Why Lean Solutions Group Is Betting on ‘Experts in the Loop’

Lean Solutions Group has grown from roughly 700 employees in 2018 to more than 10,000 today, spread across Colombia, Guatemala, Philippines, and beyond.

The company built that scale by solving a straightforward problem for freight brokerages: reducing cost per load by moving back-office functions to nearshore labor markets where the economics made more sense. According to CTO Alfonso Quijano, the calculus that powered that growth has shifted.

The 40% cost arbitrage that originally drew brokerages to Lean Solutions Group (LSG) is no longer enough. Clients are now pushing for 60% or 70% savings, and they want those gains delivered without disrupting operations. That’s where artificial intelligence enters the picture, though not in the way many in the industry might expect.

“AI is all the rage recently, but there are not a lot of people who are talking about the real stuff that needs to happen behind AI implementations,” Quijano said in an interview with FreightWaves’ Editorial Director, J. P. Hampstead.

Quijano’s central argument is that logistics is too fragmented and too varied in its processes for any single AI product to serve a broad customer base without significant customization.

Before LSG standardized its service offerings, the company supported more than 180 distinct job functions across the transportation and logistics industry (many of them minor variations on roles like track and trace that individual brokerages had tailored to fit their own workflows).

That fragmentation, Quijano says, is exactly what causes AI-first solutions from outside the industry to buckle. “You can’t just create one product that covers a wide set of different customers without change,” he said. “Each one of them requires some type of tweak and custom implementation that breaks wide scale product adoption.”

It’s a dynamic that mirrors the competitive tension LSG navigated in its early days, when rival brokerages sharing the same service provider demanded firewalled networks, branded workspaces, and siloed SOPs to protect their operational identities. That same instinct now applies to AI deployments.

“We hold a playbook for one of the most massive change management implementations that the industry has seen in terms of workforces,” Quijano said. “We know how people work, and how the work needs to change to effectively adopt AI.”

Quijano was blunt about the limitations of large language models in logistics operations, particularly when companies attempt to deploy fully autonomous AI workflows.

“The ability for it to make good high quality judgment decisions is still very far off from reality,” he said, adding that when exceptions arise in an autonomous workflow, the cost of undetected errors can cascade from the TMS through accounting and all the way to the customer.

He likened the problem to a common-sense failure: an AI chatbot advising someone to walk to a car wash rather than drive the car that needs washing. The anecdote, drawn from a viral internet trend, illustrated his broader point that AI output is probabilistic, not intelligent. “AI is not smart by default,” Quijano said.

“It’s a technology that estimates what the next word should be based on the input.” The risks compound at scale. When companies hand large volumes of work to autonomous AI agents, according to Quijano, they often end up spending far more time reviewing, correcting, and amending errors than they saved.

LSG’s alternative framework rejects the common industry shorthand of “human in the loop,” which Quijano sees as reductive. “Human in the loop has the implication of a super smart process or AI-included process that needs babysitting and that humans are just there to input ‘approve, approve, approve,’” he said. “That’s not the way to look at it.”

Instead, LSG uses the term “experts in the loop” to describe a model in which the people who previously performed operational tasks are trained as specialists responsible for identifying outlier situations, teaching the AI to handle new scenarios, interpreting performance metrics, and ensuring alignment with SLAs.

It’s a substantive change in job description, not a demotion to button pusher. Quijano pointed to LSG’s existing QA infrastructure as the foundation for this model. LSG uses a team of roughly 200 people deployed across client accounts, and those QA roles are being reworked to audit both AI and human output in accounts with active AI implementations.

“It’s an investment that you need to make in order to ensure that the AI works, at least for now until you reach that level of full autonomy,” he said. “Nobody knows, not even Jensen Huang from NVIDIA or Sam Altman, when the AI is gonna be fully autonomous.”

Through LeanTek AgentEdge and LeanTek Connect, LSG is launching AI capabilities designed to work proactively alongside operators rather than waiting to be queried. The distinction, Quijano explained, is the difference between an operator who spots a mistake and sends a screenshot to ChatGPT for analy