You’ve got a new trading partner or a new transaction type. The 850s are parsing, but now you need to map those segments and elements into your warehouse, your ERP, or your reporting schema. That used to mean opening the companion guide, the X12 spec, and a spreadsheet—then spending hours (or days) drawing lines from BEG02 to your order number column and hoping you didn’t miss a conditional segment. We wanted that first cut of the mapping to take seconds, not days. So we built an AI that suggests EDI-to-target field mappings in about 30 seconds. Here’s how it works and why we still keep a human in the loop.
The problem: mapping is the slow part
Parsing EDI is one thing. You get a parse tree—segments, elements, loops. Turning that into a usable schema for your systems is another. You have to decide which EDI path maps to which target column, handle repeating groups, deal with qualifiers and code lists, and respect the fact that one partner’s 850 might put the PO number here and another’s might put it there. Doing that by hand is tedious and error-prone. Doing it with a generic “map anything to anything” tool often means you still have to configure every relationship yourself. We wanted the first pass—a draft mapping from a real EDI instance (or a sample) to a target schema—to pop out in under a minute so you could review and adjust instead of starting from zero.
What we actually built
We run inside your Snowflake account. No EDI data leaves your environment. The flow is: you have parsed EDI (we can parse it for you, or you’ve already got structured output) and a target schema—the tables or views you want to load data into. A dedicated AI step takes that EDI structure and your target schema and produces a set of suggested field mappings. You see them in the app—source path, target column, and a short rationale for each—so you can review and adjust. The whole thing, from “run it” to “suggestions in the app,” is on the order of 30 seconds for typical 850/856/810-sized documents and target schemas. The AI runs inside Snowflake using the platform's built-in capabilities, so your EDI and your schema never leave your account.
Why 30 seconds and not “instant”
The 30 seconds is mostly the AI doing its work plus a bit of prep and turning the result into something you can use in the app. We didn’t try to squeeze it to “instant” because we don’t need to—getting a first draft in half a minute is already a huge step up from manual mapping, and it leaves room for the model to do a bit of reasoning (e.g. “matching segment semantics to column names”). So we optimized for “useful first draft fast” rather than “sub-second.”
Why we still put a human in the loop
We don’t auto-apply the suggestions. They land in the app so you can review, edit, and approve. EDI is full of partner- and guide-specific rules: conditional segments, code list meanings, and “this element is the PO number for Partner A but the reference number for Partner B.” The AI is good at the first pass—same segment and element names, similar semantics—but it doesn’t know your business rules or your partner agreements. So the AI gives you a 30-second head start; the human makes it correct. We’ve found that’s the right split: most of the obvious mappings are there, and you spend your time on the edge cases and the “except for this partner” logic instead of on the whole grid.
What’s in the box
Under the hood we use Snowflake’s built-in AI so everything stays in your account. The app has a step that takes the EDI parse output and your target schema (or a reference to it) and produces mapping suggestions. You can trigger it from the UI or via a direct call. The suggestions include source path (e.g. segment and element), target column, and a short explanation so you can see why the model suggested that link. From there you edit, add, or drop mappings and use them in your pipelines. No data leaves your Snowflake account.
We built an AI that maps EDI schemas in about 30 seconds by turning the problem into a single, focused step: EDI structure and target schema in, mapping suggestions out. It’s not magic—it’s a narrow, well-defined task that fits what modern AI is good at, with a human in the loop so your partner-specific and business rules stay correct. If you’re running EDI in Snowflake and want to try it, EDI Sentinel includes this mapping-suggestion flow along with parsing and drift detection.