Intelligent Freight Discovery

25%

Higher booking conversion on loads surfaced through Matches compared to manual search

15%

Faster load coverage across key markets due to improved routing suggestions

40%

Increase in repeat bookings from carriers who used Matches at least three times in a month

Machine Learning

Data-Driven UX

Predictive Analytics

User Research

Booking Optimization

Marketplace Growth

Overview

An iterative redesign of Arrive’s Carrier Portal homepage to surface high-probability matches through ML-driven recommendations, reducing search time, increasing instant bookings, and aligning the portal with modern 3PL technology trends.

Discovery

Carriers relied on reps manually using Arrive’s proprietary XLR8 matching engine to broker freight. The engine applied predictive algorithms considering historical carrier performance, lane preferences, equipment availability, and dynamic pricing models. While powerful, these insights were siloed and inaccessible at scale in the portal.

I collaborated with Product, Engineering, and Data Science to map algorithm outputs to carrier needs. Understanding the input features, probability scoring, and constraint validation helped define which signals mattered most to carriers and how they should be surfaced without overwhelming them.

ML application painpoints

Product owners initially pushed to surface matches on Find Loads, but real-time search behavior on that page meant dynamically pulled results could be displaced by recommended matches, creating friction and cognitive load. Surfacing high-probability matches on the homepage allowed controlled experimentation, data capture, and iteration without impacting core workflows.

Collaborative analysis and discussion

Ideation

Here’s what we learned and drew from when shaping the technical side of the matching engine. Machine learning became the core lever. Our algorithms score the booking likelihood of every carrier–load pair by looking at past acceptance patterns, seasonal demand, lane density, and carrier constraints. Real-time validation keeps the list clean by surfacing only the loads a carrier actually qualifies for based on insurance, equipment, and permissions.

The ranking model balances dynamic pricing, margin, and pickup-to-delivery efficiency so carriers see the most valuable options while brokers capture revenue. Incremental updates feed new data without reprocessing the entire database, which keeps latency low even as volume grows. ML outputs flow directly into the interface, mapping to UI components in a way that keeps relevance obvious and performance aligned with the system’s legacy requirements.

The concepts below guided us to what mattered most.

Benchmarking of ML applications

Hackweek Winning Concept - Carriers discover real-time visibility on volume, hot markets, lane price comparison assistance, reloads visibility, all while improving radius search limitations.

I considered the interplay of ML-powered confidence scores, carrier workflow psychology, and interface clarity. “Book It Now” buttons made sense during active search sessions, but homepage recommendations were meant for discovery. My customer-focused goal was balancing actionable visibility with minimal cognitive load, ensuring carriers could review detailed constraints, equipment requirements, and lane metrics before booking.

Design Execution

The first iteration of Matches tested whether Carriers would notice and engage with potential loads, moving beyond legacy table patterns that crowded the interface and created cognitive overload. I wasn’t thrilled with the use of the Oreo effect to draw attention, feeling the dark surfaces were overused and diluted hierarchy, but Product won the argument and we used it to test engagement.

Based on these insights, the second iteration refined the experience to three focused cards, highlighting both load and lane interest while maintaining a clear action hierarchy. This balanced visibility and discoverability, letting Carriers explore opportunities without feeling overwhelmed, and provided a structured framework for data-driven decision-making in future iterations.

What we delivered

Looking ahead, if the business continued iterating on the portal roadmap, I would focus on a homepage experience that emphasizes clarity, trust, and informed decision-making. The design would surface content in a way that highlights value, encouraging users to explore and understand opportunities before booking. Concepts like grouping by state and equipment filters could enable controlled experimentation with ML-driven recommendations, without overwhelming core workflows. Iterative testing and engagement tracking would validate these approaches, providing insights to guide future enhancements and optimize user confidence and efficiency.

Solution

Users can review load requirements before booking, and the system continuously refines match predictions based on real-time engagement and acceptance data. This approach merges human-centered design with AI-powered intelligence, creating a scalable, agile solution that aligns UX, business goals, and technical constraints.