Skip to main content
← Back to projects

Case study · Klaviyo

Audience Optimization

An intelligent decision engine that determines who should receive which messages — removing people likely to unsubscribe, adding people likely to convert. I led the UX from beta to GA.

Audience Optimization
Company
Klaviyo
Focus
AI Decisioning · Trust · Analytics
Impact
~5% KAV lift · −1.4% unsubscribe rate

Background

Sending to the wrong people, at scale.

When multiple campaigns run simultaneously, marketers have no way to ensure the most valuable message reaches the right person. They rely on send time as a crude priority signal — manual, unscalable, and blind to engagement risk. The result: higher unsubscribes, damaged sender reputation, and lost revenue.

No competitor offers content-aware, real-time volume control that dynamically adjusts who gets what — based on message value, channel cost, and recipient behavior. That's the gap AO fills.

Setup experience

One toggle. Maximum impact.

AO lives in the audience selection step — a single toggle that removes recipients at high unsubscribe risk before the campaign sends. Simple to enable, easy to understand.

Setup — audience selection with AO toggle

Design thinking

What changed → Who was removed → How it changed.

The model doesn't understand campaign intent — so I couldn't make it invisible. I structured the UX as a story: where is AO active, who did it affect, and did it work? Three design questions, three sections of the experience.

Showing where AO is active

AO is configured at the audience level but surfaces across campaign setup, the message list, and analytics. I introduced an AI badge with Design Systems to make it visible end-to-end.

Campaign list with AO badge
AO on dashboard
Message-level indicator
AO applied

Showing who was removed

I explored three design directions for the campaign analytics view — from minimal summaries to full transparency panels. Option 3 won: per-feature toggles that give customers intentional control without overwhelming them.

AO option 1
Option 1 — Summary box
Simple, low friction — but lacks control and transparency over what was optimized.
AO option 2
Option 2 — Expanded metrics
High transparency — but cognitively overwhelming at decision time.
AO option 3
Option 3 · chosen
Per-feature toggles — clear control, flexible, scalable to future milestones.

Analytics

Did it work? Make it obvious.

I partnered with Data Science to connect model decisions to clear, measurable outcomes — profiles removed, unsubscribe rate improvement, and aggregate lift across campaigns — surfaced directly in the campaign overview.

Campaign analytics — profiles removed and unsubscribe lift

AO reporting

Roadmap

From protection to full orchestration.

AO ships in phases — each expanding what the model can do and how much control marketers have.

M1 · Live
Remove likely-to-unsubscribe recipients
Email campaigns only. Protect list health. Single objective model.
M2 · Q2
Multi-objective removal — all channels
Remove recipients unlikely to engage or convert, or at unsubscribe risk. Email, SMS, Push, WhatsApp.
M3 · Q3
Add high-value recipients
Add recipients likely to engage or convert — unlocking incremental revenue from people not in the original audience.
M4–5 · H2
Cost-aware orchestration + customer controls
Factor in channel cost. Let marketers tune objectives, define exclusions, and debug decisions.
M6–7 · H2
Message prioritization across campaigns & flows
Send the most relevant message to each recipient, considering content value, channel cost, and behavior — across all active campaigns and flows.

Impact

Results

~5%
KAV lift for companies using the feature
−1.4%
Unsubscribe rate reduction (statistically significant)
$640M+
Projected impact as anchor feature of Marketing Analytics SKU
20%
Target customer adoption rate at scale
Reflection
The hardest design problem wasn't the setup — it was making the model's decisions trustworthy. When customers can see exactly who was removed, why, and what changed, they stop second-guessing the AI and start relying on it. Trust is the unlock for adoption at scale.