Designing AI-powered marketing dashboard that makes performance visible

The CLOVA AI message marketing solution was preparing to move from a free service to a tiered paid model, with two key goals: growing subscribed stores from 30K toward 50K and pushing paid conversion above 50%. Store owners suddenly had a concrete reason to ask what the AI was actually doing for them, and the team needed performance to be legible, not taken on trust.

I framed the dashboard as both a product-utility problem and a value-communication problem: it had to support real marketing decisions while making the impact of the AI feature understandable.

I listed up stores already subscribed to the solution with active, sizable audiences, 10,000+ subscribers with cart items in the last month, or 5,000+ with a purchase history in the last 10 days, and interviewed six store owners for an hour each: three using the AI features effectively, and three paying for message sends while barely touching them.

Performance analysis emerged as a recurring need. That finding gave the dashboard a concrete job: help users understand outcomes and turn those outcomes into more efficient marketing choices.

I worked bottom-up: first listing every input the dashboard could draw on (user data, behavior logs, message history), then grouping it by AI feature, and finally defining the metric each feature should be judged by, the expected impact a store owner would actually care about.

Team feedback reshaped the map before any UI existed:

• Set the user’s main goal per feature, so the most decision-relevant data leads each module.
• Added data types: ratio, number, and demographic information.
• Added highlights (key metrics) at the top level.
• Rethought the hierarchy between features, and noted aggregation-period limitations.

That map became the baseline I used to communicate with product and engineering while the detailed visualizations were still evolving.

Visualization was a decision tool, not decoration. I compared graph forms based on whether users needed to understand trend, magnitude, comparison, or contribution, and I worked through emphasis and hierarchy so the most important information did not disappear into equal-weight cards.

The rejected graph directions are part of the design story because they show how different visual forms answered different questions, and why the final hierarchy became simpler.

The dashboard was delivered on the CLOVA Commerce Solution component library, buttons, tags, inputs, tooltips, and chart color scales, with a fixed-plus-fluid layout spec so engineering could implement the two-column grid without guesswork.

I translated the information strategy into a dashboard that supported both scanning and deeper investigation. The visual hierarchy emphasized the most important outcomes first, while secondary details remained available without overwhelming the initial read.

Sensitive customer values are anonymized without changing the underlying structure.

After launch, end-user feedback split into satisfaction (demographic data), suggestions (cost savings, Excel export, category benchmarks), and frustration: users kept asking about aggregation criteria and didn’t fully understand the new features.

So the question became: how might we improve onboarding and minimize confusion? I answered it inside and outside the product, criteria tooltips and one-time entry-point guidance in the UI, anticipated questions added to the FAQ, feature education shared via Keynote, and this walkthrough video.

Subscriptions were maintained and grew: stores subscribed to the solution rose from 36,000+ to 37,000+ within a month of launch, against an end-of-year goal of 50,000+. Just as importantly, the work built a strong connection with enterprise users.

In 2025 I revisited the project as a quick design challenge for the agent era. The shipped product was AI-powered: AI reported message performance, and users could only view and interpret the dashboard. The rethink turns it into an AI-native “Smart Campaign Advisor” that proposes strategy and lets users adjust.

Consideration 2: balancing user authority and automation. The answer is constrained configurability: delegation and configuration are the two levels our target users live in, with full creation reserved for power users.