Behavioral design | AI interaction design | Design strategy
Designing AI as added intelligence, not a coworker
First-touch AI for non-technical users
UCaaS
SUPERFONE
Y2025
TARGET: INDIAN SMB
MULTI-LINGUAL
PROJECT_AI_CALLER_TUNE
CASE_NO_02
/OCT
LIVE IN PROD
03 WEEKS
Role
CONCEPT → LAUNCH
Design lead
Case highlights
Interaction model lets users meet AI without talking to it
A case study on first-touch AI design for users who'd never used AI for work and how an interaction model became the blueprint for an entire product's AI behaviour.
Behavioural reframing
How a behavioural insight became a product-wide AI interaction pattern
Designing for AI illiteracy
Designing the first AI touchpoint for users new to AI in a business context
AI interaction design
Building an editing-first model for users with no prompt skills
Systems thinking
Turning a single feature into a reusable pattern across four surfaces
Background
The project, the user, the surface
Superfone is a UCaaS product for India's small businesses. The kind of business run by one or two people wearing every hat. In late 2025, we began bringing AI into the product. The caller tune was the first feature it touched.
Who
Small business owners across India : clinics, salons, coaching classes, retail shops. AI-curious. AI-unfamiliar.
Why
These users had only ever used AI for entertainment (ChatGPT, Meta AI). Never for work. We didn't yet know how they'd respond; would they trust it, engage with it, break it?
What
Introduce AI to Superfone users for the first time. Build the proving ground for every AI feature that would follow.
Where
The caller tune — a low-stakes, low-engagement feature, deliberately chosen so AI-driven activity was pure upside.
When
Setup moment for new users. In-app launch promotion for existing ones.
How
Low entry requirement. Connect social media and website. AI handles the rest.
Superfone caller tune is a short greeting the user’s customer hears when they call the business.
Two-part problem
There were two problems to solve. Solving the first revealed the second.
1
The engagement gap
Most users left the caller tune on default.
2
The ownership gap
Editing AI content required skills users didn't have.
Problem 1 : The engagement gap
62% of users never customised their caller tune
A generic caller tune is automatically setup by default when a new user is registered : “Thank you for calling [business name]. Please hold.”
Current state
Users would type a message into a textbox, and a text-to-speech feature would read it aloud.
Business hours
Tune will play during of these hours
Type your message in any language
Convert a text message into a professional caller tune.
Priya
Varun
Mary
View suggestions
“Thank you for calling Tarun Dental Clinic. Please hold and someone will answer soon.
120/180 words
save & publish
Caller tune
11:42
Why it wasn’t working
Most users left the default caller tune in place for months after sign-up. Customisation was rare.
We investigated why most users left their caller tune on default. Three things came up consistently.
They didn't know what good looked like
Users didn't know what to write, what tone to strike, what details to include. So they wrote nothing.
They didn't see the value
Users didn't realise a good caller tune builds trust, signals professionalism, or can be used to announce offers, hours, or news. The feature sat in their blind spot.*
They didn't have the writing skills
Even users who wanted to customize struggled to write a polished business greeting in English, in their native language, within the character limit. Most gave up.
*A side observation: The caller tune plays only for the user's customers, never for the user themselves. So even when they did set one up, they had no feedback loop on whether it sounded good. We noted this but didn't solve for it directly, that's a different design problem.
So we asked a different question
What would make users feel proud enough of their caller tune to actively set one up?
How do we show them what good looks like without preaching, without overwhelming them?
How do we lower the skill floor, so even users who can't write well can publish something they love?
Solution 1 : Opportunity in AI
Let AI do what users couldn't
Write polished, brand-specific business communication by pulling from the user's website, Instagram, and Google Maps, AI generated a caller tune that named the business, hinted at its personality, and sounded professional.
Setup your professional business caller tune
30+ festival greetings
Special announcements
Open hours tune
Closed hours tune
Select a voice personality
2
Enter business details for AI to analyse
3
Customize your AI caller tune
Make it live on your business number
CREATE NOW WITH AI
DO LATER
11:11
English
Hindi
Tamil
Telugu
Kannada
Label
Label
Label
Label
Aditi
Modern & Growing
Rahul
Premium & Aspirational
Priya
Fun & Youthful
Arjun
Simple & Helpful
Sara
Warm & Approachable
Maya
Simple & Helpful
Kabir
Fun & Youthful
How satisfied are you with the selection available here?
Select a personality
NEXT
11:11
About
Complete your business details
Tarun Dental Clinic, HSR Layout
Brand name
Healthcare
Category
www.tarundental.com
Website URL
www.instagram.com/tarundental
https://g.co/kgs/fBx9Fqe
Google Business Profile
⚠ It is illegal to represent a business which is not yours. Your account may get banned if found in violation of Superfone’s Fair Usage Policy.
CREATE WITH AI
11:11
Business info used for AI
Add specific details about your business
Such as business hours, sale, announcement, etc.
Write message in any language
Namaste! Thank you for calling Tarun Dental Clinic. India’s leading dental chain. We're excited to help you, so please hang tight for just a moment while we connect your call.
Priya
Fun & Youthful
💡 Change personality for a different sentiment
MAKE IT LIVE
Edit Caller Tune
OPEN HOURS
11:11
Priya
|
Fun & Youthful
ENGLISH
Business info used for AI
Exclusive caller tunes for you to impress your customers and boost business!
Option
1
00:40
Namaste! Thank you for calling Tarun Dental Clinic, HSR Layout. India’s leading dental chain. We're excited to help you, so please hang tight for just a moment while we connect your call.
Option
2
00:01
Namaste! Thank you for calling Tarun Dental Clinic, HSR Layout. India’s leading dental chain. We're excited to help you, so please hang tight for just a moment while we connect your call.
Select
Option
3
00:40
Namaste! Thank you for calling Tarun Dental Clinic, HSR Layout. India’s leading dental chain. We're excited to help you, so please hang tight for just a moment while we connect your call.
More ideas
Caller Tune
OPEN HOURS
11:11
The first draft itself became the benchmark.
A brand-specific, professional-sounding caller tune
Users no longer had to write from a blank textbox.
The risk
AI alone doesn’t fix passive acceptance
We made generating the caller tune easy, but shaping it isn't. This replaces one form of disengagement with another, just enhanced by technology.
But the moment they tried to make it theirs to add a detail, change a phrase, switch the language; the skill problem came back.
We got users past the blank page and offered a personal first draft, only to drop them back into the same challenge. Ultimately, their writing skills will determine if they complete the customisation, and most likely, they won’t.
[DIAGRAM: User journey before vs. after — emotional arc showing where engagement drops in the AI-generated-only flow, and where the opportunity lies for active shaping.]
Problem 2
How do we get users to engage with the AI's output without making them work too hard?
Users needed to feel pride in their caller tune by actually shaping it, rather than just selecting from a list or approving an AI suggestion.
Solution
Possible options
A
Conversational interface
Let the user describe what they want in a chat-style prompt box.
🚨 High cognitive effort, requires prompt skills our users don't have.
B
Free-text editing
Show the AI-generated caller tune in a plain textbox. Let users edit freely.
🚨 This is what the pre-AI feature already was. They'd either publish it untouched or abandon the edit.
C
Make editing itself AI-assisted
Give users the AI-generated draft, and bring AI back into the editing flow.
✅ The user stays in control of what the caller tune says. Accepts messy input and AI helps to refine it.
Creation of AI Assisted Communication Module (AICM)
A textbox and two AI-powered buttons. Accepts messy input in any language, any length, any combination.
Not ‘Edit’ but ‘Improve it’
This single button absorbs every imperfection the user might write. Awkward phrasing, length issues, mixed-language cleanup.
‘Fit content’
Over-limit characters highlight in red. One tap and AI compresses the message without changing meaning.
‘More ideas’
Limited variations based on the current text. Keeps the experience bounded, token costs predictable.
Voice and personality, separated from text
Voice selection sat next to the text, not inside it. Tone changes happened by switching voice, not by re-prompting.
What we skipped
Undo, history, version saving
We didn't save AI iterations. We didn't keep version history. If the user doesn't like the AI's changes, they just exit without saving. Nothing's lost because nothing was committed. This kept the editing flow feeling low-stakes and exploratory
Why conversational AI didn’t fit
Once assisted-editing became the default action, a conversational interface didn't fit anymore. Not because it's wrong as a pattern, but because it pushes the user into prompting.. which is a hard skill?
Reason
Why it mattered here
The pre-AI feature was already a textbox.
Adding AI shouldn't ask users to learn a new interaction. Same flow, made better.
Chat needs threads, history, regenerate, clarification turns.
None of that serves the job of "publish one good caller tune." Every extra UI element was friction.
Our prompts were carefully tuned.
Free-form prompts would invite outputs we hadn't designed for. Tone/personality lived in the voice selector instead.
Conversational AI requires prompt skills.
Our users had used AI for fun, never for work. Prompt craft was a step too far.
Final flow
---
Final flow. Show full protoype in GIF or video format
Outcome
From default to brand identity
The feature shipped in 3 weeks. Within the first month, the behavior shift was visible in the data and in user feedback. Users weren't just adopting the feature, they were coming back to it.
Before
After
Customisation rate
[72]% of users on default caller tune
[24]% of users on default caller tune
Increase in activity
[X,XXX] caller tunes published per month
[XX,XXX] caller tunes published per month
Active re-engagement
[X]% of users returned to the feature post-setup
[XX]% returned to refine, update, or republish
Customization triggers
None
Festivals, offers, business hours changes, holidays
52% said “Feels like part of my brand"
The win wasn't that AI did the work for the user. It was that AI made the user feel proud of the work they did.
Surface adoption
From a feature to a system
AICM was built for caller tunes. But the underlying user need — I know what I want to say, help me say it well — wasn't unique to this feature. It showed up across the product. So the module did too.
AI Receptionist greeting
The opening line when an AI agent picks up
Live in production.
AI Google Review replies
Public responses to customer reviews
Live in production.
AI Message templates
Pre-saved reusable messages
Not manifested.
What changed across surfaces
The prompt. The tone. The character budget. The audience. The language defaults.
What stayed constant
The textbox. Improve it. More ideas. The user's experience of meeting AI.
Why a module, not a feature
When the same interaction works across multiple surfaces, you're no longer designing a feature. You're designing an interaction model that is stable for users to meet AI inside this product.
That's where the value compounds. Every new surface that adopts AICM costs less to design. Users meet AI the same way every time. The product gets a coherent AI personality without each team reinventing the pattern.
Reflection
What this work taught me about designing AI for non-tech users
Many AI business tools today position AI as an assistant, which works for users familiar with it as a coworker. However, our users had only interacted with AI for fun, not considering its use for business or skill development.
To bridge this gap, I redefined AI within Superfone as added intelligence. As a booster that helps users complete tasks faster and better, without the need for conversation. Our users prefer their existing methods and are not looking for disruptive changes or new skills. Therefore, we needed to demonstrate that this AI enhancement was worthwhile.
The experience should not overwhelm users. They shouldn't have to navigate when or how to interact with AI, as each decision adds cognitive load. Instead, I designed AI to perform one specific task well, seamlessly integrated into familiar interfaces. The best AI design minimizes user decisions, providing support only when necessary, allowing users to focus on their work.
Credits
Role and team
Sole product designer. Owned end-to-end design from problem framing through final QA. Worked closely with the PM on product strategy and direction, influenced the reframing of the problem and the key design decisions.
Took an active role in QA, including adversarial testing of AI outputs, and partnered with engineering through implementation.
Team 1 PM · 1 QA · 1 frontend engineer · 1 backend engineer