project44 connects logistics analyst to Snowflake Cortex AI

project44 has integrated Snowflake Cortex AI into Mo, its existing supply chain analyst, to connect questions asked in ordinary language with structured shipment data. Announced on 10 November, the integration adds a data retrieval component to a product project44 launched in July, rather than introducing Mo as a new application.

Mo can be asked which freight lanes have missed a delivery target, which containers have incurred charges or whether a carrier’s performance has changed. Before calculating a response, the system needs to establish the relevant shipments, time interval and meaning of the business measure. Those definitions can vary even between customers using similar transportation management systems.

The new process divides the work between two components. Project44 supplies its logistics data model, network context and the customer’s semantic definitions for shipments, milestones, contracts and transport modes. Snowflake Cortex Analyst translates the resulting question into SQL, which retrieves numerical results from structured records held in Snowflake. The language model is therefore not expected to generate the underlying shipment totals from memory.

Structured queries provide a traceable calculation path, but a correct database query can still answer the wrong question if the chosen definition is inaccurate. For example, a late delivery rate may change when the measured event is arrival at a port rather than arrival at a customer’s warehouse. The integration depends on the business context being applied before the query runs and on accurate milestone records being available.

Project44 says it has accumulated approximately one trillion shipment records over ten years, alongside current milestone information and estimated arrival times. That scale describes the wider company data resources; it is not a statement that every Mo user can examine the full network without access controls. The actual retrieval depends on the customer’s permitted data and configuration.

A term such as on-time performance can refer to a vessel reaching its scheduled port, a container becoming available or a delivery meeting a booked receiving slot. Contractual grace periods and approved exceptions can also change the count. An analysis that selects the wrong event will produce an irrelevant result even when its arithmetic is sound.

Mo applies logistics semantics and customer-specific business rules to the question before the query is issued. Those definitions distinguish, for example, whether a late arrival should be measured from cargo-ready date or from confirmed collection, and which movements should be excluded under a service agreement. The company says that codifying such rules helps it deliver consistent answers to repeated questions.

Cortex Analyst then generates the structured query needed to retrieve relevant figures from Snowflake. This makes the boundary between language interpretation and data access clearer, although changes to source feeds, definitions or customer data will still affect the answer. The company has not published an independent comparison isolating the accuracy contribution of the new Cortex integration.

Project44 describes its Mo product as an AI agent harness containing guardrails, workflows and execution capabilities around a language model. The company says the harness manages the logistics context used in decision support while Cortex AI supplies retrieval capability. Actions affecting customer operations must still be governed by the permissions, validation steps and workflows established for the system.

Before the Snowflake collaboration, project44 had reported 210 customers using Mo and more than 50,000 questions answered. Both numbers appeared in its 3 November update, a week before the new announcement. They establish the reach of the existing product rather than adoption attributable to the latest integration.

Project44 also cites internal testing in which Mo answered 10 of 10 selected questions correctly while a general-purpose language model answered two, with Mo returning the same answer in all ten repetitions and responding 2.5 times faster. The company conducted these tests, and the 10-question exercise was not an independent, representative benchmark. Equivalent results should not be assumed for every customer’s data or every question.

The same comparison results appeared on 3 November, before the Snowflake integration was announced. It would be misleading to present the test as proof that adding Cortex Analyst caused the difference. The new source describes how the systems are combined; it does not report a controlled before-and-after test measuring the effect of that change alone.

Customer examples cited by project44 include a $6m reduction in detention and demurrage exposure, $200,000 in released working capital and almost $750,000 in savings linked to one disruption initiative. The company attributes these to the broader Mo system and its operational data, not specifically to the newly integrated Snowflake component. The individual examples are not accompanied by externally validated measurement methods.

Detention and demurrage analysis requires more than a count of containers. Operators need reliable arrival, availability and equipment return milestones, contractual free-time limits, and the applicable charging schedules. Identifying avoidable costs in those records may inform action, but the financial outcome depends on changes to the physical movement or handling of goods.

Jett McCandless, founder and chief executive of project44, identified the separation between the company’s logistics definitions and Snowflake’s data retrieval as the basis of the integration. Snowflake’s Will Allen described using operational questions to prioritise responses to disruption and potential transport savings. Neither organisation has provided independent customer measurements attributable solely to the new connection.

The collaboration is also distinct from project44’s work on Georgia-Pacific’s factory yard operations, which IN Supply covered on 3 November. That deployment concerned vehicle check-in, trailer movements and records across manufacturing sites. Mo’s Snowflake integration instead concerns how existing supply chain records are interpreted and queried, so it represents a separate product development.

Mo remains available within project44’s Decision Intelligence Platform. Its use will depend on consistent business definitions, complete movement records and controls for any operational action taken in response to an answer. The new integration establishes a more explicit route from a natural-language question to a database result; specific performance gains from the Snowflake component have yet to be established independently.