We just recently hosted our first customer conference and the one thing we heard time and time again came down to, "how do we get our people to adopt this change?"
"The limiting factor for AI in enterprise is not technology"
Matt McKinney, CEO of Loop, argues enterprise AI is blocked by culture, and that manufacturing will adapt better than services.
"The limiting factor for AI in enterprise is not technology. It's change management."
"the exception is the rule"
Matt McKinney, CEO of Loop, says supply-chain work is dominated by exceptions, which is why his company built an exception agent.
"There were literally 20, 30 people a week processing these types of exceptions"
"By resolving exceptions in minutes
"Labor will be commoditized"
Matt McKinney, CEO of Loop, argues AI will wipe out routine labor, but reward people who keep climbing into higher-value work.
"Labor will be commoditized as we know it. Talent will never be."
"They're outrunning the tokens in the system."
Loop CEO Matt McKinney highlights the number one pain point in supply chain: untrusted data. When 60% of critical information is offline, how can AI truly deliver? This is where the biggest opportunity lies.
The result: a feedback loop where every edge case we encounter makes the next one easier to handle. New carrier formats, new invoice layouts, new extraction quirks; they surface, get diagnosed, and get addressed. Without someone manually noticing each one.
Our engineer Henry Knoll walks through the full system, including how the failure grouping mechanism works and what comes next. loop.com/engineering-bl…#supplychain#AI
Our solution: entity context rules. Natural language instructions tied to a specific entity and task type, injected into the LLM prompt only when relevant. "For invoices from [issuer], the year is often wrong. Infer the correct one from context."
When automation gets something wrong or leaves a task unanswered, we persist that as a failure instance. Over time, similar failures cluster. An agent evaluates each cluster for a shared root cause. If it finds one, it proposes a fix: a new rule, a prompt update, a system
But reactive rule-writing only works if someone notices the problem first. We needed a way to surface gaps automatically, before a human catches the same mistake for the hundredth time. That's what failure analysis does.
The core problem: LLMs are good general reasoners. But logistics data is full of entity-specific, unintuitive knowledge that isn't in any training set. It lives in the heads of people who've spent years working with specific carriers and invoice formats.
LLMs don't know that one carrier's "Document Charge" is actually a cross-border processing fee. Or that a specific invoice issuer always writes the wrong year on the date field. Experienced auditors know this. Our models didn't... at. first. Here's how we closed that gap: 🧵
The companies already running on it:
A Fortune 100 food company surfaced $9M in previously invisible inbound costs. 95% touchless invoice automation in 30 days.
A fast growing jewelry brand’s Finance, Transportation, and Ops teams operate from one source of truth.
We raised $95M last month to expand the platform across the full supply chain — POs, warehouse, customs, ERP, TMS.
The Logistics Data Platform is where that starts.
Learn more: