AI · Conversational UX · 0 to 1 · Affle 3i

A conversational AI campaign platform for non-marketers

Project
CTV AI Campaign Platform+ marketing site (in progress)
Company · Year
Affle 3i2025 to Present
My role
Product DesignerConversational UI · Figma Make · Claude Code
01 · The Challenge

The user had never
done this before

The brief: let a small business owner set up a Smart TV ad campaign without an agency, a media buyer, or technical knowledge.

The target audience had no mental model for TV advertising. They didn't know what a target audience was, what a media plan looked like, or what they needed to decide.

The team landed on a conversational interface. I worked across the conversation flows and UI alongside product and engineering.
The opening isn't
data collection.
It's a trust moment.
Before a user shares their business goals or budget, they need to feel like they're in capable hands. The first few exchanges set the tone for the entire session.
02 · What I Discovered

Early versions felt like an interview

V1
Too abrupt
Jumped straight to campaign parameters. No intro, no name, no language preference. Felt like a government form.
V2
Tested two approaches
Conversational vs. independent. The conversational version built more confidence. URL scrape became the first real trust signal: "it already knows me."
V3+
Communication at every step
Language preference, name collection, progress indicators, confirmation gates. Each addition answered: "What is happening right now?"
Scroll
03 · The Core Interaction

What the conversation
actually does

How the flow works
Welcome! 👋 Which language would you like to continue in?
English
Perfect. May I know your name?
Meher
Nice to meet you, Meher! Do you have a website or a physical store?
I have a website
Share your URL and I'll pull in your business details automatically.
mehersingh.netlify.app
⟳ Scanning website · pulling logo, brand colors, business details...
Found it! Here's what I picked up about Meher's portfolio. Does everything look right?
✓ Looks good, Continue
Every question was deliberate. Language first, because assuming English on a platform for local Indian businesses is a trust failure before the conversation starts.

The URL scrape is the key moment. Instead of "describe your business" (a blank canvas) the AI does the work and asks for confirmation. The user goes from author to editor.

Every step follows the same principle: the AI proposes, the user approves.
04 · Design Decisions

How the thinking worked

01
Script before screen
Conversation flow written as a script first, every question, every branch, before any UI. What the AI says is the product.
02
Reduce blank canvas moments
Open text fields without scaffolding spike drop-off. URL scrape solved business description. Every other input was pre-filled or choice-based.
03
Make the AI visibly working
Progress indicators, scanning animations, in-progress messages. Silence from an AI feels like failure. Every processing moment needed a signal.
04
Confirm before committing
Every major AI output (profile, campaign plan, creatives) had a review-and-confirm step. Approval gates preserve agency.
Scroll
The core principle
The AI proposes. The user approves.
The AI does the hard work (scraping, planning, generating) but never acts without explicit sign-off. An AI that supports judgment, not one that replaces it.
05 · How It Got Built

Using AI tools to design AI products

Figma Make shipped working UI from Figma's code output. Three live clickable prototypes, not mockups.

Claude Code prototyped conversation flows in hours.

Claude / ChatGPT helped iterate the conversation scripts with the team.

SME research tested two versions with real business owners. Findings shaped the direction.

Marketing site currently in build using Claude Code, carrying the conversational tone into the brand surface.
Make it real
in hours.
Learn faster.
Three shipped prototypes meant three rounds of real learning, each one changing something fundamental about the conversation design.
06 · Outcomes

Where it stands

3
Working prototypes shipped
Three live, clickable Figma Make prototypes. Each round of feedback led to a fundamental change in the conversation design.
2
SME research rounds
Tested two competing interaction models with real business owners. Findings shaped the final direction.
Live
In stakeholder review
Production-ready handoff. Marketing site for the product currently in progress in Claude Code.
07 · Takeaway

What I'd push harder on

The biggest gap is feedback during processing. Users need to feel the AI working, not just see a spinner. Richer in-progress communication is the next push.

In conversational AI, the words are the interface. Writing the script before designing UI saved weeks of rework.

More workAll work
02
Enterprise UX · AI
Optimization AI — Iris
Affle 3i · 2025–2026
AI transparency layer — CoPilot, Take Off, and Status modules that made recommendations account managers could actually trust. 75% adoption in 6 weeks.
View case study
03
AdTech · Rule Builder
Feed Manager
Affle 3i · 2025
Boolean rule builder redesigned from a developer tool into a plain-language decision interface for non-technical users.
View case study
Now playing
0:000:00