Ymax 1.0 — Brad Perlman

Context

Ymax is Yieldmo's AI-powered ad tech platform — advertisers use it to set up campaigns and deals, plan supply, build contextual targeting, manage line items, and monitor performance. I joined Yieldmo in May 2025 to take over Ymax day-to-day design from the CXO, who had been the sole designer up to that point. My role is owning feature work and ticket throughput across the platform; this case study highlights three areas where the work has been deepest.

Early on at Yieldmo, I led the PageMax interactive prototype for Cannes Lions 2025 in collaboration with the CXO. PageMax has since folded into Ymax, and the work below is what's lived in production since.

This case study covers three bodies of work: contextual targeting (live in production), the Extended Targeting redesign (built into Ymax 1.0 now), and reporting and insight surfaces (queued for Ymax 2.0).

Contextual targeting — supply package creation

Users build supply packages by describing what they want in plain language — text, image, video, or URL — instead of building inventory lists by hand. The system scans the input, returns matching segments, and bundles them into a saved package.

Within the create package modal, I designed the four input modes and the AI-Generated Attributes component, the loading sequence that runs while the search executes, and updated the Context Card on the results page.

AI-Generated Attributes block showing extracted bullet attributes below the uploaded image — AI-Generated Attributes. Bullets first ('Music enthusiasts / Tech-savvy individuals / Frequent travelers'), then prose expansion. Edit pencil lets users refine before the search runs.

The AI-Generated Attributes component shows the LLM's interpretation in plain language and populates the description field before the user commits to running the search.

Five-step loading sequence with checkmarks on completed steps — Instead of a spinner, the loading sequence shows users what the system is actually doing — five named steps, completion states, and an animated shape that runs throughout.

Before this addition, users who clicked the create button would end up staring at an empty results table while the search ran, causing a pretty poor experience. This led to product defining five states, which I then designed the sequence for.

Results page with the Context card on the left rail — Results page. The Context Card on the left rail mirrors the create-modal component.

The Context Card on the results page uses the same component as the input modal, so users can edit and re-run their search inline.

Extended Targeting — line item redesign

Product worked with our internal users to identify what was missing in our targeting capabilities, which led to me researching how DV360 handles multi-dimensional targeting and using those patterns as reference.

Before

Current targeting tab in Ymax with four fields — Current targeting in Ymax. Four fields total: Device, OS, Geo (country only), and a single Audience selector.

Internal users had been asking for frequency capping, dayparting, web-vs-app split, language scoping, multi-audience logic, and geo more specific than country level.

After

Redesigned targeting tab in default state — Default state. Most fields default to 'all' so line items run as broad network targeting until the user narrows them.

Redesigned targeting tab fully configured — Configured state. Nine dimensions including frequency capping, geo, device, traffic type, language, audience segments, supply source, and dayparting.

Geo targeting

Geo targeting has been updated to include four match types — City, Postal Code, DMA Region (a metro-area grouping used in ad targeting), and State — returned in a single query and mixable inside one selection.

Resolved postal codes shown as labeled location chips — Bulk paste resolved. Each pasted code resolves to a labeled location with remove and Exclude All / Include All toggles.

Our internal users had also requested the ability to bulk add zip codes. The parser flags invalid input with a clear constraint message — instead of silently dropping bad rows.

Reporting and insight surfaces

This work is delivered as design. Implementation is queued for Ymax 2.0.

The dashboard is a campaign-level view — KPI tiles, spend-vs-performance over time, total spend by domain, and performance comparisons across line items, segments, and URLs.

Full reporting dashboard with insight cards, KPI tiles, charts, and performance tables — Reporting dashboard. KPI tiles, dual-axis spend-vs-performance chart, total spend by domain donut, and segment / line item / URL comparison tables.

Yieldmate is the platform's AI assistant — it surfaces real-time insights when the system recognizes either an opportunity or a concern in a user's deal or campaign. The loop is straightforward: user sees an insight, navigates to where it applies, makes a change, and the deal performs better. Working with my team lead, we explored what these recommendations should look like — surfaced where the user is already working, paired with actions they can take.

Line items table with a Yieldmate tooltip suggesting a pause on the underperforming line item — Yieldmate inline. After deep-linking from the dashboard, the recommendation appears next to the row it's about.

Two things shape the pattern:

Each insight pairs a finding with a recommended action that deep-links to where it applies. Add Supply takes the user to a pre-filtered supply search. Go to Line Items lands them on the flagged row, with the recommendation inline. The user moves from observation to action in one click.
Coherent across surfaces. Yieldmate looks and reads the same way wherever it appears — dashboard, line items table, supply search — so users learn the pattern once instead of re-learning it everywhere.

What carries forward

That's the bulk of Ymax 1.0. The next case study covers Ymax 2.0, the platform rebuild currently in progress.