Simple Record Count Amazon
About me Playground · Coming soon
Simple Record Count · 2024

Adding product images during a system migration to cut item identification time by 3 seconds and save $9M/yr

Lead UX Designer · Sep 2023–Feb 2024 · ICQA associates worldwide

End-to-end UX design · Research and concept testing · Interaction design · Handoff · Scope advocacy
V2 legacy interface vs V3 TaskUI side-by-side
The problem

Text-only identification was slow, error-prone, and leaving screen space empty

SRC is the targeted count tool ICQA associates use to verify item quantities in fulfillment center bins, 100-200 counts per shift. V2 gave them a text description and nothing else. When a bin holds multiple similar items, reading a description and matching it to a physical product takes 10-15 seconds. The cost compounds fast.

10–15s
to identify an item using V2's text-only interface
100–200
counts per associate per shift, every day
99.9%+
inventory accuracy target that text-only made harder to hit
Constraints

Designing for a new platform and the warehouse floor

SRC runs one-handed under bright warehouse lighting by associates wearing gloves. The migration moved the tool from the older Zebra MC2200 to the taller MC3300. A bigger screen V2 was never built to use.

01
Hardware
Zebra MC3300 series scanner, portrait only, one-handed operation, with touch targets large enough to hit reliably with gloves on.
02
New platform capabilities
The migration was an opportunity to understand and use the new platform's features properly, not just replicate what existed before.
03
A taller screen to fill
The MC3300 display is meaningfully taller than the MC2200 (425px vs 293px). That extra real estate is exactly where the product image went.
Hand scanner comparison: Zebra MC2200 vs MC3300 screen size
Research

An iPhone, a scanner handle, and a paper keypad

I didn't wait for engineering. I built a clickthrough Figma prototype on an iPhone attached to a real scanner handle, with paper standing in as the keypad: actual warehouse hardware, actual warehouse environment.

01
Visual confirmation was the missing piece
Associates consistently said a photo would let them confirm the item at a glance instead of reading and matching text.
02
Screen space was being wasted
V2 left significant real estate empty. Associates immediately noticed the image in the prototype and responded positively to seeing it fill that space.
03
Hardware shaped the interaction model
One-handed use with gloves meant every interactive element needed to be reachable with a thumb and large enough to hit reliably.
04
Similar items were the real failure mode
The worst misidentification cases came from items with nearly identical text descriptions. A photo made those instantly distinguishable.
iPhone attached to a hand scanner handle with a paper keypad, held up to a warehouse bin
iPhone attached to a hand scanner handle with a paper attached as a keypad.
Concept testing pod with products laid out on a table
Concept testing pod setup in a training room.
The solution

One image, placed right, changed everything

One image, placed at the top. Large enough to read at arm's length, positioned before the item details so it's the first thing an associate sees after scanning.

01
Product image as primary identifier
Prominent placement, large enough for arm's-length scanning. Falls back to a generic icon if the catalog image isn't available.
02
Clear information hierarchy
Image → item name → ASIN/SKU → quantity. Each layer gives associates a different confirmation point.
03
Optimized layout
Previously unused screen space filled purposefully. No added clutter, no new workflow steps.
04
High contrast design
Text and interface elements designed to read clearly under warehouse fluorescent lighting and on a small handheld screen.
Before · V2
V2 text-only item identification screen

Text description only. Associates read, matched, and identified with no visual reference: 10-15 seconds per item.

After · V3
V3 screen with product image as primary identifier

Product image as the primary identifier. Associates confirm at a glance before counting. 3 seconds saved per count.

Pushed for it

Product images weren't in the original migration scope

The brief was a tool migration: move SRC to the new platform, maintain existing functionality. Product images weren't planned. I built the case from research: the scrappy prototype had already shown associates responding immediately to the image. I brought that evidence to PM and engineering and got it added as a priority feature.

Outcome

A feature that wasn't in the brief became the most impactful part of the migration, and the model for similar improvements across other ICQA tools.

Validation

Tested in the warehouse, not the conference room

Validation happened in the environment the tool would actually be used in. The scrappy prototype gave us early signal, and we continued testing through iterations before shipping. Most of what we refined was about image placement and loading states, getting the experience right for slow WiFi and busy associates.

01
Discovery & research
On-site observations
02
Scrappy prototype
iPhone + scanner handle
03
Design & iteration
Warehouse testing
04
Migration launch
Worldwide rollout
Impact

Three seconds. Multiplied across the network.

~1250 hrs
estimated savings across the organization every day
~456K hrs
estimated savings annually across Amazon's fulfillment network
99.9%+
inventory accuracy target, now easier to hit with visual confirmation
Reflection

Small changes have enormous impact at scale

01
A photo turned out to matter enormously
The associate enthusiasm for such a "simple" change was surprising. Removing the cognitive work of text-based identification on a fast-paced warehouse floor adds up fast.
02
Establish baselines earlier
Stronger upfront metrics would have told the story more precisely. The impact was real, but the numbers had to be reconstructed after the fact rather than measured directly.
03
Test with newer associates sooner
Day-one associates have a different interaction pattern than tenured ones. Getting them into testing earlier would have surfaced edge cases we only discovered later.