin applications in manufacturing routinely incorporate lifecycle assessment, standardized reliability metrics, and formal evaluation frameworks, highlighting the relative methodological immaturity of the agricultural domain. This imbalance between rapid technical development and slow empirical validation undermines the credibility of strong performance claims, complicates decision-making for farmers, investors, and policymakers, and risks eroding trust if highly visible systems fail to deliver under the variability of commercial conditions.

4.2. Structural Reasons for Persistent, Interdependent Gaps

The persistence of these gaps reflects shared structural constraints rather than isolated oversight within individual studies. One major constraint is the divide between research and commercial contexts. Many digital twin deployments are evaluated in research barns with dedicated technical support, stable infrastructure, and simplified management practices. These conditions reduce exposure to the failure modes, maintenance pressures, and economic trade-offs that strongly influence performance in commercial poultry houses. Evidence generated under such conditions is therefore likely to overestimate robustness and underestimate practical barriers to adoption.

A second constraint arises from the developmental stage of many systems. Frequent hardware revisions, software updates, and short pilot durations limit opportunities for multi-cycle evaluation. Researchers and technology providers may also be reluctant to subject early-stage systems to rigorous trials that could yield ambiguous or unfavorable results. A third constraint is embedded in publication and reporting practices that favor positive outcomes, novel architecture, and short-term success. Studies documenting performance degradation, high maintenance burden, or technology abandonment are rarely published, biasing the literature toward systems that appear successful while obscuring the full distribution of outcomes.

A fourth constraint concerns the absence of shared metrics and reporting standards. Without common definitions of uptime, failure, abandonment, or carbon payback, and without standardized reporting of feed efficiency, welfare outcomes, or maintenance requirements, even well-designed studies remain difficult to compare or synthesize. These structural factors act across all four domains simultaneously. They limit comprehensive carbon accounting, constrain rigorous feed efficiency trials, discourage transparent reliability reporting, and impede the development of shared performance frameworks. Addressing them will require not only improved study designs but also changes in incentives, funding structures, and data-sharing norms.

4.3. Methodological Priorities for Future Work

Given the stakes associated with deploying complex digital infrastructures in livestock systems, several methodological priorities emerge. First, lifecycle carbon accounting tailored to poultry digital twins is essential. Such assessments should build on parametric approaches applied in other agricultural sectors while extending system boundaries to include cloud services, communication networks, hardware replacement cycles, and the environmental stresses imposed by dust, ammonia, and humidity in poultry houses. Full Scope 1, 2, and 3 accounting, aligned with IPCC guidance and ISO 14040/44 standards [25,26], is required if environmental claims are to be expressed as net effects rather than unqualified narratives.

Second, feed efficiency effects should be evaluated through properly randomized, multi-cycle field trials with sufficient statistical power and transparent reporting. These trials should disentangle the contributions of specific system components, including climate control, feeding algorithms, and health monitoring, while reporting contextual variables such as genetics, management practices, and climatic conditions. Where feasible, pre-registration of hypotheses and analysis plans would further reduce selective reporting.

Third, performance frameworks suitable for poultry systems must be developed and validated in real deployments rather than remaining at a conceptual level. Such frameworks should integrate technical metrics such as accuracy and latency, economic metrics such as cost per bird and payback period, welfare indicators, environmental impacts, and reliability measures within a unified structure that enables meaningful comparison across systems.

Finally, operational reliability and technology abandonment should be treated as primary research questions rather than incidental observations. Addressing these issues will require multi-year monitoring of deployed systems, with detailed logging of failures, repairs, configuration changes, and software updates, alongside structured engagement with users to document reasons for continued use, modification, or discontinuation. While concepts from dependability engineering and technology adoption research offer useful guidance, the poultry sector will require its own empirical datasets before robust generalizations can be made.

4.4. Small-Scale, Deeply Instrumented Deployments as Integrated Evidence Generators

Small, carefully instrumented research flocks provide a practical pathway for generating the multidimensional evidence currently lacking. A digital twin deployment involving approximately 2000 birds and operated over multiple production cycles can function as a living laboratory in which hardware energy use, cloud computing demand, sensor performance, feed and water intake, growth, mortality, welfare indicators, labor inputs, and maintenance requirements are measured at high temporal resolution. When combined with lifecycle carbon accounting, such deployments enable direct estimation of both the environmental cost of digital infrastructure and the environmental benefit of observed improvements in feed efficiency or related metrics.

The value of these deployments lies less in their scale than in their design. When planned with explicit evaluation objectives and harmonized protocols, a single well-instrumented flock can address multiple evidence gaps simultaneously by providing a common dataset for carbon accounting, feed efficiency analysis, reliability assessment, and field testing of candidate performance frameworks. This integrated approach contrasts with the prevailing pattern of disconnected case studies and single-outcome trials. While small-scale deployments cannot replace large commercial evaluations, they can substantially de-risk them by clarifying which performance claims are plausible, which design choices are consequential, and which variables must be monitored when digital twins are deployed at farm scale.

The deployment described in this section is presented as a conceptual evaluation blueprint (Figure 3) and has not yet been implemented or tested within the present study.

5. Broader Implications: Toward Cross-Species Digital Twin Research

The evidence–practice chasm identified for poultry digital twins is unlikely to be unique to this sector. Digital twin initiatives are increasingly reported across swine, dairy, beef, small ruminants, and aquaculture systems, often accompanied by similar expectations regarding productivity, animal welfare, and environmental sustainability. Yet, as in poultry, these expectations are typically supported by limited empirical documentation of lifecycle carbon footprints, field-validated performance gains, standardized evaluation metrics, or long-term operational reliability. Extending the analytical framework developed in this review beyond poultry therefore offers an opportunity to anticipate and address these challenges before they become entrenched across the broader landscape of precision livestock farming.

From a technical standpoint, digital twin architectures comprise elements with differing degrees of transferability across species. Core components such as sensor networks, edge and cloud data pipelines, storage infrastructures, and general machine-learning approaches can often be adapted with relatively minor modification. In contrast, physiological models, welfare indicators, and key performance metrics are closely tied to species-specific biology and management practices. Fast-growing broilers, long-lived dairy cows, and extensively grazed beef cattle differ markedly in production cycles, behavioral expression, and environmental exposure. In intensive swine and dairy systems, individual animal monitoring is more common and may enable richer digital representations, but this also increases demands on data volume, model generalization, and system reliability, particularly when animals move between pens, barns, or pastures. Aquaculture introduces additional complexity through underwater sensing and dynamic water quality processes, which place distinct stresses on hardware while raising analogous questions about reliability, carbon cost, and net environmental benefit.

A cross-species research agenda must therefore balance transferability with biological and operational specificity. The integrated framework articulated in this review, linking carbon accounting, resource efficiency, performance standardization, and reliability, provides a common template for evaluating digital twins across livestock systems. At the same time, meaningful application requires expansion of the variable space beyond conventional productivity indicators. Relevant dimensions include detailed behavioral metrics, physiological stress markers, high-resolution environmental and energy use data, product quality attributes, and explicit economic and social measures of adoption. Together, these variables shape whether digital twins deliver practical value under real farm conditions rather than remaining technically impressive but operationally marginal.

Longitudinal reliability and adoption studies are especially critical when extending digital twin research beyond poultry. Extensive and semi-intensive systems such as grazing beef, sheep, goats, and open-water aquaculture are characterized by greater environmental variability, intermittent connectivity, and incomplete data streams. These conditions increase the fragility of digital systems while also amplifying the potential benefits of timely intervention for welfare and sustainability. Multi-year, multi-farm studies that document failure modes, maintenance requirements, technology upgrades, obsolescence, and abandonment across species are essential to move the field beyond short-lived prototypes toward durable and trusted infrastructures.

By applying a shared evaluative lens while respecting species-specific biology and management contexts, cross-species digital twin research can generate a comparative evidence base capable of informing realistic, context-sensitive decisions. Such an approach clarifies where digital twins are most likely to deliver sustained benefit, how systems should be designed for robustness and efficiency, and which claimed advantages are supported by evidence rather than aspiration.

6. Conclusions

Digital twins in poultry farming have come to occupy a strikingly prominent position in the discourse of precision livestock farming and sustainable agriculture. They are frequently presented as transformative tools that can unlock measurable improvements in efficiency, environmental performance and animal welfare. Yet the evidence compiled in this synthesis makes clear that the empirical foundation supporting these claims is still remarkably narrow. No peer reviewed study provides a full lifecycle carbon assessment of the digital infrastructures involved, despite the centrality of environmental claims to the digital twin narrative. Only a single field validated study has demonstrated any improvement in feed conversion ratio, and that result is constrained by methodological limitations that preclude generalization across breeds, production systems or seasons. No validated or standardized performance assessment framework exists that can compare digital twin architectures or quantify their multidimensional impacts in poultry production. Perhaps most troubling, the field lacks any systematic documentation of reliability, failure patterns or abandonment timelines, leaving the long-term durability and real-world feasibility of these systems essentially unknown.

These gaps do not invalidate the potential of digital twins. Instead, they highlight an urgent need to shift the center of gravity in digital livestock research away from conceptual claims and prototype demonstrations toward rigorous, longitudinal and multidimensional evaluation embedded within commercial or near commercial production environments. Small scale but deeply instrumented deployments represent a practical and scientifically powerful route to generating the missing evidence. A research flock of approximately 2000 birds, outfitted with comprehensive sensing, energy monitoring, environmental instrumentation, feed and water intake measurements, behavioral analytics and reliability logging, can function as a living laboratory for systematic validation. Over multiple production cycles, such a deployment can quantify genuine feed efficiency effects, calculate full Scope 1 to 3 carbon footprints, document failure and maintenance patterns, capture abandonment or disengagement decisions and test emerging performance assessment frameworks in real operating conditions.

By grounding digital twin development in evidence derived from real production dynamics, the field can transition from aspirational rhetoric to scientifically defensible practice. In doing so, digital twins have the potential to mature into robust tools that meaningfully support sustainable intensification, reduce environmental burdens and promote higher welfare poultry production. The opportunity now lies not in advancing ever more elaborate conceptual architectures, but in generating the empirical record required to demonstrate that these systems deliver on their promise when exposed to the complexity, unpredictability and constraints of commercial agriculture.