Build or buy AI quoting: what the giants show

The most credible proof that AI quoting works in freight comes from companies that built their own. That evidence is public, it is recent, and any forwarder weighing a build deserves an honest account of it. This guide lays out what the largest players have disclosed, what their programmes required, and a practical frame for the decision. Nexcade sells a bought alternative, so read with that in mind; every claim below traces to the public sources listed at the end.

What the builders have disclosed

C.H. Robinson started embedding large language models in workflows in 2023. Its October 2024 announcement described generative AI reading incoming email and automating tasks across the freight lifecycle, starting with price quotes. By mid 2025 it reported a fleet of more than 30 AI agents at scale, with quoting agents past one million quotes and response times around 32 seconds. By 2026 it described hundreds of connected agents orchestrated by its Always On Logistics Planner. Two details matter for anyone studying the build path. First, the agents sit on top of pricing engines the company built earlier: Procure IQ, Market Rate IQ, and a Dynamic Pricing Engine. Second, the company describes a disciplined portfolio approach: agents get built where ROI is tangible.

DSV presented an AI-supported quote-and-rate workflow for Air & Sea at its May 2026 Capital Markets Day: quote requests arrive by email, AI extracts details and missing information, queries the rate engine, and returns a structured quote through the DSV portal, with steps logged in a sales copilot under human control. DSV frames the opportunity at 3 to 3.5 million spot quote requests per year and targets roughly DKK 6 billion of AI and technology productivity by 2030. The platform strategy underneath is explicitly in-house: a count-to-one consolidation onto its own TANGO and STAR systems, reducing third-party software dependency.

DP World offers the most instructive architectural detail, published in a vendor’s press release: it is developing in-house quoting tools while partnering with cargo.one to aggregate airline rates directly into its systems. Even a build at global scale buys the rate layer where a market already exists.

What the build path actually requires

Read together, the disclosures describe programmes with common ingredients:

• Years, plural. C.H. Robinson dates its first agents to 2023, built on pricing engines from 2020 and 2021. DSV’s workflow rides on a multi-year platform consolidation.
• Owned data and pricing foundations. The agents are the visible layer; rate engines, pricing science, and clean data pipelines sit underneath.
• A large permanent technology organisation. These are core-business programmes at companies with thousands of engineers, with executive sponsorship measured in capital-markets commitments.
• Selective buying anyway. DP World’s rate aggregation deal shows the pattern: build the differentiating layer, buy commodity layers.

None of this says smaller forwarders cannot build. Mid-size forwarders build real systems, most often orchestration layers and customer portals stitched over an existing TMS, and some work well. The honest questions are narrower: which layers differentiate your business enough to own, and what does maintenance cost once volumes, carriers, formats, and edge cases drift? Industry guides put full custom forwarding systems at 12 to 24 months and significant six-figure budgets before maintenance, and the maintenance is the part that compounds.

A practical frame

A quoting build decomposes into layers, and each layer has a different build-versus-buy answer:

The pattern at the top of the market supports the middle rows: even builders buy rate layers, and even buyers keep pricing judgement in-house as configuration. The genuine fork is the workflow-agent layer, and the deciding variables are engineering capacity, volume, and how much of the quoting pain lives in edge cases. Edge cases are where in-house tools quietly die: the agent email chase, the out-of-gauge request, the customer with a nonstandard SOP, the TMS writeback mapping that breaks on a charge code.

Where Nexcade sits in this picture

Nexcade is a bought workflow-agent layer: a team of AI agents covering triage, quotations, tenders, bookings, and analytics, running over a forwarder’s existing rate sources and TMS, with review gates and configurable autonomy. The pricing policy stays the forwarder’s own. The architecture mirrors what the builders converged on, with the engineering maintained by a vendor whose whole business is the edge cases.

For the wider map of categories, see the field guide to freight quoting software.