Prioritizing Road Freight Risks: An Integrated Analytic Hierarchy Process-Likelihood Framework

Road freight is essential to supply-chain continuity but is exposed to interconnected operational, infrastructural, financial, regulatory, environmental, and cargo-related risks. These risks can disrupt deliveries, increase costs, damage goods, and reduce service reliability [1, 2]. Their management is particularly challenging because freight risks are multidimensional, interdependent, and context-dependent, while the information needed to estimate their likelihood and consequences is often incomplete [2, 3]. Under such conditions, expert judgment remains important, but it should be incorporated into a structured prioritization framework that makes priorities transparent and supports defensible resource allocation decisions [4, 5].

This challenge is particularly relevant to Vietnam, where the rapid expansion of freight activity places increasing pressure on transport and logistics systems. In 2024, the country transported approximately 2.67 billion tons of freight, representing a 14.0% increase from the previous year, while freight traffic reached 545.1 billion ton-kilometers [6]. Road transport remains central to domestic freight distribution because it connects production centers, ports, inland container depots, warehouses, and consumer markets. This role is especially pronounced in Ho Chi Minh City and its surrounding industrial region, where concentrated port, manufacturing, and consumption activities generate substantial truck traffic and expose operators to congestion, restricted delivery windows, and bottlenecks at port access points and on urban roads. These operational pressures extend beyond physical transportation, as freight firms must coordinate warehousing, documentation, customs clearance, cargo handling, and regulatory compliance among multiple parties. The challenge is compounded by the fragmented structure and limited technological capacity of many domestic logistics enterprises, while Vietnam ranked 43rd among 139 economies in the World Bank’s 2023 Logistics Performance Index, behind several regional competitors [7, 8]. Because resource-constrained firms cannot address all potential risks simultaneously, identifying which risks require earlier intervention offers greater managerial value than compiling an extensive but unranked risk register.

Previous studies have identified diverse risks in road freight and supply chain operations and applied multi-criteria decision-making methods to support risk assessment [9, 10]. Among these methods, the Analytic Hierarchy Process (AHP) is useful because it structures complex decisions, derives relative weights from pairwise comparisons, and permits the examination of the consistency of expert judgments [11]. AHP and related approaches have consequently been used to prioritize supply-chain, transport, customs, and logistics risks [12, 13]. However, conventional AHP rankings primarily represent the relative importance assigned to risk factors. A highly important risk may occur infrequently, whereas a moderately weighted risk may require more immediate attention because it is encountered repeatedly. Studies that consider occurrence often embed it as one criterion within more complex models, leaving limited evidence on a transparent procedure that combines hierarchical importance with separately assessed occurrence in data-constrained road freight settings. Moreover, the robustness of the resulting rankings is not always systematically tested.

To address these limitations, this study develops an AHP-likelihood framework for prioritizing enterprise-level road freight risks in Ho Chi Minh City. Risks identified from the literature were refined through interviews with logistics professionals and screened to distinguish risk sources and events from downstream consequences. Expert pairwise comparisons were used to derive group-level AHP weights, which were combined with zero-based perceived-likelihood scores to calculate a relative Risk Priority Index (RPI). Robustness was evaluated using category-weight scenarios, alternative normalization and aggregation rules, and joint simulations of weight and likelihood uncertainty.

The study makes three contributions. First, it develops a contextually validated hierarchy of road freight risk sources and events for the participating Ho Chi Minh City logistics enterprises. Second, it distinguishes expert-perceived relative managerial importance from perceived likelihood and examines the incremental ranking information contributed by likelihood. Third, it evaluates the stability of the leading priorities across alternative analytical specifications and simultaneous uncertainty.

Table 1. Road freight risk sources/events and direct supporting evidence

Twenty logistics professionals from 10 enterprises were purposively selected for the quantitative assessment, including seven participants who had previously validated the hierarchy. Participants had at least one year of direct experience in logistics coordination, documentation, cargo handling, road freight operations, or related managerial work in Ho Chi Minh City. The qualitative interviews and quantitative assessments were conducted from August to December 2025. Table 2 summarizes the panel characteristics.

Table 2. Available profile of the Ho Chi Minh City expert panel

The panel included six managerial or supervisory participants and 14 operational or professional participants, with experience ranging from one year to more than ten years. All experts were located in Ho Chi Minh City, and their judgments are interpreted within this urban logistics context. The experts were assigned equal influence during aggregation, and personal and organizational identities are omitted for confidentiality.

The questionnaire comprised two components. First, experts evaluated the general relative managerial importance of the seven categories and the factors within each category using Saaty's nine-point pairwise-comparison scale [11]. These comparisons captured overall managerial importance for prioritizing intervention rather than separate measures of severity, financial impact, controllability, urgency, or disruption potential. The revised hierarchy contained 25 risk sources and events after source-event-consequence screening. Because RTC1 was the only retained factor in the financial-risk category, no within-category pairwise comparison was required for RTC; its local weight was fixed at 1. Second, experts rated the perceived likelihood of encountering each factor during the preceding 12 months using five ordered bands based on the proportion of relevant shipments or operations affected: 0% to <5% (1), 5% to <10% (2), 10% to <20% (3), 20% to <40% (4), and ≥40% (5).

For each pairwise comparison, the majority direction and modal intensity of the 20 responses were used to form a representative reciprocal group matrix. Local weights were obtained by column normalization and row averaging, and global weights were calculated as the product of category and local weights. Consistency was assessed for the seven group matrices and for each expert's category-level matrix and five within-category matrices containing at least three factors. The maximum CR across these matrices was used as the individual-level consistency measure. The two-factor RMT matrix is inherently consistent, while a within-category CR is not applicable to the single-factor RTC category.

$w_{i j}^G=w_i^C \times w_{i j}^L$

where, $w_i^C$ denotes the category weight and $w_{i j}^L$ denotes the factor’s local weight within that category.

Judgment consistency was evaluated using:

$C I=\frac{\lambda_{\max }-n}{n-1}, C R=\frac{C I}{R I}$

where, $\lambda_{\max }$ is the maximum eigenvalue, $n$ is the matrix order, and RI is the corresponding random consistency index. Aggregated and individual matrices with $\mathrm{CR} \leq 0.10$ were considered acceptable [11].

3.3 Perceived-likelihood integration and sensitivity analysis

Perceived likelihood was assessed separately from the AHP comparisons for the preceding 12 months. Experts selected one of five ordered bands defined by the estimated proportion of relevant shipments or operations affected: 0% to <5% (1), 5% to <10% (2), 10% to <20% (3), 20% to <40% (4), and ≥40% (5). These common anchors improved comparability across respondents while retaining an ordinal expert-judgment format. To make the lower endpoint equal to zero, the primary normalized score was calculated as follows:

$L_i=\left\{\left[\left(\Sigma_j x_{i j}\right) / N\right]-1\right\} / 4$

where, L_i is the zero-based perceived-likelihood score for factor i, x_ij is expert j's ordinal 12-month rating, and N = 20. The transformation maps a mean rating of 1 to 0 and a mean rating of 5 to 1. Although the categories use quantitative percentage anchors, L_i remains an expert-rated ordinal likelihood score and is not interpreted as an observed annual probability or incident rate.

The RPI was calculated by multiplying the global AHP weight of each risk factor by its normalized perceived-likelihood score:

$R P I_i=G W_i \times L_i$

where, $GW_i$ is the global AHP importance weight, and $L_i$ is the zero-based perceived-likelihood score. Higher RPI values indicate greater relative priority for managerial intervention. The RPI is not an absolute probability of loss, an expected-loss measure, or a conventional probability-impact score; it is a multiplicative screening index combining relative managerial importance with expert-rated 12-month perceived likelihood.

Robustness analysis comprised four components. First, a one-at-a-time test increased and decreased each of the seven category weights by 10%, with the remaining weights proportionally rescaled to sum to one. This produced 14 category-weight scenarios while holding local weights and likelihood scores constant:

$w_c^{(s)}=(1+\delta) w_c$

The weights of the remaining categories were proportionally adjusted to preserve their relative structure and ensure that the complete weight vector summed to one:

$w_i^{(s)}=w_i\left(\frac{1-w_c^{(s)}}{1-w_c}\right), i \neq c$

where, $w_c$ and $w_i$ are the baseline category weights, while $w_c^{(s)}$ and $w_i^{(s)}$ are their corresponding weights under scenario $s$. Scenario-specific global weights were then obtained by multiplying each adjusted category weight by the unchanged local weight of the corresponding risk factor:

$G W_{i j}^{(s)}=w_i^{(s)} \times L W_{i j}$

The RPI was subsequently recalculated for every scenario as:

$R P I_i^{(s)}=G W_i^{(s)} \times L_i$

Second, the zero-based likelihood transformation was compared with the alternative x/5 rule. Third, a diagnostic equal-weight additive score combined min-max-scaled global importance with zero-based likelihood, testing whether conclusions depended on multiplication. Fourth, a 20,000-iteration Monte Carlo analysis varied all category and local weights simultaneously using independent U(0.90, 1.10) multipliers, renormalized weights at the category and within-category levels, and bootstrapped 20 likelihood ratings for each factor from its observed five-level response distribution. A fixed seed of 45364 ensured reproducibility. Robustness was summarized using Spearman rank correlations, maximum rank movement, top-five and top-ten inclusion rates, and changes in the leading set.