by

Slashed tech debt by reengineering a data modeling workflow

Several product teams were building complex, interrelated data models to power both the product UI and a new generation of AI agents at a Fortune 500 company. The agents depended on that data model content — names and descriptions for hundreds of fields — to understand customer data and act on it. But the existing workflow gave data model content only a brief review window. That wasn’t enough time for the iterative, cross-functional work that clear, AI-ready content requires. I redesigned the review workflow so we could vet the content without putting the release date at risk. The result was a 15% ongoing cut in technical debt and rework, freeing the team to pull more from the backlog.

UX writing & content strategy
Workflow & standards design

Results

15%

Ongoing tech debt reduction

The situation

Data models were getting larger and more complex 

The company’s data models were growing in size and complexity as teams added agents to their roadmaps. 

The challenge

The current workflow was a bottleneck 

I had been the technical writer for many product teams, and data model content review was typically treated as a superficial cleanup pass. Now, with AI agents drawing context directly from larger models, the stakes of getting field names and descriptions right had quietly gone up — but the workflow hadn’t caught up.

The release deadline couldn’t slip, but the planned 3-day review window wasn’t enough time to assess a large data model, collaborate on review, and finalize the data model content. 

I needed to work with engineering to redesign the content review workflow so there would be time for iterative review and collaboration.

“Cate impressed me not only with what she produced but the process by which she produced it. Never did I feel more secure that the needs of my team were being met than when I was working with Cate.”

Lead Member of Technical Staff, Fortune 500 Company

My approach

1. I asked for earlier, staggered delivery

Instead of waiting for a single handoff near the deadline, I asked engineering to flag objects as soon as they were modeled — weeks earlier. That way, I could deliver final content in parallel with development, across sprints, instead of in a single crunch at the end.

2. I built a simple visual system to track progress

To drive content review and make the process more transparent, I inserted color-coded columns into the model template. The colors showed progress toward final content at a glance. I left the original columns alone, as a record of engineering’s starting names. 

3. I rolled out the new workflow and iterated

I walked the team through the redesigned workflow, addressed questions, and implemented a few requested tweaks to the template.

Planned fields in a data model

Two rows in a spreadsheet showing planned fields in a data model, with column headers indicating draft or approval. In first row, "Approved" cells have green background: API name draft is Id. API name approved is UserId. Description draft is Id for the user. Description approved is The user's login ID. Must be formatted as an email address. In second row, "Approved" cells have yellow background: API name draft is Type. API name approved is UserRole. Description draft is The type of user. Description approved is The permission level assigned to the user: owner, admin, member, or viewer.

API names and descriptions for two fields on an object. Yellow indicates review is still in progress.

More ways to improve the process

I always look for opportunities to use technologies to automate and streamline workflows. 

In the context of data modeling workflow in particular, I would look at using AI to help gather context from code and to track and prioritize field reviews across data models.

To judge what made the content clear, I’d keep a human in the loop.