Role Staff Product Designer, Strategic Lead

(System architecture, exec alignment, business case development)

Company Pinterest

Timeline 2024–2025

Domain Ads · Monetization · Value-Based Bidding

Impact $700M addressable market · $100M NCA opportunity · DLT-aligned MVP

We had automation momentum—$523M in P+ revenue, ~33% of daily lower-funnel spend. But advertisers couldn’t tell us what was worth more. They could optimize for conversions, but not for valuable conversions. Value Expression was the system to close that loop.

The Situation

In 2024, Pinterest had a competitive problem hiding in plain sight.

Performance+ was working. Revenue was growing. But we were optimizing for the wrong thing. When an advertiser ran a P+ campaign, we optimized for conversions—any conversions. A new customer worth $500 in lifetime value was treated the same as a repeat customer worth $50.

Meta and Google had solved this years ago. Their value-based bidding systems let advertisers tell the platform what outcomes were worth. Their ML systems learned from those signals and optimized accordingly. Pinterest was 8 years behind.

The gap wasn’t just technical—it was strategic. Sophisticated advertisers allocate budgets based on LTV, not just conversion volume. They were spending those budgets on Meta and Google, not Pinterest. We were losing share of wallet because we couldn’t speak their language.

I was tasked with designing the system that would close this gap.

The System: Encode → Inherit → Prove

I designed Value Expression as a closed-loop system with three parts:

[Image: System diagram showing Encode → Inherit → Prove loop]

Encode: Advertisers set value rules at the account or campaign level. “New customers are worth 2x.” “High-intent signals are worth more.” These rules live in a central place and apply across campaigns.

Inherit: When campaigns run, the rules flow into optimization. P+ uses the value signals. Bid Multiplier adjusts bids accordingly. Audience targeting respects the rules. The advertiser’s intent propagates through the system.

Prove: Reporting shows whether value was honored. Not just ROAS, but value-adjusted ROAS. Not just conversions, but the mix of new versus existing customers. Advertisers can verify the system is doing what they asked.

This wasn’t three features bolted together. It was a system designed to work as a loop—where proving feeds back into encoding, and encoding shapes inheritance.

What I Did

Driving Scope with JTBD Framing

Creating Executive Alignment Materials

Getting DLT (Design Leadership Team) alignment on the Q1 MVP required a specific kind of storytelling. Executives don’t read 30-page decks. They need one loop and one proof.

I created a combined deck that showed:

1. Momentum: P+ is working ($523M in revenue)

2. Gap: But we can’t capture value-based budgets (competitive disadvantage)

3. Opportunity: ~$700M addressable market in NA alone

4. Ask: Commit to shipping the closed loop in Q1

The “one loop + one proof” structure—showing the system and proving the opportunity—got alignment in a single meeting.

[Image: Market sizing / DLT deck slide]

Defining Success Metrics

Value-based bidding requires value-based measurement. I defined the proof metrics that would show whether the system was working:

Value-adjusted ROAS: Not just return on ad spend, but return weighted by the value the advertiser assigned.

New vs. existing customer mix: Showing that the system is actually prioritizing the audiences the advertiser cares about.

Signal coverage: What percentage of conversions have value signals attached?

These metrics became the success criteria for the Q1 MVP.

The project had natural scope creep. Every team had ideas for what Value Expression should do. I used Jobs-to-be-Done framing to separate concerns:

Govern: Account-level value rules. “Across all my campaigns, new customers are worth more.”

Express: Campaign-level fine-tuning. “For this specific campaign, I want to bid more aggressively on high-intent signals.”

This distinction—govern versus express—reduced cross-org churn. Teams knew which layer they were building for. Scope decisions became clearer. The JTBD framing turned philosophical debates into practical product boundaries.

Authoring the Strategic Narrative

Before any design work, I had to frame the opportunity in a way that would move executives. “Value-based bidding” is technical jargon. “Closing an 8-year competitive gap” gets attention.

I authored the narrative that positioned Value Expression as: “from any outcomes → higher-value outcomes.” This became the framing the cross-functional team used in every discussion.

[Image: Strategic narrative deck/slide]

Ship the closed loop first. The temptation was to ship encode, then inherit, then prove sequentially. I pushed to ship the full loop—even if minimal—because the value is in the connection, not the individual pieces. A value rule that doesn’t affect optimization isn’t useful. Optimization without proof isn’t trustworthy.

Parity before innovation. We weren’t trying to out-innovate Meta and Google. We were trying to catch up to table stakes. I framed this as “parity unlocks spend; clarity builds trust.” Don’t sell innovation—sell competence. The sophisticated advertisers will appreciate not being oversold. Executives don’t dig into appendices. I kept the market sizing to one slide with directional footers. The modeling lived in backup slides for anyone who wanted it. The headline number moved the decision.

Value Expression doesn’t exist in isolation. It’s the “fine-tuned automation” layer of a spectrum:

• Full Auto (P+): Set a goal, we do the rest

• Fine-Tuned (Value Expression + Bid Multiplier): Express what matters more

• Recommendations (Learning Phase, Guidance): Here’s what we think you should do

• Manual: You decide everything

Value Expression uses the Bid Multiplier backend as its expression mechanism. When an advertiser says “new customers are worth 2x,” Bid Multiplier is how that intent gets executed in the auction.

~$700M addressable market (NA, directional)

~$100M new customer acquisition opportunity

DLT alignment to Q1 MVP via “one loop + one proof” storytelling

Sales resonance — Clarity on account vs. campaign controls reduced escalations

Value-based bidding adds a layer of abstraction. The advertiser says what matters, but the system decides how to act on it. That’s a trust problem.

Where might this fail? The system inherits value rules but might optimize in unexpected ways. An advertiser says “new customers are worth 2x” and expects to see mostly new customers. Instead, the system might find a pocket of high-value repeat customers the advertiser didn’t anticipate.

How do users recover? Proof metrics let advertisers verify the system is honoring their intent. Value-adjusted ROAS shows whether the value rules are affecting outcomes. New vs. existing mix shows who’s actually being reached. Advertisers can adjust rules based on evidence, not just faith.

What feedback improves the system? Every value rule is training data. When advertisers tell us what matters, we learn about their business. Over time, we can make smarter recommendations about value—not because we guessed, but because advertisers taught us.

REFLECTION

The JTBD framing (“govern vs. express”) was the unlock. Before that, every conversation was a scope debate. After, teams knew which layer they were building for.

The other lesson: senior audiences move on one loop and one proof. I spent weeks on modeling and analysis. The decision happened because of one slide that showed the loop and one slide that showed the opportunity. Everything else was credibility-building backup.

If I were starting over, I’d invest earlier in the proof layer. Encode and inherit are exciting to build. Prove is where trust lives.

Value Expression is the strategic layer that sits on top of Performance+ automation. It answers: “P+ optimizes—but for what?” The system connects to P+ targeting, audience creation, campaign insights, and Bid Multiplier.

Related work: Performance+ · Bid Multiplier · Guidance System

← Back to Work

[Image: Example slide showing system sizing/scale]