Appier grew through product acquisition, expanding from a single ad platform into a suite of enterprise marketing tools covering audience data, campaign automation, personalization, and performance analytics. Each product solved a real problem. Together, they created a new one.
As clients gained access to more services, the workflows required to connect them multiplied. A marketer running a personalized campaign now had to move across multiple products, each with its own interface, logic, and learning curve. The more powerful the platform became, the harder it was to use.
AIQUA's AI-powered platform helps businesses personalize customer experiences across all channels, from smart recommendations to data-driven campaigns.
(Lead designer from Jun 2018 to Mar 2025)
AIRIS is a real-time customer data platform that unifies behavioral, transactional, and offline data into a single customer view, enabling marketers to build segments and generate predictive insights without relying on data teams.
(Lead designer from Mar 2025 to Feb 2026)
User pain points
Creating Consistency Across Independent AI Features
More products
Appier expanded the marketing blueprint by acquiring tools, expanding the platform's surface area across audience, campaign, and content layers.
More fragmented workflows
Clients had to navigate across products to complete tasks that should have felt like a single flow.
More CSM dependency
As complexity grew, clients leaned on Customer Success Managers to bridge the gaps the UI could not.
Design challenge
The platform had become expert-dependent. Clients with deep product knowledge thrived. Everyone else waited for help.
The AI agent was not a feature addition. It was a rethinking of how clients interact with the platform entirely. Instead of requiring users to know which product to open and which workflow to follow, the agent offered a conversational layer where clients could describe their goal in plain language and be guided to the right action across the full product suite.
Internal research
Looking for Users Patterns and Needs
We started with a rapid Lovable prototype to explore the concept, then conducted interviews with Customer Success Managers who worked directly with enterprise clients every day. CSMs were a proxy for client behavior at scale: they could describe what questions clients asked before they could begin any campaign task.
Internal Research Finding
A Clear Pattern Emerged Across Every Interview.
Marketers never opened the platform thinking "I need to create a segment." They opened it with a business question: "Why is my push notification CTR dropping?" or "Which users should I target for this flash sale?"
Scenario 1
Open-Ended Entry
Users asked open-ended business questions and expected the AI to guide them toward building a segment, rather than knowing in advance which action to take.
Scenario 2
Data Accuracy
Users did not trust the numbers or results the agent surfaced. They needed explanations behind the data before they felt confident acting on it.
Scenario 3
Analyze Purpose
Some users had no predefined strategy. They preferred to review past campaign performance before deciding how to define their audience.
Design process
Designing the Conversation
Three behavioral patterns emerged consistently:
Users started with open-ended business questions rather than specific tasks
They needed data explanations before trusting AI-generated results
Many wanted to review past performance before committing to a strategy.
With those findings, I redesigned the agent flow using Appier's design system, moving from a rough prototype to a testable, high-fidelity experience ready for client validation.
The new flow guided users from a business goal through audience options, past performance comparison, and revenue analysis, before surfacing recommended strategies and completing the segment in a single conversation.
Client Research Finding
A Clear Pattern Emerged Across Every Interview
Eight client interviews surfaced four recurring personas, revealing that marketers approach the agent with fundamentally different goals depending on their role.
Use Case A
CRM Management
Use Case B
Product Revenue Focused
Use Case C
Campaign Optimization
Use Case D
User Acquisition and Growth
Design process
No handoffs. Four disciplines, shared loops
The AI agent was not built in sequence. Research, design, engineering, and data science worked in parallel throughout, with multiple feedback loops connecting every discipline at once.
User research fed insights directly into design decisions. Design and engineering iterated in sync, with build constraints shaping the next round of UI decisions. Research findings traveled further, informing how the model recognized patterns and formatted results for marketers. And client feedback collected at every stage looped back across all four teams simultaneously, keeping everyone aligned on what was actually working in practice.
Each loop tightened the agent further. Not just incrementally better, but more grounded in how real marketers think and work with every cycle.
Design outcome
The Agent Scope Expanded in Deliberate Stages
We scoped the agent deliberately, starting narrow and expanding as we learned. Segmentation came first, giving us a contained problem to validate the core conversation model. From there we extended to single push campaigns, then to full customer journey orchestration across channels.
Each new scope revealed behaviors we had not anticipated. The more ground the agent covered, the richer our understanding of how marketers think across different tasks, and the stronger the design patterns became as a result.
Design process
Brought the AI Agent to Journey Maps
Journey maps were one of AIQUA's most powerful features and one of its least used. In user interviews, marketers consistently described the same hesitation: building a journey required configuring multiple campaigns, defining timeout logic between nodes, and manually sequencing every touchpoint across channels. The setup cost was high enough that many clients avoided it entirely, defaulting to simpler single campaigns even when a journey would have driven better results.
Design outcome
The Hardest Challenge Was to Help Users Stay Oriented Inside Agent
A journey map contains multiple nodes, each with its own content, timing rules, and logic. Through user testing we found that marketers needed to move constantly between two levels: the overall map structure and the individual node details. They wanted to drill into a node, adjust the content, and immediately see how it fit within the larger journey without losing their place.
This back-and-forth between macro and micro became the central design challenge for the agent interface.
Design outcome
Design Components for AI Agent
Designing the agent required building a component system from the ground up. The layout established three modes: homepage, conversation, and side panel, giving users clear context at every stage. Within the conversation, components covered filter buttons, template cards, image-aware inputs, and a thinking status indicator that made the agent feel responsive during processing. Spatial rules governing message alignment, button placement, and suggested actions kept complex multi-turn conversations scannable and easy to follow.
Proposal cards and side panels addressed decision-making inside the conversation. Segment and campaign cards surfaced enough information for users to choose confidently without leaving the chat. Side panels kept both the conversation and detailed content accessible at once. Element created cards closed each interaction with a clear confirmation and a direct link to verify the output in AIQUA or AIRIS.
Success Story
LANEIGE Drove a 3.3x Increase in Campaign ROI with Smart Segmentation
24x
Efficiency
Audience Agent accelerates marketing decisions from an average of 3 Days to 1 Hour, leveraging Industry Expertise, Historical Customer Activity, and AI-Powered Predictions.
3.3x
ROI Boost
By leveraging Audience Agent’s AI algorithms, LANEIGE identified a high-intent segment accounting for roughly one-third of total members and focused its ad spend on this group. The result was a 3.3x increase in ROI, rising to 21.3%, with higher conversion rates and revenue achieved at a lower marketing cost.
219%
Uplift CTR
AI-driven smart segmentation and performance optimization delivered a 3.3x increase in ROI, a 219% uplift in CTR, and a 200% improvement in CVR, creating a highly efficient and continuously optimized data-driven growth loop.
1.3x
Average of purchase value
AI-powered audience segmentation identified higher-intent buyers, lifting the average order value by 1.3x compared to manual-targeted campaigns.
2x
Add to cart
Personalized campaign delivery reached users at the right moment in their journey, doubling the rate at which shoppers added products to their cart.
14x
Analysis efficiency
The AI agent reduced the time marketers spent analyzing campaign data and building audience segments, making the workflow from 2 weeks to 1 day.
Reflection
What We Learned
Defining AI limitations early matters more than we anticipated. When users understood what the agent could and could not do before they encountered the boundaries, they worked with it more effectively and complained less when it fell short.
The iteration process across four disciplines was harder to coordinate than expected. Engineering, research, product management, and data science each shaped the product in different ways. Clearer shared milestones would have made the parallel workstreams easier to align.
AI output is unpredictable and designers cannot fully control it. The real design challenge was finding the right boundaries: defining what the agent should attempt, using pre-set templates to constrain outputs, and building status components that help users understand what is happening and how to proceed.
Hallucination required a dedicated design response. We designed a reminder to surface when AI-generated results may be unreliable, shifting the responsibility from the user to notice errors toward the system to flag uncertainty proactively.
What the work delivered
LANEIGE reduced marketing decision time from three days to one hour using the Audience Agent.
Campaign ROI increased 3.3x, CTR improved by 219%, and CVR improved by 200%.
These results confirmed that closing the gap between a marketer's business question and a confident next action is the right problem to solve.
Diane Lee
Updated Sep, 2026