IFS · 2026

Duplicate Work Order Detection through AI

Avg. minutes to identify duplicate 3

Role: Senior Lead UX Designer
Duration: 3 months
Year: 2026
Client: IFS

Overview

Project overview

In enterprise asset management, multiple work orders can sometimes be created for the same underlying maintenance issue. Identifying these duplicates can be difficult when records contain different descriptions, statuses, or operational details.

The challenge was to explore how the user experience could help maintenance teams recognise potential duplicates, compare relevant information, and make informed decisions without disrupting legitimate maintenance activities.

The problem: A duplicate work order is not always an exact copy.
The decision: Designed an AI-assisted duplicate detection flow where the AI suggests and the planner decides.
The outcome: 3 Avg. minutes per duplicate

The challenge

Understanding the problem

A duplicate work order is not always an exact copy. Two work orders may describe the same fault using different wording, originate from different users, or refer to the same equipment and maintenance issue.

Work Order	Description	Equipment
WO-1042	Motor overheating	Pump P-102
WO-1078 Potential duplicate	Excessive motor temperature	Pump P-102
WO-1091	Replace conveyor belt	Conveyor C-204

The first two work orders could represent the same maintenance issue, while the third is unrelated.

The UX challenge is that similarity does not automatically mean duplication. Users need enough context to make an informed decision without accidentally losing valid maintenance work.

Key user pain points

Identifying related work orders across potentially large lists.
Comparing descriptions, equipment, status, and other relevant details.
Distinguishing genuine duplicates from legitimate separate tasks.
Resolving duplicates without losing important maintenance information.
Making decisions efficiently without disrupting existing workflows.

Key question: How might we help users confidently decide whether two or more work orders are the same issue, quickly and safely?

The approach

How I approached it

07DeliverWorked with Product Management, Program Management, Engineering on feasibility and handoff

01Discover

01DiscoverIdentified research opportunities and ran stakeholder workshops to gather requirements and built a story board for framing the right problem.
02Map the current stateDocumented how duplicates are handled today.
03DefineIdentified Personas and Synthesised pain points into objectives and goals
04Map the future stateDesigned the ideal flow from flagged duplicate to resolution
05Wireframe and validateTested early concepts with stakeholders
06Design and testBuilt high-fidelity screens and ran usability testing
07DeliverWorked with Product Management, Program Management, Engineering on feasibility and handoff

Swipe for more steps →

The solution

What I designed

Designed an AI-assisted duplicate detection flow where the AI suggests and the planner decides.

I designed a flow in the work order list that follows established human-AI design patterns. A planner selects one work order and asks for duplicates. The system finds likely matches, scores them, and explains why each one matched. The planner then chooses which ones to merge into the original. The AI never merges anything on its own.

How it works

Select and generate. The planner picks one work order in the list (1,000 records, 20 per page) and presses Generate duplicates.
Set the search scope. A side drawer opens with the site, date range, and organisation pre-filled from the selected order. The planner can change any of them.
Review matches. The original sits at the top. Below it are cards for each possible duplicate, sorted by match score, with the reasons for the match.
Decide. The planner ticks the orders that are really the same fault and merges them. The rest stay separate.

AI design patterns followed

Pattern	What it means	Where it shows up
Human-in-the-loop	A person makes every consequential decision.	The AI only suggests. Merging needs the planner to tick cards and press the merge button.
Suggest, don't automate (assistive AI)	The AI proposes and the user stays in control.	Matches are shown as possible duplicates, never as merged or closed.
Confidence indicator	The system shows how sure it is.	Each card has a match score (for example 94%) and a label: High, Possible, or Low match.
Explainability ("why this suggestion")	Users can see the reasoning behind a suggestion.	Each card has a "Why it matched" line, such as same equipment, same site, similar description, or reported 2 days apart.
Smart defaults, editable	The system fills in what it can and lets the user change it.	Site and date range are pre-filled from the selected order but remain editable.
Scoped context	The AI works within limits the user can see.	Filters appear as chips in the drawer, so the planner knows what was searched.
Progressive disclosure	Show what's needed now and reveal the rest later.	Two steps, filters first and matches second, instead of one crowded screen.
Consequence clarity	Make the result of an action clear before it happens.	The button reads "Merge 2 with WO-1042", and a note says history is kep