evolve

Designing Explainable Money Advice for First-Time Earners: 85% Could Explain What Navi Recommended and Why

AI product design, Interaction design, Trust & safety, Prototyping

Confidence Lift

+30%

Confidence lift pre/post viewing explainability widgets.

Comprehension

85%

Could explain the AI's recommendation clearly.

Engagement rate

75%

Engaged with Explain or Source indicators before acting.

Concept validation

55%

Preferred segmented Insight pattern over baseline layouts.

Now in validation. The MVP is in ongoing user testing to assess market fit: iterating continuously on real feedback rather than assumptions.

01

Context

Who this is for, and where today's AI advice loses their trust.

Introduction

Goal

Set the scene: the product, my role and the timeline.

About Evolve Money

Evolve is a social finance app for young adults in India. It makes managing money feel less like a solo journey, helping first-time earners budget, save, invest, and learn from others while building confidence with money.

My Role

Founder & Product Designer. Led the end-to-end process solo - research, AI system spec, interaction design, voice, failure states, and prototyping.

Timeline

2 months · Students & early professionals, India · MVP in validation

Unpacking the Problem

Gaps in the AI Dialogue Box

Goal

Find exactly where the current AI overlay loses the user.

Closed state. Scroll to open the overlay.

Mapping Where User Confidence Breaks Down

16

Interviews conducted

128

Surveys conducted

6

Field Observations

H1

Recommendation Without Direction

“I know I should switch, but what do I actually do next?”
The advice showed what to do, but not how to act on it.

Hypothesis

If recommendations explain how to act, users will feel more confident taking the next step.

H2

Lost in Financial Jargons

“I don’t even know what half of this means.”
Financial jargon made the advice harder to understand and evaluate.

Hypothesis

If financial terms are explained in plain language, users will better understand the advice before deciding whether to act.

H3

Conflicting Advice

“The app says one thing, my friend says another. Who do I trust?”
Conflicting advice made users unsure what to trust.

Hypothesis

If AI explains the trade-offs and lets users question the advice, they can better decide what’s right for them.

H4

The Human Check

“Can I just talk to someone before I do this?”
Users wanted human reassurance before making higher-stakes decisions.

Hypothesis

If users can question and discuss advice with an accountable human expert, they’ll feel more confident making a decision.

H5

The AI Feels Distant

“It just feels like I’m talking to a bot.”
The interaction felt functional, but impersonal.

Hypothesis

Giving the AI a recognizable personality and presence could make conversations feel more natural and engaging.

Problem Statement

The Gap Between Interest and Trust

Goal

Turn the research into one problem worth solving.

The overlay above isn't broken by accident, it's what happens when an AI feature is built to sound helpful without being checkable. Users are willing to try AI for money decisions. What stops them is that trying isn't the same as trusting.

User Opportunity

Novice investors are open to AI for financial advice

Novice investors are open to AI advice. Human advisors feel costly and intimidating, making AI their only accessible option.

User friction

Personalisation Didn’t Make the Advice Trustworthy

Used real data but lacked transparency. Framed the switch as all upside with no fees, tradeoffs, timestamps, or reasoning behind the confidence score.

The Problem Statement

How might we design the presentation of AI-driven financial recommendations so that novice investors can understand the reasoning behind the advice, evaluate trade-offs with confidence, and take appropriate action without needing to blindly trust the system

02

Principles

The rules every later design decision is checked against.

Design Principles

Turning User Hesitation Into Design Direction

Goal

Turn user hesitations into rules for every design choice.

Each hypothesis revealed a different reason users hesitated to trust AI financial advice. I translated those gaps into four principles that guided every design decision that followed.

H1

Actionable next steps

Path over point

Every recommendation pairs a clear next step with the reasoning behind it.

H2

Lost in financial jargon

Plain language over jargon

Terms are explained in context so users can retell the advice confidently.

H3

Conflicting advice

Question the advice

Every recommendation pairs a clear next step with the reasoning behind it.

H4 + H5

Human trust gap

Keep a human reachable

For high-stakes decisions, a visible path to a human advisor closes the trust gap.

03

Ideation

Four design problems, each explored, tested and narrowed to one direction.

Content & Voice

Teaching the AI how to talk about money

Goal

Decide how the AI should talk about money.

Steps
  1. 01Three directions
  2. 02Testing them
  3. 03The final voice

The recommendation pulled from the user's real data. But the language flattened every nuance into a single confident line. No trade-offs. No uncertainty. Just a directive dressed as advice

Three directions

Three possible content directions

The next step was content and voice. I took the same Kotak recommendation and wrote three genuinely different versions. Not rewrites, but different assumptions about how financial advice should speak to someone making a real decision.

Direction 01

Evidence First

Led with reasoning before making the recommendation

Your savings could be earning ₹4,482 more each year

Your average HDFC balance is ₹62,000. At Kotak's current 7.1% rate, that balance could earn more than HDFC's 3.5%.

worth Considering:

Kotak Active Money

But

Check the minimum balance requirement before switching

View Calculation View RBI SOURCE

Voice:

Factual, transparent, restrained

Direction 02

Conversational-First

Make the AI sound like an advisor opening a conversation rather than delivering a verdict

There's an opportunity to earn more on your savings.

Based on the balance you usually keep in HDFC, Kotak ActiveMoney could earn your around ₹4,482 more per year.

But higher returns aren't the only thing that matters. Your minimum balance, fees, and how you use the account could change whether switching makes sense for you.

Want me to compare the trade offs?

Voice:

Warm and approachable, but not overly human

Direction 03

Trade-Off Upfront

Led with reasoning before making the recommendation

Switching to Kotak could earn you more interest, but only if the higher balance requirement works for you.

What will you gain

₹4,482 more / year
Higher interest on your current savings balance.

Why this came up

Your current HDFC balance + current bank interest rates.

View Calculation View RBI SOURCE

Voice:

Balanced & Neutral

Testing the Content Directions

The next step was content and voice. I took the same Kotak recommendation and wrote three genuinely different versions, not rewrites, but different assumptions about how financial advice should speak to someone making a real decision.

Method

10

Participants

Active savers and investors

1

Recommendation

HDFC → Kotak, ₹4,482/yr

3

Voices tested

Evidence, Conversational, Trade-Off

1 → 3

Exposure

One voice first, then all three ranked

We measured

Comprehension

“What is this telling you to do, and why?”

Trust

“How much do you trust this advice?”

Action readiness

“Would you act on this now?”

Tone

“Describe it as a person.”

Findings

Direction 01

Evidence First

Trusted, but didn't move people.

High trust, low action. Caveats made people pause.

Comprehension
Trust
Trade-off awareness
Feels human

Direction 02

Conversational-First

Felt human, but the point got lost.

Warm, but the soft opening buried the recommendation.

Comprehension
Trust
Trade-off awareness
Feels human

Direction 03

Trade-Off Upfront

Clearest, but least trusted.

Fastest to understand, but read like a spec sheet.

Comprehension
Trust
Trade-off awareness
Feels human

No direction won on everything: trust, warmth and clarity each came from a different voice.

Synthesis

From Trade-Off Upfront

Structure

Condition and trade-off stated upfront

+

From Conversational-First

Warmth

Carried by who's speaking, not the wording

=

Final voice

Mentor Voice

Clear advice from a warm, human guide

Structure and warmth are separable

They can sit on different layers.

Warmth in wording is fragile

"Want me to compare?" read as withholding.

Evidence First played it safe

Never failed, never fixed "AI feels distant".

Sources and trade-offs

Comes fromContributesWhy it survived
Trade-Off Upfront
Leads with the condition, not just the upsideHighest comprehension + trade-off awareness
Conversational-First
The warmth, the sense of being spoken toHighest "feels human". Answers the earlier "AI feels distant" finding
New
A persistent mascot carrying the warmth outside the copyRemoves the need to soften language to sound human
✕ Rejected

Evidence First's caveat-as-afterthought (reads as disclosure, not guidance); Conversational-First's soft opening (warmth shouldn't delay comprehension).

The final voice

Designing Direction 4: Mentor Tone and Voice

The formula Defined
What to say and whyStructure from Trade-Off Upfront, plus warmth from Conversational-First, kept on separate layers rather than blended.
Mentor Voice Formula
Give warmth a persistent presenceBuild the layer the formula never specified, so copy can stay direct and structural without softer wording to feel human.

Principles for the Mentor Voice

The mentor voice adapts to the moment: explaining, cautioning, guiding, or listening, so the advice stays factual while the experience stays warm.

Mascot

Giving the AI a face, without pretending it’s human

Goal

Give the AI a name and a face without faking a human.

Steps
  1. 01Naming
  2. 02Look and feel

The advice needed personality, but not at the cost of credibility. I explored a mascot that could bring warmth and continuity to the experience while keeping financial recommendations direct, neutral, and grounded in evidence.

Brainstorming Names

A human name for familiarity, or a role-based tag for clarity?

Direction 1

Human Identity

Direction 2 · Selected

Role Based Persona

Question

Could familiarity feel approachable?
Could the name explain the AI's job?

Names

AriKiranNoorIsha+14
FinDadSageWise MentorEvolve Guide+2

Strengths

WarmMemorableEmotional
Clear purposeNot a fake personFlexible

Risks

Cultural biasLock-inHard to scale
GenericSounds authoritativeLess emotional

Verdict

Warmer, but boxes the product in.
Clearer and can grow with the product.
Naming the guide

A role beat a human name. But "Evolve Guide" was clear and forgettable, so I looked for a name inside the role.

Look and feel

Look and feel For Navi, your Evolve Guide

I explored Navi as more than a mascot. He needed to feel like someone users could comfortably turn to when financial decisions felt confusing: approachable enough to ask, experienced enough to trust, and expressive enough to make AI guidance feel less transactional.

Face: approachable, not authoritative

Explored

AgeHairFacial hairFeatures

Chose

MoustacheGlassesSofter features
Four rough face sketches for Navi, with the first selected

Experienced, warm, and recognizable.

Body: a presence to match

Explored

BuildsClothingPosturesFormal → casual

Chose

Softer buildCasual layersEasy stance
Six full-body sketches for Navi, with the relaxed pose selected

Relaxed, grounded, and approachable.

Expressions: respond, don't just appear

Moments

ReassuranceCuriosityUncertaintyConcernExcitement

Result

12 expressionsOne recognizable face
Twelve expression sketches for Navi, labeled from Neutral to Worried

Final render: a reusable system

Poses signal

ListeningExplainingThinkingReassuringGuiding

Role

The guide, not the authority
Final Navi character in six standing poses

AI Confidence Cue

Exploring how we can give AI confidence meaning

Goal

Make the AI's confidence score mean something.

AI confidence was easy to show as a number, but harder to make actionable. I explored how to communicate it visually so users could quickly gauge certainty and decide when to trust or question a recommendation.

How Navi brings meaning to AI confidence

Rather than treating confidence as a percentage users had to interpret, I explored ways Navi could translate uncertainty into cues users could recognize at a glance, without implying that his expression itself was proof.

Direction 1: Expression

Navi's expression changes with certainty.

Navi at high confidence
High Confidence
Navi at medium confidence
Medium Confidence
Navi at low confidence
Low Confidence

Direction 2: Language

Navi changes how confidently he phrases guidance.

“This is strongly supported”

High Confidence

“This could work”

Medium Confidence

“I'd verify this first”

Low Confidence

Direction 3: Evidence Cue

Navi's posture and gestures shift with AI certainty.

Switch to Kotak for Better Returns

● Supported by 4 signals

Try a recurring SIP top-up

● Supported by 1 signal
Choosing a direction

I moved forward with language and evidence cues as the clearest way to communicate AI confidence. Changing Navi's expression for every confidence state added another visual signal to an interface that already contained recommendations, trade-offs, and supporting information.

Instead, confidence would be communicated through what Navi says and what evidence he points to, keeping the interface lighter while giving users something concrete to evaluate.

“This is strongly supported”

High Confidence
+

Switch to Kotak for Better Returns

● Supported by 4 signals

AI Insight Layout

Finding the right layer for AI Insight Overlay

Goal

Choose where AI insights should live in the app.

The provenance layer gave users a clear reason to trust or question each recommendation, turning AI output into something they could evaluate on their own terms.

available bank balance

₹34,284.96

3 new transactions made today

₹4,482

more interest per year

by switching your idle savings balance from HDFC (3.5%) to Kotak ActiveMoney (7.1%).

Source: Kotak.com Why am I seeing this?
92% AI confidence
Was this advice helpful?

KOTAK Mahindr...

2 3

₹17,985.00

02 nov

HDFC BANK

7 4

₹1,230.00

02 nov

HDFC BANK

3 8

₹19,251.96

02 nov

Direction 01 Inline Expansion

available bank balance

₹34,284.96

3 new transactions made today

This insight is strongly supported

Your savings could be earning ₹4,482 more each year

Average HDFC balance of ₹62,000 earns more at Kotak’s 7.1% than HDFC’s 3.5%.

But, check the minimum balance requirement before switching. Supported by 4 signals.

Why now?

Rates were checked today. The gap between 3.5% and 7.1% is large enough to matter on your average balance.

What’s the catch?

Kotak Active Money has a minimum balance requirement. Check that before you switch.

KOTAK Mahindr...

2 3

₹17,985.00

02 nov

HDFC BANK

7 4

₹1,230.00

02 nov

HDFC BANK

3 8

₹19,251.96

02 nov

Direction 02 Conversational

available bank balance

₹34,284.96

3 new transactions made today

KOTAK Mahindr...

2 3

₹17,985.00

02 nov

HDFC BANK

7 4

₹1,230.00

02 nov

HDFC BANK

3 8

₹19,251.96

02 nov

Direction 03 Modal Sheet

available bank balance

₹34,284.96

3 new transactions made today

KOTAK Mahindr...

2 3

₹17,985.00

02 nov

HDFC BANK

7 4

₹1,230.00

02 nov

₹4,482

more interest per year

Anchored to this HDFC balance. Switching idle savings to Kotak ActiveMoney (7.1%) from HDFC (3.5%) could earn more.

Source: Kotak.com 92% confidence

HDFC BANK

3 8

₹19,251.96

02 nov

Direction 04 Anchored to the data

Method & criteria

8

Participants

Each tried every design

4

Interaction models

Counterbalanced order

5

Tasks per model

Scored, not preferred

4

Scenarios

Equivalent, so no recall

We measuredQuestion askedSuccess =
Comprehension"Why is Evolve recommending this?"Explains the reason
Trade-off awareness"What would you gain / give up?"Names ≥1 gain and a compromise
Actionability"What would you do next?"States a next step
Discoverability"What else would you want to know?"Finds supporting info unassisted
Follow-up abilityRate confidence 1–5 and explainResolves an open question

Comparing four directions (n = 8)

No single layout won everything; each solved a different part of the decision.

ComprehensionTrade-offsActionabilityDiscoverabilityFollow-upTotal
Inline Expansion7/87/87/87/83/878%
Conversational7/86/86/88/88/883%
Modal sheet8/88/87/86/83/880%
Anchored to the data4/83/84/82/82/838%

Scores are participants (out of 8) who completed the task. Total bars show success across all 40 tasks.

Final Selection

Based on our evaluation, the conversational direction emerged as the strongest approach. It uniquely enables users to resolve doubts in context, builds trust through upfront confidence signals, and keeps expert consultation always accessible.

Direction 02 Conversational
  • Conversational direction is the only direction that lets a real doubt get resolved in place, the exact gap the other three left open
  • Modal sheet and inline expansion scored close behind but capped at 3/8 on follow-up. Users with questions had nowhere to go.

Evolve Guide: Navi

This insight is strongly supported

Your savings could be earning ₹4,482 more each year

Average HDFC balance of ₹62,000 earns more at Kotak’s 7.1% than HDFC’s 3.5%.

But, check the minimum balance requirement before switching. Supported by 4 signals.

What if I don’t switch

You’d continue earning 3.5% at HDFC, missing out on roughly ₹4,482 per year.

Rates were checked today. The gap between 3.5% and 7.1% is large enough to matter on your average balance.

Kotak Active Money has a minimum balance requirement. Check that before you switch.

AI-generated insight. Verify independently before acting.

Confidence Stated Upfront
Case stated upfront with minimal to
no financial jargons with trade off
Doubts stated upfront
Possible doubts and concern stated upfront
Option to consult an human alway available upfront

Supporting Components

Supporting components extend the conversational direction by helping users navigate multiple insights, understand confidence at a glance, and read Navi’s intent through a clear expression system.

01: Showing more than one insight without stacking dialog boxes

A compact insight stack surfaces additional recommendations without interrupting the conversation with overlapping dialogs.

02: Confidence Spectrum

A clear confidence spectrum shows how strongly each insight is supported, helping users decide when to act and when to verify.

This insight is strongly supported

Your savings could be earning ₹4,482 more each year

Average HDFC balance of ₹62,000 earns more at Kotak’s 7.1% than HDFC’s 3.5%.

But, check the minimum balance requirement before switching. Supported by 4 signals.

You’d continue earning 3.5% at HDFC, missing out on roughly ₹4,482 per year.

This could work

Your savings could be earning ₹4,482 more each year

Average HDFC balance of ₹62,000 earns more at Kotak’s 7.1% than HDFC’s 3.5%.

But, check the minimum balance requirement before switching. Supported by 4 signals.

Worth checking the catch first. The upside is real, but the minimum balance may not fit every month.

I’d verify this first

Your savings could be earning ₹4,482 more each year

Average HDFC balance of ₹62,000 earns more at Kotak’s 7.1% than HDFC’s 3.5%.

But, check the minimum balance requirement before switching. Supported by 4 signals.

I’d treat this as a lead, not a decision. Verify the rate and balance rules before moving money.

03: Navi’s expression layer

Navi’s expression layer mirrors the tone of each moment, making the assistant feel more responsive while helping users interpret its intent at a glance.

Navi expression set: Neutral, Thinking, Curious, Concerned, Skeptical, Excited, and Worried.
04

User Testing

Checking whether the assembled experience actually earns trust.

Validation

Validating the Assembled Experience

Goal

Test whether the assembled experience earns trust.

Voice and layout were tested separately, but neither revealed how Mentor tone, Navi, the confidence spectrum, and the insight stack worked together. This round tested the complete experience with a new cohort, navigating the flow independently without facilitator guidance.

Method & criteria

20

Participants

First-time earners, India

1

Full prototype

Tested end to end

5

Tasks

Unmoderated, then a survey

0

Prior exposure

None had seen earlier versions

We measuredQuestion askedSuccess =
Comprehension"Why is Evolve recommending this?"Explains the reason
Trade-off awareness"What would you gain / give up?"Names ≥1 gain and a compromise
Actionability"What would you do next?"States a next step
Discoverability"What else would you want to know?"Finds supporting info unassisted
Follow-up abilityRate confidence 1–5 and explainResolves an open question

Findings

“It’s not telling me to do something, it’s telling me why I might want to.”

Participant, Full system Validation

85%

Comprehension

17 of 20 participants (85%) could independently explain what Evolve recommended and why, without prompting.

+30%

Confidence lift

Average confidence rose from 2.6 to 3.4 after participants explored an explainability widget, showing how verification helped build trust.

75%

Engagement with provenance

15 of 20 participants opened Explain or View Source before deciding, showing they wanted evidence rather than relying on Navi’s tone alone.

55%

Concept validation

11 of 20 preferred segmented insights, leaving its effectiveness an open question.

Synthesis

The problem statement asked for advice users could evaluate, not just receive, and the numbers here answer that directly.

Comprehension and engagement moved together: people weren’t just understanding the recommendation, they were checking it before acting on it, which is the behavior the whole provenance layer was built to produce.

The 55/45 split on layout is the honest exception. It’s a reminder that validating the ideal path isn’t the same as validating the edges, which is exactly what the next section takes on.

05

System

Designing for when the AI is wrong, unsure or out of scope.

Failure States

Failure states are the job

Goal

Design for when the AI is wrong, unsure or off-limits.

Trust isn’t built on the ideal path alone. In finance, the experience must remain clear and responsible when data is stale, confidence is low, experts disagree, or a request falls outside what we can legally provide.

Stale data

No transfer prompt

If rates are older than 7 days, show balances but don't recommend a switch.

Rates last checked 12 days ago. Refresh before moving money. Comparison hidden until then.

Low confidence

Say the difference is small

A 0.2% gap isn't a “boost your savings” headline. No dramatizing a rounding error.

This difference is small. Transfer friction may eat the gain. Not recommending a move.

Refuse

Off-policy questions

Stock picks, insurance, loans, crypto: hand off to Learn or a human expert. Never improvise.

I can’t advise on that. I can show a beginner explainer, or you can book a verified expert.

Conflict

Experts disagree

If sources conflict (3 vs 6 months’ emergency fund), show both. No fake certainty.

Two experts disagree on emergency-fund size. Here’s both. Your call.

Correction

“That rate looks wrong”

A user flag hides the suggestion, logs it, and pauses that rate until a human checks it.

Thanks, we hid this suggestion. We won’t use that rate until it’s checked.

Compliance

Not a footer

Disclosures sit on the action itself: not advice, which data was used, and that Evolve doesn’t earn from it.

Not financial advice. Using your linked HDFC & Kotak balances. Evolve does not earn if you switch.

06

Reflection

What I'd keep, and what I'd do differently next time.

Closing the Loop

What project taught me

Goal

Lessons I'm taking into the next project.

01

MEASURE

Comprehension, trust, and action are three different metrics

A design can win one and lose the other two. Conversational won not because it scored highest everywhere, but because it was the only direction that let a real doubt get resolved without the user leaving the screen.

02

EMBODY

Voice and interface aren't separate decisions

The mentor tone only worked once it had somewhere to live. Navi's persistent presence carried the warmth, not the copy itself, which stayed direct and factual throughout.

03

VALIDATE

The human handoff is designed, not yet proven

It assumes advisors respond quickly and are available on demand. With more time, I’d validate that against real advisor capacity before treating “always reachable” as a solved promise.

04

SEQUENCE

Failure states came late, and it showed

Working solo across research, AI spec, voice, and interaction in two months meant the failure-state work happened after the ideal path was already designed. Sequencing it earlier next time would change how the ideal-path screens are built, not just what sits behind them.

Made this far?

Let’s make the next thing matter.

If this way of turning complex product questions into clear, useful experiences resonates, let’s talk.