Why Logistics AI Still Needs “Experts in the Loop”

In the world of logistics and freight, artificial intelligence (AI) is rapidly expanding its presence, but the critical question remains: how much control should be relinquished to these AI tools? On the latest Lean Quarterly Dive, host Malcolm Harris interviewed Alfonso Quijano, CTO of Lean Solutions Group, to explore this issue.

Quijano emphasized the importance of a model known as "experts in the loop," where human judgment is not seen as a stopgap for AI limitations, but rather as the very element that makes AI worthwhile.

Quijano believes that AI has revolutionized productivity across the logistics industry, but its true potential can only be realized when paired with human expertise. He stated that AI should not be allowed to make decisions unsupervised, but rather should work in tandem with human experts. The model splits the workload, with AI handling repetitive tasks and humans providing judgment to guide the system, ensuring its continuous improvement and reliability in more complex scenarios.

Standard operating procedures (SOPs) are currently undergoing a transformation as AI becomes integrated into more workflows. Most companies have yet to develop SOPs that can be effectively utilized by AI, resulting in a static document that is rarely reviewed. Quijano drew a parallel to CLAUDE.md or AGENTS.md files used by companies like Anthropic and OpenAI, noting that these files function as living sources of truth, constantly evolving alongside the AI system.

While AI excels in document-heavy workflows such as invoice intake, discrepancy checks, email routing, and accounts payable/receivable tasks, there are clear use cases where AI outperforms humans. For instance, an AI system is less likely to miss errors compared to a human. However, the handoff point for these tasks often involves dollar thresholds and client-specific rules.

When an invoice includes unfamiliar accessorial fees, an expert must verify the details, and their decision then trains the system on handling similar cases in the future.

Quijano expressed cautious optimism about AI's potential in carrier sales, but cautioned against full autonomy due to the critical role of human trust-building in the early stages of a client relationship. He emphasized that trust is built through human interactions, which are difficult to automate. AI can triage low-value noise in spot-market inquiries, but outbound sales remain a challenge, with companies still determining the optimal approach to automation.

According to Quijano, decision intelligence combines 14 years of operational knowledge with an AI layer that acts as an embedded assistant. This approach, which can be presented as a sidebar, desktop app, or always-on presence, watches for ambiguous data or risky decisions in real-time and flags them before they become problems.

New hires at Lean typically offload half of their tasks to a manager during their first day on the job, but with decision intelligence providing suggestions based on SOPs, operational history, and peer behavior, this drag in onboarding is significantly reduced. Across document-based automations, this can result in a savings of 65-70 hours of time per month, or approximately a quarter of an operator's working hours.