In a recent post, we introduced the capability of asking questions about your trading partner activity directly in the AI assistant you already use. Getting details on orders, shipments, and invoices without a portal or additional login required. Quick Read here.
Now we’re getting into what happens after you ask one of those questions. Because getting data back isn’t the win. Getting a real answer is.
Matching is Harder Than it Looks
Ask most people what it means for two EDI documents to match and they’ll describe something simple: the numbers line up, or they don’t. Anyone who has actually compared a purchase order to an invoice knows that’s not quite true.
When you scan two documents side by side, you’re not running a simple check. You’re doing something closer to pattern recognition, built from experience. You know which fields matter, which small differences are normal, and which ones signal a real problem. For a system to do this well, it has to check exact fields, apply tolerance thresholds, and understand expected sequences, not just compare two numbers and call it done.
That’s why “mismatch” isn’t really one problem. It’s a category of different potential problems that all look the same on the surface: two documents that don’t agree.
The Anatomy of a Mismatch
A few of the most common patterns behind that surface-level disagreement:
Timing discrepancies. A document gets generated before a status update lands on the other side. The invoice says one thing because the ASN hadn’t been updated yet, not because anything is actually wrong.
Unit-of-measure conversions. One document counts in cases, the other in eaches. The quantities look different because they’re measuring different units, not because either side made an error.
Partial shipments and partial invoicing. An order ships or bills in pieces. Comparing a single invoice to the full PO looks like a mismatch until you account for what’s already been fulfilled elsewhere.
Rounding and tolerance thresholds. Small numeric differences fall inside an acceptable range but still trip a naive line-by-line comparison that doesn’t know a threshold exists.
Data entry versus system-generated errors. Some mismatches are simple typos. Others come from the system itself, formatting or logic differences between two platforms that were never talking about the same thing in the same way.
You’ve likely diagnosed all of these before, probably without ever seeing them laid out as a list. That instinct, knowing which of these you’re looking at before you’ve finished reading the second document, is expertise. It’s just never been written down.
Asking Instead of Hunting
This is what’s actually happening when you ask your AI assistant, in plain language, “Why doesn’t this invoice match the PO for Retailer X?” instead of comparing both documents line by line yourself.
That capability, connecting your AI assistant directly to live ECGrid network data, is called MCP, an open standard now built into ECGrid. It’s what lets your AI assistant reach into your actual trading partner data and reason about it, instead of guessing or sending you back to a portal to look it up yourself.
The AI isn’t just pulling up two records and putting them side by side. It’s checking across these same categories, the same ones you’d check, and telling you which one applies. That’s the difference between looking something up and getting it analyzed. You ask a question. You get an answer, not two documents to reconcile yourself.
If you manage a team that does this all day, the same shift applies at scale. Less time hunting for the answer, more time already resolved.
What’s Next
Understanding why something doesn’t match is one thing. Filtering, drilling down, and resolving it interactively, right inside your chat, is next. That’s coming in our next post.
In the meantime, see what this conversational access means for you. Visit the ECGrid B2B MCP page