project44 Connects AI Analyst Mo to Snowflake for Grounded Logistics Answers

project44 said on November 10 that it has integrated Mo, its AI supply-chain analyst, with Snowflake, so that answers to logistics questions are grounded in project44 data rather than in a model’s recalled figures. In a Chicago press release, the company said the integration pairs its logistics ontology, semantics, and network data — about 1 trillion shipment records over the last decade, it said — with Snowflake Cortex AI. Teams, it said, can ask questions in plain language and get answers and recommended next steps.

The company described Mo as an “AI agent harness”: the guardrails, workflows, and execution layer around a language model. In logistics, it said, even straightforward terms can mean different things to different companies. On-time performance, for example, may depend on customer-specific definitions, contract terms, shipment modes, and the milestones used to measure transit time; some shippers measure transit from cargo-ready date and others from pickup. Mo, project44 said, uses those customer definitions to interpret a question, and then uses Cortex Analyst to translate the interpreted question into SQL and retrieve results from project44’s data in Snowflake.

“Reliable data, reliable signal, reliable action. That’s the sequence, and project44 provides it with our logistics ontology, customer contracts, a decade of network data, and real-time milestones. Cortex AI retrieves the figures from the data. Together, that’s what turns a capable model into something operations teams can trust with real decisions,” said Jett McCandless, founder and chief executive officer of project44. Will Allen, head of Snowflake CoCo and CoWork at Snowflake, said the combination is meant to help teams “ask questions in business terms and use the answers to respond to disruptions and identify opportunities to reduce costs.” Doug Cantriel, head of North America transportation at Ford Motor Company, said Mo helps put “accurate, data-grounded answers directly in the hands of the people who need them, without the back-and-forth of pulling reports or waiting on analysis.”

project44 also cited examples from individual enterprise customers, including a $6 million reduction in detention and demurrage exposure, $200,000 of working capital identified “with a single question to Mo,” and nearly $750,000 in savings tied to a single disruption-management initiative. Those are company-described results at unnamed or individually cited customers, not independently audited averages. The company said 210 customers now use Mo, which has answered more than 50,000 questions to date, and that Mo is available now for project44 customers. It also reported an internal test comparing Mo with a general-purpose large language model on the same underlying data; those scores are noted under Additional sources.

What it means for manufacturers

The instructive part of this announcement is not the trillion-record figure. It is the reminder that “on time” is a definition, not a universal metric, and that an assistant is only as useful as the query it runs against a company’s own records. Midsize manufacturers that ask a chatbot about late inbound loads often get a fluent answer that does not match how the plant actually measures a milestone. That definition problem is a recurring theme in supply chain AI work, and it also sits behind freight tools such as Freightos’s forwarder platform.

A practical next step is to write down, for a handful of recent inbound and outbound shipments, the milestone the plant uses as “on time,” the source of the actual date, and whether detention, demurrage, or accessorial charges are visible without a manual chase. Our note on October dry van spot rates is a starting point for the cost side of the same visibility question.

Questions about shipment milestones, detention charges, or whether logistics data is consistent enough to support an assistant?

Contact us

Sources

Additional sources

  • In the same November 10 release, project44 said that in internal testing, with both systems working from the same underlying data, Mo answered 10 of 10 verified test questions correctly while a general-purpose large language model answered 2 of 10 correctly; that Mo gave the same answer all 10 times while the general-purpose model did so 2 of 10 times; and that Mo responded 2.5 times faster. Those scores are company-reported results from an internal test, not an independent benchmark.
  • Forwarder News, “project44’s Mo, InstaLILY’s survey and the AI walls around freight pricing data” (Nov. 11, 2026), a contemporaneous trade item that discusses Mo in a broader freight-data context. This report relies on the project44 release for product facts and customer figures.

Photo: Frank Schulenburg, Wikimedia Commons, CC BY-SA 4.0 (illustrative)

Try the Statadventure live tools

Inventory tools for US manufacturers: reorder point, safety stock, EOQ and landed cost, with the inputs and assumptions shown.

Open the live tools

Contact us

Scroll to Top