________________________________________________________________________________________________________________________
The situation isn ’ t that much better for shipping customers of ugly freight , who can face numerous issues if an item ’ s dimensions or weight are not calculated correctly . A price estimate can prove wildly inaccurate , or an unprepared carrier could be unable to fulfill a shipment . In either case , discrepancies between different items of similar shapes and sizes ( a couch built in 2020 is vastly easier to move than a sleeper sofa from the 1980s ) can escape a carrier ’ s eye with serious consequences . AI could do better if it ’ s equipped with high-quality data .
Ensuring data quality
Recent advancements in deep learning algorithms have vastly transformed AI ’ s ability to conduct complex visual recognition tasks , with Google ’ s Lens technology now processing over 20 billion searches a month with a 92.6 percent accuracy rate . If the same technology could be applied to larger-thanparcel shipping , it could fill gaps that carriers may otherwise miss and may even amend a user ’ s inaccurate dimensions or weight calculations for a better , more accurate quote . Yet , this accuracy is contingent on strong underlying training and inferencing data .
Unfortunately for the logistics space , this data seems easier to find than it ultimately is . Despite mountains of proprietary historical pricing and shipping data , most of this data isn ’ t properly structured or suitable for AI training . A recent study found more than 45 percent of newly created data records fall into this category , with at least one critical fault . Before utilizing datasets for model training ( or evaluating the dimensions of an ugly freight piece ), logistics pros must first comb through their training data for missing values , outliers , or other abnormalities that may cause an AI model to return inaccurate results once online . These values can be replaced with high-quality synthetic data , but only if they ’ re caught before model training . This is why data validation is the most timeconsuming step in AI development , and why shoring up one ’ s data warehouse is the most critical step to leverage AI . Ugly freight shipping may still be transformed by the
16