Matching Explainability

Quantifies the marginal contribution of each input variable to the final match decision. For every carrier-load pairing, the engine outputs a ranked list of factors—such as on-time performance, price competitiveness, or lane affinity—with their respective influence percentages. This allows brokers to instantly understand that Carrier A was selected over Carrier B primarily due to a 15% higher predicted acceptance probability and a 10% lower detention risk score, despite a slightly higher rate.

Generates 'what-if' scenarios to clarify decision boundaries. The system articulates the minimal changes required to alter the outcome, such as: 'Carrier B would have been selected if their safety score exceeded 92 or if their estimated time of arrival was within a 30-minute window.' This technique, rooted in causal inference, helps shippers understand the precise thresholds governing automated decisions and provides actionable feedback to carriers seeking to improve their match eligibility.

Converts complex mathematical optimization results into plain-English rationales using large language models. Instead of displaying raw vectors or constraint satisfaction logs, the engine generates a human-readable summary: 'Acme Trucking was matched to Load #4521 because they are the highest-rated carrier on the Chicago-to-Dallas lane with available hours and a refrigerated trailer, saving an estimated $340 compared to the next-best option.' This bridges the gap between algorithmic complexity and operational trust.

Maintains an immutable, timestamped record of every input, constraint, and weight used in the matching process. Each decision is logged with:

• Input snapshot: All data points available at decision time (rates, ETA predictions, carrier scores)
• Constraint evaluation: Which hard constraints (equipment type, hazmat certification) passed or failed
• Objective function state: The multi-objective weights applied (cost vs. speed vs. carbon) This auditability is critical for regulatory compliance and defending against claims of biased or unfair matching.

Goes beyond single-decision explanations to reveal the engine's overall behavior patterns. Techniques like SHAP (SHapley Additive exPlanations) dependence plots and partial dependence plots show how the model's matching preferences shift across different contexts. For example, a global analysis might reveal that the engine's sensitivity to price increases sharply when market capacity is abundant, while carrier reliability dominates during peak season. This macro-level transparency enables strategic tuning of the matching algorithm.

Provides a visual dashboard where users can explore match decisions dynamically. Brokers can drill down into a specific match, toggle the visibility of different feature attributions, and adjust hypothetical parameters to see how the decision would change. This interface often includes confidence bands around predictions—such as a ±15 minute ETA uncertainty—so users understand not just what the model decided, but how certain it was. This transforms explainability from a static report into an exploratory tool for operational learning.

The broader field of techniques used to understand how complex models make decisions. In freight matching, this involves using feature attribution methods like SHAP or LIME to quantify the impact of factors such as lane density, carrier scorecarding, and detention risk scoring on a specific match recommendation. This ensures that a model's logic can be audited by human operators, moving beyond a simple 'black box' output.

An automated performance evaluation system that is a primary input for explainable matching. It rates carriers on metrics like on-time delivery, safety records, and digital compliance. Matching explainability directly surfaces these scorecard components, providing a clear, data-driven justification for why a high-performing carrier was prioritized over a lower-scored one for a critical load.

A mathematical framework that finds the optimal match by balancing conflicting goals like lowest cost, fastest transit, and lowest carbon emission. Explainability is critical here to unpack the trade-offs the algorithm made. It can show that 'Carrier A' was selected over 'Carrier B' not because of price, but because their superior predictive ETA and lower emissions profile outweighed a 2% cost difference.

An algorithmic engine that finds valid carrier-load pairings by strictly enforcing hard requirements. Explainability in this context provides a clear 'certificate of compliance,' showing that a match is valid because it explicitly satisfies all constraints: equipment type (reefer), time windows (0700-0900), and hazmat certification. If no match is found, it explains which constraint could not be satisfied.

A machine learning system that infers a carrier's implicit lane and load type preferences from historical booking data. Explainability connects a match to this profile, showing a carrier why they received an offer: 'This load was tendered to you because your profile indicates a 95% acceptance rate for this lane density and your continuous move optimization model shows it minimizes your idle time.'

A predictive model that forecasts the likelihood of a primary carrier refusing a shipment. When a match is made with a secondary carrier, explainability uses this model's output to justify the decision, stating: 'Carrier A was bypassed due to an 85% predicted rejection risk based on their current hours-of-service and historical aversion to this receiver's detention risk score.'