Thursday, November 12, 2026
The news that moves freight
Tech & AI Nov 12, 2026 · 7 min read

project44’s Mo, InstaLILY’s survey and the AI walls around freight pricing data

The AI that knows your rates: why data walls will decide who wins the tender

project44 announced on 10 November that Mo, its AI supply-chain analyst, now runs inside Snowflake. The company says Mo draws on “approximately 1 trillion shipment records”, has 210 customers and has answered more than 50,000 questions. On the same day, FreightWaves reported a survey by InstaLILY, a company that builds AI agents for distributors. It found that 1 in 6 managers and executives at AI-using firms said supplier or customer data had been exposed through an AI tool in the past year.

Both sets of figures come from vendors with something to sell. Read together, they still tell one clear story. AI analysts are being placed directly on top of the most sensitive data in freight: who pays what, on which lane, at what margin. The answers are getting quicker. The walls around the data are not keeping up.

Our view is simple. For forwarders, the hard question about AI is no longer “is it accurate?” It is “what can it see, and who else can see what it says?” The firms that can answer that in writing will win business. The ones that can’t will start losing tenders.

What project44 actually launched

Snowflake is a cloud data warehouse, where companies store and query their operational data. Mo now works alongside Snowflake’s own AI tools, called Cortex AI. In project44’s description, Mo interprets a plain-language question using its logistics “ontology” (a structured map of what freight terms mean and how they relate). It also applies each customer’s own business rules, transport modes, contract terms and exceptions. Snowflake’s Cortex Analyst then turns the question into SQL, the standard language for querying databases, and runs it against project44’s data in Snowflake.

The company says answers come from structured records rather than the model’s memory. That is a sound design choice. It is also why these tools need such wide data access.

The claimed customer results are large, and they are claims. project44 says one enterprise customer cut detention and demurrage (D&D, the charges for holding containers and equipment past free time) exposure by $6 million. It says individual customers have freed up $200K of working capital with a single question to Mo. It says one disruption management initiative saved nearly $750K. None of these customers is named.

The benchmark is also internal. project44 says Mo and a general-purpose large language model (LLM, the kind of AI behind public chatbots) worked from the same underlying data. Mo answered 10 of 10 verified test questions correctly; the LLM got 2 of 10. Mo gave the same answer all 10 times; the LLM managed that 2 of 10 times. Mo was 2.5 times faster. It is a 10-question test that project44 designed and ran. The company credits the results to Mo’s “harness”: the guardrails, workflows and execution layer around the model.

Founder and CEO Jett McCandless summed up the pitch: “Reliable data, reliable signal, reliable action.” Doug Cantriel, Head of North America Transportation at Ford Motor Company, is quoted in the release: “Supply chain decisions move fast, and the teams making them need answers that are just as fast.”

What the InstaLILY survey found

The survey was fielded through Pollfish, an online polling platform, and launched on 9 October 2026. It collected 250 responses from managers and executives at supply chain, logistics, distribution and wholesale firms that use AI in daily operations. FreightWaves reported the results on 10 November.

The headline figure is that 1 in 6 reported an incident in the past year where supplier or customer information was exposed through an AI tool. The access figures matter more. 78% said their AI tools can reach customer pricing and margins. 67% said the tools can reach supplier pricing and rebate information.

The most telling gap is about confidence. 87% said they were very or somewhat confident their AI tools keep supplier and customer data properly separated. Yet 63% acknowledged some overlap between that data across their platforms. People believe the walls hold while admitting there are holes in them.

The commercial signal is clear. 63% said a supplier or customer had asked how the company protects their information. For 30%, those questions influenced a decision about a software provider, vendor or AI tool. 90% said their employers would be more likely to choose an AI tool with stronger data controls, and 92% called data separation a high or moderate priority for the next year. The most cited obstacle was that existing AI tools were not designed for it (22%), followed by cost (21%) and lack of technical expertise (20%).

The caveats are real. InstaLILY develops AI agents for distribution, so it has a stake in the answer. The exposures were self-reported, not verified. Severity was not measured. The sample is distribution and wholesale, not freight forwarding.

Adam Isenberg, head of distribution at InstaLILY, gave the line that travels best: “AI working in distribution should know that one customer’s contract price never belongs on another customer’s quote. Buyers should make vendors prove that before they sign.”

The independent number behind the vendor numbers

A larger, older study points the same way. IBM’s 2025 Cost of a Data Breach report, conducted by the Ponemon Institute and covering breaches at 600 organisations between March 2024 and February 2025, found that 13% of organisations reported breaches of AI models or applications. Of those compromised, 97% said they had no AI access controls in place. Firms with high levels of “shadow AI” (staff using AI tools the company hasn’t approved) saw breach costs on average $670,000 higher.

That report is over a year old, and IBM sells security services too. But it describes the same weakness: AI tools given broad access and few limits.

Why this lands on forwarders

Swap “contract price” in Isenberg’s quote for the data a forwarder holds. Carrier contract rates. Named-account rates, where a carrier gives a specific shipper its own price through the forwarder. Each customer’s sell rate and margin. Many carrier and shipper contracts restrict who may see those numbers. A quoting agent that pulls from one shared rate database can, if built carelessly, surface one customer’s buy rate or margin in another customer’s quote. Or it can expose it to a staff member who should not see it.

That risk grows as AI quoting spreads. project44 itself launched an AI freight procurement agent in February that compares contracted rates with market rates, runs mini-bids and, in AJOT’s report of the launch, can “autonomously negotiate renewals” within limits the customer sets. Forwarders are building or buying similar tools for quotes, tracking and rate management. Every one of them needs broad access to pricing to be useful.

The second hit is commercial. Mo gives large shippers D&D exposure figures and lane answers straight from their own data warehouse. That is the reporting, D&D auditing and “control tower” work many forwarders sell as value-add. When the shipper’s own AI analyst answers those questions, the forwarder’s monthly KPI pack is worth less. What still sets a forwarder apart is execution, and proof that it can be trusted with data.

Third, the question is already being asked. In the InstaLILY sample, 63% had been asked how their data was protected, and for 30% it changed a buying decision. Shippers will put the same question to their forwarders, and forwarders will put it to their software vendors.

What forwarders should do now

Practical steps, before a customer asks:

  • Map what your AI tools can reach. List which rate tables, margins and customer files each quoting, tracking or email agent can query. If the honest answer is “everything”, that is the finding.
  • Separate by customer and by contract. Named-account and customer-specific rates should be walled off at the data level, not just hidden in the screen.
  • Test the obvious failure. FreightWaves’ report suggests checking whether a tool can prepare a customer quote without exposing supplier pricing. For a forwarder, ask the tool to quote customer B and see if customer A’s rate or margin leaks into the answer or the reasoning.
  • Set human sign-off points. Decide when a person must approve before an AI agent sends a quote or commits to a rate.
  • Ask vendors for consistency as well as speed. project44 is selling “same answer every time” as a feature. Ask your own vendors for repeat-run results on your data, not a demo.
  • Write it down. A one-page AI data policy you can attach to a tender response will be worth more than a feature list.

Our call

Our prediction: by the end of Q2 2027, major shippers’ annual forwarder tenders (RFQs, requests for quotation) will include a dedicated section on AI data handling. It will ask which customer and carrier data AI tools can access, how one customer’s rates are kept out of another’s quotes, and whether there is human sign-off before AI-generated quotes go out. Forwarders that can’t answer in writing will be marked down. We also expect visibility and freight-software vendors, project44 among them, to start publishing data-separation controls alongside their accuracy and speed benchmarks within the same window. That is our view, not a vendor’s.

Sources

Read us on Google? Add Forwarder News as a preferred source to see more of our stories in Top Stories.