Designing AI-Powered Routing & Automation Recommendations (CX Product)
AI recommendations for identifying routing and automation opportunities
Company
Lang.ai
Role
Product Designer
Industry
AI · SaaS · Automation
Duration
3 Months
Overview
CX teams wanted to automate more of their support workflows, but knowing what to automate - and whether the AI could be trusted to do it correctly - was a major barrier.
Agents were manually routing tickets and managing repetitive workflows, while lacking visibility into how successful an automation would be or how much time and money it could save.
I designed Automation Recommendations to turn AI from a black box into a transparent decision-support system, giving teams the evidence and control they needed to confidently automate.
Impact:
↓ 20% ticket resolution time
↓ Manual routing and repetitive work
↑ Confidence in automation decisions
✓ Successfully rolled out to real customers
The problem
Support teams were sitting on large amounts of repetitive work, but automation wasn't as simple as telling users:
“Let AI handle this.”
CX agents had several questions:
What should I automate?
Will this automation actually work?
How confident is the AI?
What happens if it gets something wrong?
How much time and money will this save me?
Routing was particularly repetitive.
For example, a ticket about a payment issue could potentially be routed automatically to the appropriate team instead of being manually reviewed and assigned.
But without evidence that the automation was reliable, users were understandably reluctant to trust it.
The opportunity was therefore not just automation.
It was building enough transparency and control for users to confidently adopt it.
My role
I owned the design end-to-end, working closely with Product and Engineering to turn the automation capabilities into a customer-facing experience.
I:
Conducted research with CX teams to understand automation barriers
Identified trust and discoverability as key adoption problems
Designed the recommendation experience and configuration flows
Defined how confidence and success should be communicated
Worked through edge cases and failure scenarios
Iterated through prototypes and customer feedback
Supported progressive rollout to real customers
The solution
Instead of expecting users to identify automation opportunities themselves, the product proactively surfaced recommendations based on their existing ticket data.
01. AI-powered recommendations
The system identified recurring routing and automation opportunities and recommended them to users.
→ Users didn't have to guess what was worth automating.
02. Transparent confidence & success
Each recommendation showed evidence behind the opportunity, including:
How frequently similar tickets occurred
How often they were successfully automated or routed
Expected time savings
Potential cost savings
Confidence in the recommendation
→ Users could evaluate the recommendation rather than blindly trusting the AI.
03. Control over automation
Teams could configure the system around their own risk tolerance and workflows.
They could define:
Confidence thresholds
Ticket types
Destination teams
Recommendation frequency
Other automation parameters
Recommendations could be delivered daily, weekly or monthly, depending on what worked best for the team.
→ AI remained assistive and controllable, rather than taking decisions away from users.
Designing for trust
One of the biggest design challenges wasn't making the recommendations intelligent.
It was making them understandable enough to trust.
I focused on answering three questions directly within the experience:
Why am I seeing this recommendation?
→ Show the underlying pattern.
How confident should I be?
→ Make confidence visible.
What's in it for my team?
→ Clearly communicate time and cost savings.
This turned automation from an opaque AI capability into a transparent product decision.
How I approached it
The project combined continuous research, rapid prototyping and close collaboration with Engineering and Product.
I:
Interviewed CX teams about automation workflows
Mapped routing and automation pain points
Explored different ways of communicating AI confidence
Prototyped and tested recommendation patterns in Figma
Iterated based on customer feedback
Worked closely with engineering on implementation constraints
Used progressive rollouts to validate the experience with real customers
The goal was to find the right balance between automation, transparency and user control.
Outcomes
Automation Recommendations were progressively rolled out to customers and ultimately shipped as part of the Lang.ai product.
Results:
20% reduction in ticket resolution time
Reduced manual routing and repetitive operational work
Helped teams identify automation opportunities they might otherwise have missed
Increased transparency around AI recommendations
Gave customers control over how and when recommendations were surfaced
Enabled CX teams to make automation decisions with greater confidence
Key takeaway
The fastest way to make AI useful isn't always to give it more autonomy. Sometimes it's to give users better reasons to trust it.
This project shaped how I approach AI product design: recommend first, explain clearly, quantify the value and give users control.
Other projects
Turning Customer Conversations into AI Summaries & Sentiment Insights (CX Product)
AI-powered summaries and sentiment analysis for faster customer insight discovery
Designing a Multi-Controller Integration & Setup Experience (DevOps)
Introducing a new CI integration and rethinking existing integration workflows for platform engineers
Redesigning the Release Notes Page (DevOps)
Rebuilding a searchable, structured release notes experience with Claude Code
Defining an AI-Powered Analytics Dashboard (DevOps)
Turning fragmented engineering data into actionable insights
Reimagining CI Troubleshooting with AI Dashboards (DevOps)
An AI-assisted workflow for investigating pipeline failures
Building a Design System from Scratch (CX Product)
Creating a documented component library that improved team delivery by 25%
Building a Design System for consistent AI Prototyping (DevOps)
Structuring the Jenkins design system for Claude Code and more consistent AI-generated prototypes