Rebuilding AI Carrier Onboarding One Step at a Time

Our old carrier onboarding process took about sixty days per carrier. You'd go to one carrier at a time, gather example files, rework the mapping until it met our requirements, then finally begin auditing invoices. Fixing that became the clearest example I have of how AI carrier onboarding should get built: one problem at a time, not one sweeping platform.

I talked through this on a recent Trax Live conversation, and the methodology is worth expanding on, because it matters more than the tool we ended up with.

The Process Nobody Fully Understood

Before we brought in any AI, we sat down with operations, every single person in the role, and went through what they were doing step by step. I thought I knew that process well. I'd been in this industry long enough. But when you sit down with the person doing the work and find out what they're doing day to day, you realize no one knows the whole thing, top to bottom. Everyone's holding a piece of it, and nobody's ever had to hold all of it at once.

That's the sixty days, right there. Nobody had ever seen the whole map, just their corner of it.

Why We Didn't Wait for One Big AI Platform

Once we had the process mapped out, step by step, we picked one piece of it and used AI to automate that piece alone. Then the next piece. Then the next, building each one out before moving forward. We built the resulting tool in about six weeks working this way, taking one small thing at a time instead of trying to solve onboarding all at once.

What AI Carrier Onboarding Looked Like in Practice

We already had AI Extractor handling data entry replacement, using AI computer vision to ingest paper invoices. That gave us clean, structured data to build on.

What we added takes every carrier invoice from a new client, across multiple carriers and multiple submission types, and analyzes them in bulk rather than one or ten at a time. It tells you where a carrier's data needs to change, what to configure so that carrier loads correctly, which EDI specification they're using, and how to map them, all at once instead of carrier by carrier.

What This Means If You're Evaluating AI for Your Own Operations

Traditional product management isn't cutting it anymore. You have to incorporate AI, and that means rethinking how you build, not just what you build.

The approach that worked for us was straightforward, if not easy: map the process fully before you automate any of it, and then take one piece at a time instead of waiting for a single tool to solve everything at once. That's what let us move from a sixty-day, carrier-by-carrier process to something built and working in six weeks.

Start With the Process You Know Best

Waiting for a complete AI solution before starting feels prudent, but it usually means waiting too long. Map the process you're trying to fix, pick one piece of it, and build from there.