UX Case Study · Oracle

Oracle Session Delivery
Management Cloud

Redesigning the Work Order Admin experience, turning a chaotic, 50-step manual firefight into a calm, intelligent, one-click resolution, and cutting execution time by up to 90%.

SS
Designed by
Sandip Sarkar
ProductOSDMC, Work Order Admin
ClientVerizon Network Ops
Duration12 Weeks
RoleSenior Designer
OSDMC · WORK ORDER ADMIN EXECUTION TIME -90% ROLLBACK SUCCESS 98% DEVICE ROUTE RESOLVED
90%
Reduction in task execution time
+52%
User satisfaction improvement
−82%
SLA breaches per quarter
98%
Automated rollback success
The Challenge

A 2 AM crisis with no escape hatch

“
Imagine it's 2 AM. A critical service update in the Verizon network has just failed. For millions of customers, this could mean a dropped connection, a failed transaction, or a missed emergency call. The network specialist on call has to act now.

But the tools they have are working against them. Instead of a single 'undo' button, they're faced with a manual, 50-step rollback procedure, a digital minefield where a single typo could make the problem ten times worse.

This wasn't a rare, once-a-year crisis. This was the daily reality.

My challenge was to step into this high-pressure environment and ask: How can we do better? The goal wasn't just a new interface. It was an intelligent experience that turns the firefight into a calm, controlled, one-click resolution.

Methodology

Design Thinking Process

A user-centric approach across five phases, grounded in real stakeholder observation and iterative validation with Verizon network operations teams.

Empathize01
Define02
Ideate03
Design04
Test05

12-Week Sprint

W1
Strategy & Research
W2
Interviews
W3
Empathy & Personas
W4
Problem & Goal
W5
Competitive Analysis
W6
User & Task Flows
W7
Info Architecture
W8
Moodboard & Lo-Fi
W9
Visual Design
W10
Prototyping
W11
Usability Testing
W12
Handoff
Empathize Phase

Understanding the people behind the network

Conducted interviews and job shadowing across 5 user roles, observed 20+ real-world scenarios, and collected quantitative data on task durations and failure rates.

I
Issac
IT Administrator
Goal
Get through the upgrade window without waking up at 2 AM to fix it.
Frustration
Keeps a printed copy of the 18-step MOP taped inside his desk drawer. Says he's memorized it but still checks.
D
Devika
NOC Specialist
Goal
Close her shift with zero open incidents, or at least nothing that follows her home.
Frustration
Has her own spreadsheet of error codes she trusts more than the official documentation.
R
Rick
Network Admin
Goal
Route around problems before Devika has to call him about them.
Frustration
Doesn't fully trust automated routing yet. Wants to see the reasoning, not just the recommendation.
S
Sarah
System Architect
Goal
Catch a failing device before it becomes a Sunday-night incident report.
Frustration
Her three monitoring tools disagree with each other about what "uptime" even means.
A
Adam
Admin
Goal
Approve a work order without becoming the bottleneck everyone blames later.
Frustration
Finds out about delays secondhand, usually from Devika, well after he's already signed off.
Empathize Phase · Research

User Interviews

Semi-structured interviews per role surfaced the real friction inside daily work order execution. A sample of the questions asked:

I

Issac

IT Administrator · Software Upgrades

  • QWalk me through the last upgrade that actually went wrong.
  • QHow do you know a device is ready to upgrade, versus just assuming it is?
  • QYou mentioned the MOP earlier, do you ever skip steps on routine ones?
  • QWhat's the first thing you check when a rollback fails?
  • QNot asking for a wishlist, but what's the one step you'd cut if you could?
D

Devika

NOC Specialist · Executes Work Orders

  • QWalk me through a night shift where three things went wrong at once.
  • QWhen something fails, do you trust the logs or your own gut first?
  • QHow do you decide when to escalate versus handle it yourself?
  • QYou mentioned a personal spreadsheet, can I see it?
  • QIs there anything you've stopped reporting because nobody acted on it anyway?
R

Rick

Network Admin · Route Manager

  • QOkay, so what actually takes eight hours in that process?
  • QIf a tool recommended a route, what would make you not trust it?
  • QHow much of this is written down anywhere versus just in your head?
  • QDo you and Devika ever disagree on a fix?
  • QSorry, back up, why does the spreadsheet route through email at all?
S

Sarah

System Architect · Health Monitoring

  • QWhich number do you actually check first thing in the morning?
  • QHave your three tools ever told you three different uptime figures for the same week?
  • QWho gets to decide which definition of "breach" is the real one?
  • QHonestly, do you trust a predictive alert as much as a confirmed failure?
  • QWhat would it take for you to stop double-checking the numbers yourself?
A

Adam

Admin · Approvals & Oversight

  • QWalk me through what you're actually approving when you sign off on a work order.
  • QOnce it's approved, how do you find out whether it went well?
  • QHas a delay ever changed a decision you'd already made?
  • QWhen something fails, how far back do you have to dig to find out why?
  • QEmail, text, or the platform itself, which one actually reaches you in time?
Framework

The Double Diamond

Divergent exploration followed by convergent focus, twice. From discovering the true problem to delivering a validated, tested solution.

Discover Define Develop Deliver Problem Solution
Discover
Observed 5+ user roles performing tasks across 20+ scenarios. Collected quantitative data on task durations and failures.
Define
Grouped the research into a short list of what was actually slowing people down.
Develop
Built future workflows reducing time by 80–90% using automation and an AI intelligence layer.
Deliver
Interactive prototypes tested with users showed improved clarity, speed, and confidence.
Quantitative Research

The Shape of Data

Beyond interviews, we captured numerical task-time data. The gap between current and future state made the case for redesign undeniable.

TaskCurrent Avg. TimeTarget Time (To-Be)Improvement
Work Order Software Upgrade30–120 mins/device5–10 mins/device↓ ~90%
Manual Rollback Procedure30–120 mins (50 steps)5–10 mins (automated)↓ ~90%
Route Manager Changes (CLI)8+ hours10–15 mins↓ ~97%
Manual Error Diagnosis120 mins<10 mins (AI-assisted)↓ ~92%

Task Time: Before vs After (minutes, log-adjusted view)

Upgrade Rollback Route Changes Diagnosis Before (As-Is) After (To-Be)
Empathize Phase · Synthesis

Empathy Maps

Mapping inner states, observed behaviours, and environmental cues surfaced the emotional texture of working inside OSDMC every day.

I
Issac
IT Administrator
Think & Feel
  • Last time it said success and it wasn't actually done.
  • I just don't trust the version number until I check it myself.
  • Frustrated by silent failures more than loud ones
  • Genuinely proud of his 98% rollback record, brings it up unprompted
See
  • A desk with two monitors, one permanently open to the MOP PDF
  • A sticky note with three device IDs that "always cause trouble"
Hear
  • Did you actually check, or did the tool check?
  • We had this exact failure back in March.
Say & Do
  • I'll believe it when the log confirms it.
  • Re-reads the MOP even on upgrades he's done a dozen times
  • Keeps a mental list of which devices need extra attention
D
Devika
NOC Specialist
Think & Feel
  • If I page Issac at 3 AM it better actually be worth it.
  • Runs the night shift alone most weeks, escalation is a last resort not a first move
  • More annoyed by noisy false alarms than by real failures
See
  • A personal spreadsheet of error codes, updated more often than the wiki
  • Four terminal tabs open, one of them usually forgotten
Hear
  • Same error as last Tuesday, isn't it?
  • Wake up Issac or wait it out?
Say & Do
  • Give me ten minutes before you call anyone.
  • Cross-checks the error against her own spreadsheet before trusting a dashboard
  • Documents the fix herself even when a tool already logged it
R
Rick
Network Administrator
Think & Feel
  • I can build the spreadsheet in ten minutes, it's the waiting on Devika that takes eight hours.
  • Doesn't mind the manual work itself, minds not knowing if it landed
  • More territorial about routing decisions than he'd admit out loud
See
  • A whiteboard sketch of the MPLS topology, redrawn every few months
  • Route history he keeps in his head better than in any tool
Hear
  • Just tell me which core it's hitting.
  • Why would I take the AI's word for it over 90 days of data I already have?
Say & Do
  • Show me the path, not just the score.
  • Recalculates a suggested route by hand before accepting it
  • Will accept a slower path if he understands why it's better
S
Sarah
System Architect
Think & Feel
  • I can't sign off on a number if I don't know which tool it came from.
  • Spends more time reconciling reports than reading them
  • Cares more about consistent definitions than about the dashboards looking impressive
See
  • Three browser tabs, each claiming a different uptime number for the same week
  • A shared doc where the compliance team argues about SLA definitions
Hear
  • Which tool are you quoting, the old one or the new one?
  • That is not what Infrastructure told me this morning.
Say & Do
  • Let's agree on one definition before we agree on one number.
  • Pushes back when a metric looks too clean
  • Keeps her own footnotes on what each SLA figure actually includes
A
Adam
Admin
Think & Feel
  • I signed off on this three days ago, why am I only hearing about it now?
  • Uneasy approving something he can't independently verify is on track
  • Wants to be looped in earlier, not just when something's already gone wrong
See
  • An inbox where approval requests and status updates arrive as separate emails, hours apart
  • Dashboards he checks only after Devika has already flagged a problem
Hear
  • You approved this, didn't you know it was going to fail?
  • It's in the system somewhere, I just have to find it.
Say & Do
  • Just tell me the one number I need to know before I sign this.
  • Calls Devika directly rather than waiting for a dashboard to update
  • Prefers platform notifications over email, but still gets most updates by email
Define Phase

Surfacing the core pain points

Most of it came down to two things: people waiting on manual steps, and nobody having a clear view of what was actually happening.

Time Inefficiency
Work order creation and upgrades take 30–120 mins per device, blocking real-time response.
Manual Rollback Risk
A single misstep in a 50-step CLI rollback can amplify the original failure tenfold.
Coordination Dependency
Every step depends on human hand-offs, with no shared visibility into status.
No Intelligence Layer
Zero predictive alerts or root-cause identification. Users rely on memory and logs.
Visibility Gaps
NOC specialists have no visual feedback on progress during execution.
No Proactive Alerting
Stakeholders learn about failures through verbal escalation, not the system.
Ideate Phase

Features that eliminate the pain

Six core capabilities mapped directly to user needs, anchored by an AI intelligence layer.

Core
Smart Work Order Templates
Pre-defined workflows based on past device behaviour, no blank-slate setup.
Core
Status Tracking
Clear visual transitions: Ready → Running → Completed, replacing CLI polling.
Core
Auto-Rollback
Triggered on failure with minimal delay. 50 manual steps become one click.
AI-Powered
AI Log Analyser
Surfaces probable root cause from logs with fix suggestions: 120 down to under 10 mins.
Core
Notification Engine
Automated alerts to stakeholders at key events. No more verbal coordination.
AI-Powered
Recommendation System
Suggests alternative routes, rollback strategies, and retry logic from history.
Intelligence Layer

Inclusion of Artificial Intelligence

Four AI capabilities woven into the core workflow, turning reactive operations into proactive network intelligence.

Auto-identifies device-specific issues from logs, surfacing root cause with context-aware fixes.

Suggests optimal routing paths and device upgrade schedules from network topology.

Predicts potential failure points from historical data before an upgrade begins.

Offers intelligent rollback decisions with context-aware fixes, not generic retry loops.

AI CORE Detect Route Predict Rollback
Ideate Phase · Structure

Information Architecture

A role-based navigation structure organizing every function around the mental model of the operators who use it. Four levels deep where the workflow demands it, flat everywhere else.

Scroll to see the full diagram →
Dashboard
Notifications
Work Orders
Create New
In Progress
Completed
Failed
Auto-Rollback & Suggestions
Devices
Upgrade History
Logs
AI Recommendations
Settings
Manual Review
Navigation node
AI-assisted node
Ideate Phase · Flows

Process Flow

The redesigned end-to-end work order lifecycle, from detection through automated resolution, with AI decision points where the old flow relied on manual guesswork.

SYSTEM AI LAYER USER Issue Detected Auto-Identifyaffected devices Smart Templatepre-loaded steps Batch Execute Live Tracking OK? Auto-Rollback + Fix Notifyall roles yes no
Design Phase · Transformation

Before vs After

Five workflows, shown as they were in the old Fusion Classic interface and as they exist now in Redwood. Some of this landed exactly as planned. One didn't, and I've left that in.

IS

Device Selection

Issac · IT Administrator
Before and after: Issac device selection
~30 min
Manual, stale inventory
Under 2 min
AI auto-select
122 of 127
Auto-verified
122 of 127 devices auto-verify against live inventory now, up from zero. The other 5 still get flagged for manual review, mostly older NF-SBC units where the ping response is inconsistent enough that we didn't trust full automation yet. Issac still opens those by hand.
IS

Upgrade Execution

Issac · IT Administrator
Before and after: Issac upgrade execution
700 min
20 devices, CLI
~11 min
AI batch
–98%
Time reduction
The 18-step MOP used to live in a PDF Issac kept open on a second monitor. It's baked into the batch flow now, so there's nothing to refer back to mid-upgrade. In testing, two of five participants still asked what happens if it fails partway through before they'd trust it. The live progress view helped, but it took a run or two before people stopped double-checking.
DK

Work Order Monitoring

Devika · NOC Specialist
Before and after: Devika monitoring
No visibility
CLI logs only
38 sec
avg. detection time
Auto
Stakeholder alerts
Devika's workaround before any of this existed was a personal spreadsheet of which error strings usually meant what. Some of that logic is in the AI root-cause matching now. Not all of it. She still knows things the system doesn't, and she'll tell you that.
RN

Change Request

Rick · Network Administrator
Before and after: Rick change request
8+ hours
Spreadsheet CRQ
10–15 min
for single-path fixes
~60%
of Rick's CRQs
This is the one that didn't fully land. AI path recommendation works well for straightforward degraded-route cases, about 60% of what Rick handles. For multi-hop reroutes with more than two viable alternatives, the confidence score kept coming back too close to call, and Rick said flatly he wasn't going to accept a route he couldn't reason through himself. We scoped it down: AI handles the simple cases end to end and hands off to Rick's original manual flow, telemetry pre-filled, for anything more complex. Smaller win than the dashboard implies, but the more honest way to ship it.
SM

Health Monitoring

Sarah · System Architect
Before and after: Sarah health monitoring
47-min stale
3 siloed tools
30-sec refresh
1 unified view
78%
confidence, 96hr window
Sarah's three old tools measured uptime slightly differently from each other. One counted scheduled maintenance against the SLA, two didn't. Merging them into a single dashboard forced an actual decision about which definition was correct, which turned into its own two-week conversation with the compliance team before we could ship one number everyone agreed on.
Design Phase · Hi-Fidelity

The redesigned product

Twenty-five hi-fi screens built on the Oracle Redwood Design System, one tailored console per persona, each turning the pain points into calm, guided, AI-assisted workflows.

Issac · IT Admin
OSDMC Work Order Admin dashboard: Issac, IT Administrator
Work Order Dashboard
Work Order Overview
Real-time KPIs and a proactive maintenance banner replace the old CLI-and-spreadsheet workflow, with AI flagging what needs attention before Issac has to go looking for it.
AI · Auto-Rollback
AI Auto-Rollback screen with root cause analysis
The One-Click Resolution
AI Auto-Rollback, WO-2025-4820
Live rollback progress, AI root-cause analysis, and an "Apply AI Fix Plan" action. The 50-step manual procedure reduced to a single confident click.
Devika · NOC
NOC Dashboard live work order monitor for Devika
NOC Specialist
Live Work Order Monitor
A single queue with visual status and AI quick-insights ends Devika's reliance on memory, scripts, and scattered terminal windows.
Devika · NOC
Failure triage screen showing AI-assisted diagnosis
Failure Triage
AI-Assisted Diagnosis
Root cause surfaced automatically with suggested fixes, cutting manual error diagnosis from 120 minutes to under 10.
Rick · Network
AI route optimisation comparing paths for Rick
Network Administrator
AI Path Optimisation
Current, recommended, and alternate routes compared side by side with latency and packet-loss, replacing Rick's spreadsheet hand-offs.
Sarah · Architect
SLA Compliance report screen for Sarah, System Architect
System Architect
SLA Compliance Report
Auto-tracked SLA metrics with AI predictions turn Sarah's lagging, siloed dashboards into a single source of truth. The numbers mirror the results below.
Built with the Oracle Redwood Design System v2.0. 25 hi-fi screens across 4 personas
Ideate Phase · Journeys

As-Is vs To-Be Journey Maps

Mapping every stage of the current workflow against the redesigned experience to quantify savings at each step.

Scenario 1: Software Upgrade (Issac · IT Admin)

StageAs-IsTo-Be
Identify IssueBug found on NF device; opens support ticketSame
Patch DevelopmentTakes 6 hours for dev teamSame, this part didn't change
Device SelectionManually checks specs (~30 mins)System auto-identifies affected devices
Procedure Follow-upRefers to lengthy 18-step MOPPre-loaded automated instructions
ExecutionManual upgrades per device (~35 mins × 20)Batch upgrades (5–10 mins/device)
Status TrackingCLI-based; frequent manual checksUI-based with real-time tracking
CompletionManually commits upgradesSystem auto-commits
NotificationsNone; needs follow-upsAI-powered alerts to all stakeholders

Patch development stayed a 6-hour manual job for the dev team. That part was never in scope, the redesign only touches what happens after the patch exists.

Scenario 2: Routing Update (Rick · Network Admin)

StageAs-IsTo-Be
Problem DetectedPoor route performance in OSDMCSame
PlanningSpreadsheet-based path updatesSystem suggests optimal paths
Change RequestManually filled formAuto-generated from detected anomalies
Approval & ExecutionManual via CLI; 8 hours of effortPre-built template; 10–15 mins for simple cases
Status MonitoringManually watched terminalDashboard with progress bar
Commit ChangesManually doneAuto-committed by system
Failure HandlingManual CLI rollbackInstant rollback + AI-recommended fix
Final NotificationVerbally sharedAutomated multi-user notifications

The "10–15 mins" row only holds for single-path fixes. Multi-hop reroutes still route back through Rick's original manual process, see the Before/After section above for why.

Error Taxonomy

Known failure modes & how the system resolves them

Error TypeCauseImpactAs-IsTo-Be
Device Not ReachableNetwork connectivity issueWork Order failsManual rollback & CLI diagnosticsAuto-rollback & AI log detection
Version MismatchOutdated device inventoryUpgrade failsManual verificationSystem auto-checks versions
MOP Step MissedHuman error in 18-step flowIncomplete executionRerun full MOPAuto-guided step validation
Route ConflictOverlapping routing pathsCall flow degradationManual CLI auditAI route suggestions
Dependency DeletedResources removed in rollbackRollback failsManual escalationSystem alerts + retry logic
Notification MissedNo automated communicationDelayed responseManual status requestsPush notification engine

This list is what showed up during the 12-week testing window. It's not exhaustive, Devika's team still catches error patterns the taxonomy doesn't have a row for, and probably will for a while.

Test Phase · Results

SLA Compliance Impact

Interactive prototypes tested with real users produced measurable gains across every SLA metric tracked by Verizon network operations.

Work Order Completion Rate
65%94%
+29%
Mean Time to Resolution
120m10m
−91%
User Satisfaction Score
3.14.7
+52%
Rollback Success Rate
55%98%
+43%
SLA Breach Incidents / Qtr
234
−82%
Downtime / Upgrade Window
45m8m
−82%

These numbers are from the Q2 2025 reporting window, the first full quarter after rollout. Q1 was rougher than either quarter shown here, since the team was still learning when to trust the AI rollback versus stepping in manually. Sarah's team expects the numbers to hold, but hasn't called it a trend yet.

Test Phase · Narrative

Story Board

The redesigned experience, told as the specialist would live it, from that 2 AM alert to a calm, one-click resolution.

Step 01
The 2 AM Alert
A critical upgrade fails. The specialist gets a push notification, not a phone call hours later.
Step 02
AI Diagnoses
The AI Log Analyser identifies the root cause and surfaces the affected devices instantly.
Step 03
One-Click Rollback
Instead of 50 manual steps, auto-rollback executes with an AI-recommended fix in minutes.
Step 04
Resolved & Notified
Service restored. Every stakeholder is auto-notified. The specialist goes back to sleep.
Conclusion

From reactive to intelligent

Most of this shipped the way it was designed to. Issac's upgrades run in minutes instead of hours. Devika stopped digging through raw logs at 2 AM. Sarah's three dashboards became one number everyone trusts. Rick's is the exception, AI handles his simpler routing calls and still hands the hard ones back to him, which is probably closer to how AI-assisted tools should work than a dashboard that claims to solve everything.

What I took from this project is that the win wasn't the automation itself. It was figuring out, screen by screen, where automation earned trust and where it didn't. That distinction is the actual deliverable, more than any single metric on this page.

Shipped as designed
Device selection, upgrade execution, work order monitoring, and health monitoring all run close to the original plan.
Scoped down, not shipped as pitched
Route optimization covers the simple cases well. Complex multi-hop reroutes still go back to Rick.
Still being watched
One quarter of clean SLA data isn't a trend yet. Sarah's team is waiting for Q3 before calling it settled.