How Flexport is using AI to cut ocean freight costs by 8-10%

Executive overview

Logistics is one of the most labor-intensive industries in the world, yet it runs on email, phone calls, and manual data entry. Flexport, a $2B revenue freight forwarder, is systematically replacing that work with AI — from container routing to warehouse address verification.

The core bet: automate enough of the freight-forwarding layer that the cost savings can be passed to customers, driving volume growth in a scale-economies business.

The company expects to automate 80-95% of its work by end of next year, cutting ocean freight prices by 8-10%.

AI applications already in production

• Natural language querying of shipment data — customers type questions, get charts and tables without SQL or dashboards; eliminated ~25% of account management time spent on reports
• Container routing optimization — ML model selects ship, route, and contract; saved 2% of ocean freight spend while improving transit time 20%
• Automated container rebooking — system runs 10x per day, finds canceled containers and moves forward bookings that were scheduled a week out; produces the 20% transit time gain
• LLM agent verifies warehouse delivery addresses and books appointments via email and voice before every delivery not made in the past 3 months
• Sentiment detection in customer messages triggers automatic escalation to managers when a customer appears upset
• LLM-based parser for logistics contracts — large Excel files with thousands of rows and multiple tabs, previously requiring manual extraction

Automation trajectory

• Started 2025 at 20% of work automated
• Target: 50% by end of 2025
• Original ceiling estimate: 80% automatable
• Revised ceiling: 90-95%, with further upside as LLMs improve
• Labor costs in freight forwarding represent ~10% of total shipping cost — full automation could reduce international freight prices ~8-10%

Scaling AI skills internally

• Employees can join a 90-day AI bootcamp: one day per week, covering vibe coding, Cursor, Streamlit, and workflow automation
• Managers approve participation; the promise is a 10x productivity return
• Program started organically in the Amsterdam engineering office, now going global
• Goal: domain experts build and automate their own workflows without relying on engineers
• Hackathons now run twice yearly; ~90% of projects are LLM-based (vs. a handful 18 months ago); hackathon ideas increasingly feed directly into the product roadmap

Incumbents vs. startups in AI adoption

• Incumbents have three structural advantages: proprietary data at scale, domain knowledge to identify which problems are worth solving, and existing customer distribution
• Flexport's additional edge: owns its own tech stack, so AI can be integrated anywhere without waiting on third-party software vendors
• Most logistics competitors treat technology as an IT purchase — they can't modify the codebase
• Startup disadvantage: must beg for customer data, earn compliance trust, and then acquire customers before any AI can operate at scale

Capital allocation and headcount discipline

• Raising a large round is fine as long as it is an up-round (price per share is the only metric that matters, alongside control)
• Money has a strong tendency to spend itself — defaulting to hiring to solve problems leads to bloat and slows the company
• Recommended practice after a large raise: 90-day hiring freeze to set the cultural expectation that money does not solve problems
• Headcount discipline is harder than it sounds and requires active, ongoing top-down pressure

Flexport's 2035 roadmap

• Current: ships to/from 147 countries, employees in 22 countries
• Target: own-staff coverage of 95% of container trade by 2028
• Target: employees in every legally accessible country by 2035
• Vision: logistics as a utility — any cargo, any mode, any quantity, globally, fully accessible via API or voice, at the lowest automated cost