AdTech · Dynamic Product Feeds · Affle 3i

When the ad converts, but the product can't

Project

Feed ManagerEcommerce product feed rules

Company · Year

Affle 3i2025

My role

Product DesignerInteraction design · Rule architecture

01 · The Challenge

When the ad converts, but the product can't

You scroll Instagram, see a red dress in a carousel ad, tap through, ready to buy. The size is sold out. You bounce. That's not just a wasted click. It's a lost purchase, an inflated CAC, and a shopper who now associates the brand with disappointment.

Feed Manager is the tool ecommerce clients use to control which products from their catalogue feed into Meta, Google, and other ad platforms. The rules are boolean: include if in stock AND size available, exclude if discontinued, OR if rating is below 3.

The existing interface exposed this logic raw. Account managers made errors, skipped the feature, or escalated to engineering. The brief: make this feel like a decision tool, without hiding the power technical users needed.

Every wrong product
shown is a lost sale,
not just a lost click.

When stock data is stale or rules are misconfigured, the ad still spends, the shopper still taps, but the conversion never lands. CAC inflates. Trust in the brand drops. Feed rules sit at the centre of all of it.

The platform · Real screens

Three steps. One flow.

Map Fields step showing conditional Only IF rules per catalogue field, with AND/OR logic, date ranges, and a Preview button

01Map Fields · Conditional rules per field

Exclude Associated Variants. Drop variants when a percentage match price ranges or have blank sale prices

02Exclude Variants · Stop wasted ad spend

Auto Categorisation tab. AI assigns Google product taxonomy categories

03 · AutoAI assigns Google taxonomy

Manual Categorisation tab with custom When clauses

03 · ManualCustom rules when AI isn't enough

Advanced rule editor with nested conditional logic and Combine fields

+Power user · Nested conditions

Scroll

02 · The Core Problem

Query builder vs. decision tool

Before: query builder

Technically correct. Humanly unusable.

Asked users to think in SQL-like logic. Correct for engineers, alienating for account managers who think in product outcomes.

"I just want to show products that are on sale and have good reviews. Why does this feel like writing code?"

After: decision tool

Complex logic. Plain language surface.

Rules grouped into plain-English blocks: "Include products where…" and "Exclude where…". Logic lives under the surface; decisions sit on top.

"I can see exactly which products will be in the feed before I save."

03 · Design Decisions

How the thinking worked

AND / OR / IF chips, not dropdowns

Always-visible chips made logic readable at a glance. AND = blue. OR = amber. IF = green. Color made it scannable.

Separate Include and Exclude

Distinct named groups halved the cognitive load. Same logic underneath, completely different surface.

Live matched product count

Every rule change recalculates in real time. Mental model shifted from "setting up a filter" to "sculpting a feed."

AI suggestions with reasoning

Not just "add this rule" but "here's why this matters for your campaigns." Trust requires context.

Scroll

The key insight

No raw logic exposure

Fields named in business language, operators written out, values formatted. Power users still get full complexity. They just don't think about implementation while using it.

04 · Outcomes

What it delivered

↓

Engineering escalations

Account managers who previously escalated to engineering could operate the tool independently.

Core design decisions

Chips, group separation, live count, AI suggestions. Each targeted a specific friction point identified through user conversations.

48K+

Products manageable

The rule engine supports feeds of 48,000+ products with complex nested logic, made accessible to non-technical users.

05 · Takeaway

What stuck with me

Boolean logic is the product. You can't design it away. The job was making it legible.

The live count was the breakthrough. When users see the consequence before committing, they stop being afraid of mistakes. That principle applies to every complex configuration tool.