22%
of drivers reported difficulty finding correct delivery points
Amazon - Geocode Correction
The research, design decisions, and outcomes in this case study reflect my work at Amazon. The case study itself was built by me using Claude as a coding assistant — a deliberate choice to present my work in a custom, coded format rather than a template.
Amazon · UX Case Study · 2025
Geocode Correction
Geocode Correction
Workflow
Redesigning how Customer Service Associates submit delivery point corrections inside Amazon's contact center — reducing handling time by 45% and improving delivery success for millions of last-mile packages.
My Role
L6 UX Designer
Platform
AC3 LMS (Web)
Timeline
Q1–Q3 2025
Launched
August 2025
Overview
A unified tool for a fragmented problem
Amazon delivery drivers navigate 21.4 million support contacts per year. A significant portion stem from a single, solvable issue: inaccurate map data. When a geocode is wrong, drivers can't find the delivery point, customers don't receive packages, and Customer Service Associates (CSAs) spend nearly 9 minutes manually correcting data across legacy tools.
As part of Amazon's SDS 3.0 redesign, I led UX for the geocode correction workflow — a new native experience inside AC3 LMS that replaced a fragmented, error-prone process with a guided, intelligent workflow.
21.4M
Annual driver support contacts
15%
Contacts from geocode or geofence issues
8.2 min
Avg handle time per geocode call
38%
YoY increase in location-accuracy contacts
The Problem
Broken maps, broken workflows
Amazon Flex drivers and Delivery Associates operate in the real world — where new developments open, access points change, and map data lags reality. When a geocode is wrong, the driver calls support. But the CSA's experience wasn't much better than the driver's.
"CSAs were toggling between the legacy Driver Contact Landing Page, external mapping tools, and SIM ticketing — all to complete a single correction. The process was error-prone, slow, and untrained associates got it wrong constantly."
The downstream effects were measurable:
15%
reduction in delivery success in new residential developments
65%
first-attempt delivery success rate in affected geocode areas
12%
increase in customer delivery complaints
My Role
End-to-end design across a complex, multi-stakeholder system
I was the UX designer for the geocode correction workflow (Epic 6.0) within Milestone 3 of the SDS 3.0 roadmap — working embedded with product, process engineering, program management, and engineering.
My work spanned the full design lifecycle: from shadowing CSA sessions and translating complex SOPs into intuitive workflows, to delivering high-fidelity mocks and facilitating reviews with SDS leadership.
Discovery
Stakeholder interviews
Mapped current-state workflows with process engineering. Audited legacy DCLP and identified failure points.
Research
CSA shadowing
Observed 45 CSA interviews and 30 usability sessions. Identified where CSAs deviated from SOP and why.
Design
Workflow mapping
Translated dense SOP requirements into guided, branching AC3 LMS workflows with interactive prototypes.
Validation
UAT + pilot
Facilitated UAT with 100+ CSAs across global marketplaces before August 2025 worldwide launch.
Who I Designed For
Two users, one workflow
The geocode correction workflow serves two distinct users with different mental models, technical proficiency, and time pressure. Getting the balance right was critical.
👩💼
Expert CSA
Sarah Martinez
5+ years. High technical proficiency. Trains new CSAs and handles complex cases.
Efficient workflow, minimal friction
Advanced override options when needed
Quality assurance visibility
👨💼
New CSA
James Chen
3 months. Learning systems. Follows strict protocols and seeks guidance at each step.
Clear, guided step-by-step flow
Error prevention before submission
Confidence the correction was right
🚚
Delivery Driver
Michael Thompson
2 years, urban route specialist. Time-pressured and frustrated when the map is wrong.
Get back to deliveries fast
Trust the fix will stick
Not held on the phone longer than needed
Key Design Decisions
Solving the right problems
The geocode correction workflow required balancing SOP compliance, CSA efficiency, and driver experience. Four decisions defined the design.
Decision 01
Move only the pins you need to move
The legacy DCLP required CSAs to reposition both the road entry pin and the delivery point pin — even when only one was incorrect. This caused erroneous corrections and confused CSAs about what they were actually changing.
I designed the new workflow so AC3 does not require moving a pin unless the CSA actually needs to. Each pin is independently optional with its own contextual question set.
Before
CSA forced to move both pins. Frequent accidental corrections. High error rate.
After
Pins are independent. Move only what's wrong. Questions only appear for moved pins.
Decision 02
Surface risk before submission, not after
Corrections under 25 meters are statistically less likely to reflect a genuine map error — they often indicate a driver preference. Submitting these without scrutiny added noise to the LMAQ review queue.
I introduced a warning gate for sub-25m moves that prompts the CSA to select a reason. This flows directly to LMAQ, reducing false corrections and giving the team structured context.
Before
All corrections submitted identically. No context for small moves. Queue bloated with low-signal submissions.
After
Sub-25m warning + reason capture. LMAQ gets structured context. Higher signal-to-noise ratio.
Decision 03
Let the driver go — then finish the work
CSAs kept drivers on the phone while submitting corrections. Submissions could take several minutes, inflating AHT and frustrating drivers with remaining deliveries.
I designed the workflow to guide CSAs to disconnect the driver before submitting. After-call work is deferred to the end. The driver gets back on the road faster; the CSA submits without time pressure.
Before
Driver stays on call during submission. Avg 8.2 min handle time. Both parties under pressure.
After
Driver released early. CSA submits post-call. ACW deferred. AHT drops 45%.
Decision 04
Don't let CSAs submit duplicate corrections
When an address already has a pending geocode correction in the LMAQ queue, submitting another created duplicate tickets that slowed review and could result in conflicting data.
AC3 now checks for pending corrections before the CSA begins. If one exists, the workflow surfaces it, blocks a new submission, and ends gracefully.
Before
No duplicate check. Repeat submissions common. Conflicting corrections created bad data.
After
Pending correction surfaced upfront. Duplicates blocked. Workflow ends cleanly.
Impact
Measurable results at scale
The geocode correction workflow launched globally in August 2025 as part of the Milestones 1–3 release. Results measured against legacy DCLP baseline.
45%
Reduction in average handle time
70%
Decrease in repeat contacts for location accuracy
30%
Improvement in delivery success in corrected areas
95%
CSA satisfaction with the new workflow
$12M
Annual cost savings
50%
Reduction in driver escalations
35%
Decrease in delivery complaints
25%
Fewer delivery attempts in corrected areas
Reflections
What I learned
This project stretched across multiple milestones, global marketplaces, and a dense web of stakeholders. A few things I'd carry forward:
What worked
Embedding deeply with process engineering early meant designs were grounded in SOP reality — not idealized flows that fell apart during UAT.
The driver-release-before-submission pattern reduced AHT significantly. Simple sequencing decisions can have outsized metric impact.
Bundling features into coherent milestone releases avoided CSA confusion about which tool to use for which case.
Challenges navigated
Integrating with legacy systems (DCLP, LMAQ, CSALT) created constraints requiring creative design compromises and close engineering collaboration.
Marketplace-specific variations (US, UK, IN, JP, etc.) added significant complexity to workflow branching that had to be legible in every market.
Change management for experienced CSAs with strong legacy DCLP habits required careful onboarding design and training alignment.
What's Next
Laying the foundation for intelligent correction
The geocode correction workflow in its current form is guided and SOP-compliant. The roadmap points toward a smarter system: ML-powered suggestion engines that surface likely corrections before the CSA has to search, predictive geocode recommendations based on historical delivery patterns, and real-time validation against live map data.
The data instrumentation built into this milestone — structured reasons, pin coordinates, geofence context — is the foundation that makes those future capabilities possible.
Amazon · Shipping and Delivery Support · AC3 LMS · 2025