Moving an AI feature out of the modal, onto the page

My Contribution

1. 01Research

   Identified a hidden user behavior (later named Double Review) by mapping customer interviews into a Journey Map and validating the pattern in follow-up conversations

2. 02Alignment

   Led a cross-functional workshop with 10+ teammates across engineering, design, and product to align the team on the core pain points; the PM used that alignment to adjust the Scrum schedule

3. 03Proposal

   Designed and pitched a new audit experience that runs alongside the existing flow, grounded in three observations about how users actually work

4. 04Prototype

   Designed the end-to-end prototype by Claude Code, covering every step of the new audit flow

5. 05Migration

   Designed follow-up extensions that helped existing customers adopt the new flow, by carrying capabilities they relied on in the old one

The original design: review inside a Modal

GoFreight's AI Document Intelligence feature had already shipped before I joined the team. After a user uploaded a PDF, a popover Modal would open: PDF preview on the left, AI-extracted fields on the right. The user confirmed each field through a dropdown, corrected anything that looked wrong, and pressed Save. The data was then written to a Shipment (creating a new one, or updating one the user had selected).

The flow had been running for a while. The accuracy of the AI extraction itself wasn't the focus of this project. What needed reassessing was where the review action sat: entirely inside a Modal.

The pattern users didn't tell us about: Double Review

2a. Mapping feedback during PM interviews

The PM was running a series of customer interviews at the time. As the interviews unfolded, I tracked each user's feedback in a Journey Map: their actions, pauses, and any verification steps that surfaced in the audit stage.

One pattern stood out. After pressing Save, users would reopen the same Shipment Entry and check that the data they had just saved was really there. To verify it wasn't an outlier, we started asking about it directly in the interviews that followed. Most users confirmed it was a routine part of their workflow.

The team came to call this behavior Double Review.

Double Review: after the review inside the Modal, users still went back to the Shipment Entry to review whether the data had been saved correctly.

2b. Validating the pattern with engineering

To check that Double Review wasn't isolated to a handful of customers, I led a cross-functional workshop with 10+ engineers from the product team. The session had two purposes: share the customer observations I'd gathered so far, and walk the group through an impact / feasibility prioritization exercise.

By the end of the session, the team had aligned on accuracy and Double Review as the core pain points for this iteration. The PM used that alignment to adjust the Scrum schedule for the upcoming sprints.

Why move audit into the Entry page

Working through the directions together in the workshop, we aligned on a direction: build a new audit experience on the Shipment Entry page itself. Three observations from the research reinforced this:

Decision: ship v2 alongside v1

These three reasons set the direction for v2, but we decided not to replace the Modal outright. v1 would stay live for customers already using it, and v2 would ship alongside it on the same product. The plan was to run them in parallel with real workflows, to see whether the new flow actually helped both new and existing customers, before deciding the long-term direction.

The new design: review on the Entry page

Past the prototype: two extensions in development

After the initial prototype, the design kept moving. Two extensions came out of further use-case review.

Extensions added during development

Two features were added to the Review Guide panel after the initial prototype. Both carried capabilities from the old modal into the new panel, so existing users could move over without losing the moves their workflow depended on.

(1) Copy From an existing Shipment

When creating a new Shipment from a PDF, users sometimes want to start from the data of an existing Shipment. Copy From lets them select any past Shipment by file number; the new Shipment opens pre-filled with that data, and the AI-extracted values land on top of those fields. This came out of cases where users had recurring Shipments they wanted to clone, and starting from a blank form meant retyping known information.

(2) Manual select Update Shipment

Update Shipment originally relied on the system to detect which Shipment the uploaded PDF belonged to. The detection worked in most cases. There were also cases where users needed to update a Shipment the system hadn't picked as the match. The panel now lets users select the target Shipment manually.

Both extensions reflect a broader product call: the new panel only succeeds if existing users actually move into it. Keeping the moves they rely on lowers the cost of switching and keeps the migration on track.

What I take from this round

There were a lot of design tradeoffs in this project, and the case study above is where they live. Here I want to land on one principle: when the calls got hard, I kept following the happy path. Stay with me.

In any project, timeline and technical constraints show up at some point. What I can protect, as a designer, is the base experience: the path most users actually take. Stretching the design to absorb every edge case has a cost. It complicates the main flow and pushes complexity onto every user, even for cases they rarely run into.

The way I think about edge cases shifted on this project. An edge case is a problem to fix. Designing the interface around it locks the problem into the UI. Source Documents is one example. The PDF panel could have been pinned open by default, since it's the source of truth. Most users only reach for it when something looks off. Pinning it open would have served a rare moment at the expense of the main flow. Keeping it collapsed gave that space back. That shift is one of the most useful things this project taught me.