Varsha Lakshmanan: Senior Product Designer

Project 01 of 04

Project 01 · Prosperr.io · AI-assisted tax experience

Designing Trust in AI-Assisted ITR Filing

Balancing AI automation with user trust in high-stakes tax filing. Reduced errors 38% while improving completion by 24%.

Role: Design Manager, B2C
Team: 14-person cross-functional
Duration: 12 weeks
Ownership: AI UX strategy, research, design

+24%Completion rateFirst filing cycle vs. manual baseline

↓38%Fewer filing errorsAI-assisted review vs. manual entry

↓31%Support queriesMilestone clarity reduced queries

4.3/5Confidence ratingPost-filing confidence survey

Measured 2–3 weeks post-launch across live filing workflows.

The short version

1Problem

45% of first-time filers abandoned manual filing because of error anxiety, form complexity and distrust of AI.

2Insight

Users accepted AI only when they could verify its reasoning and keep control over final decisions.

3What I did

Led an explainable AI strategy for a 14-person team: users review AI suggestions with sources, confidence scores and edit controls, instead of filling blank forms.

4Result

+24% completion and 38% fewer errors. The trust principles became company-wide standards for all AI features.

ContextBusiness problem

Business problem and constraints

User segment: First-time tax filers overwhelmed by compliance terminology. 45% abandoned manual filing due to error anxiety and form complexity.
Job to be done: File accurately without legal risk, minimize cognitive load, and maintain control over financial data.
Key constraint: AI accuracy was 85–92%. I couldn't hide uncertainty, so I had to design for it. Regulatory compliance required user verification for all automated suggestions.

ResearchWhat we learned

Insights that shaped decisions

Users feared costly mistakes

Compliance risk outweighed convenience. Trust required transparency, not just speed.

68%

“What do I do next?” dominated support

68% of tickets showed navigation confusion. Users needed progress visibility and clear next steps.

28%

Manual entry created high error rates

28% of submissions contained errors. Users wanted help but didn't trust “black box” automation.

AI explainability was non-negotiable

Users accepted AI only when they could verify its reasoning. Control over convenience.

Key findingUsers accepted AI assistance only when they could verify reasoning and maintain control over final decisions.

What we tried firstAnd why it failed

“Magic” auto-fill failed user testing

First attempt

Fill everything automatically

Early flows were “magic” auto-fill: the system filled in tax details for users, with little visibility into where the numbers came from.

What testing showed

Users didn't trust what they couldn't check

Those flows failed user testing. Users feared costly mistakes and accepted AI only when they could verify its reasoning and keep control of the final decision.

What changed

Review, don't fill

I redesigned around transparency: every suggestion shows its source, a confidence score and edit controls. Beta testing confirmed users valued a self-aware system.

ApproachHow it grew

Three decision phases

Mapped trust gaps

Identified where automation eroded confidence, which shaped the decision to prioritize transparency over speed.

Why it mattered: without trust insights, we'd have optimized for the wrong metric.

Established design principles

Made explainability and user control non-negotiable, and defined confidence thresholds that balance automation with verification.

Why it mattered: it created an alignment framework that prevented endless debate on every AI interaction.

Validated transparency patterns

Tested AI explanations to prove that showing reasoning increased trust. Beta confirmed users valued self-aware systems.

Why it mattered: it validated that “magic” automation would fail, saving months of wrong direction.

Try itTiered confidence

Designing for uncertainty, not hiding it

AI accuracy ranged 85–92%, so confidence thresholds decided what was auto-filled and what needed review. Drag the slider to see how a suggestion behaves.

AI confidence for a suggestion90%

80%85%90%95%100%

80–95%: you review95%+: auto-acceptedTypical AI accuracy, 85–92%

✎

RecoveryWhen the AI is wrong

Making a wrong suggestion easy to spot and fix

At 85 to 92% accuracy, some suggestions will be wrong. The design makes that visible and recoverable instead of hiding it. Try it: check the source, then fix the value.

Suggested by AI · House rent allowance exemption₹1,20,000

82% confidence · needs your review

An illustration of the recovery pattern, not a real user's data.

Key decisionsTrade-offs I made

Critical choices

Milestone-driven journey (system over screen)

Broke filing into 6 milestones with progress visibility, accepting more screens for clarity instead of fewer screens for speed.

Chose confidenceover efficiency

↓31% support queries. Framework reused across products.

Shift from entry to review (intentional friction)

Users review AI suggestions instead of blank forms. Slower interactions instead of instant automation, to build trust.

Chose verificationover speed

↓38% errors. Repositioned the product as an intelligent assistant, not an automation tool.

AI explainability as the foundation

Every suggestion includes source documents, confidence scores and edit controls. Hiding complexity for simplicity was considered and rejected, because users needed control.

Chose controlover hidden simplicity

4.3/5 confidence rating. Became the company-wide AI standard.

Tiered confidence scoring

95%+ confidence auto-accepted; 80–95% required review. Constraint-first thinking: 85–92% accuracy meant designing for uncertainty.

Chose honesty about uncertaintyover hiding it

+24% completion while maintaining accuracy. Self-aware systems increased trust.

ExecutionThe experience

Strategic tax journey

Turning a fragmented process into a milestone-driven experience.

Tax journey

Milestone-driven guidance

✓Tax planningActiveUnderstanding your tax obligations
✓Upload salary slipDocuments verified
•••Book consultationSchedule with tax advisor
4Review declarationUpcoming
🔒File tax returnLocked

Milestone-driven guidance reduces uncertainty across the tax lifecycle.

AI-assisted decision

Smart data import · tap to compare

✓ Read-only access. No credentials stored; data fetched securely.✓ AI-assisted validation: intelligent error checking and suggestions.

System status

Clear feedback

✓Salary slip uploadedForm 16 verified

✓ITD data fetchedIncome details verified

✓Consultation scheduledDec 23, 2025 at 2:00 PM

→Next step: review declaration with advisorYour tax advisor will review all submitted documents before filing.

Clear system feedback reduces post-action anxiety.

InterfaceScreens

Wireframe of the ITR pre-fill step, with a three-step progress rail, data fetching state, FAQs and chat — Loading state showing AI fetching personal, income, and tax information from ITD with progress indicators.

File ITR screen showing completed details, document verification in review, and upcoming computation and filing steps — User verification workflow showing source documents linked to AI suggestions.

Dashboard with ITR filed status, filing progress, tax advisor support and financial overview — Completion dashboard showing filing progress, tax advisor support, and financial overview.

AI strategyBeyond this feature

How AI changed product direction

Explainability as non-negotiable

Early “magic” auto-fill flows failed user testing. I redesigned around transparency: every suggestion with source references, confidence scores and edit controls. That shifted the product from an automation tool to an intelligent assistant.

Human-in-the-loop safeguards

Confidence thresholds determined auto-fill versus review. Users trusted the system's self-awareness, which paradoxically increased confidence.

Responsible automation principles

Principles balancing speed with verification became company-wide standards for all AI features.

Never auto-submitAlways show the sourceMake overrides penalty-freeEducate, don't just alert

ReflectionWhat I carry forward

Leadership principles

Trust-first AI design

In high-stakes contexts, transparency matters more than speed. Explainability isn't a feature; it's a foundation for adoption.

Systems over screens

The milestone framework became reusable infrastructure. Systems thinking creates leverage that scales across products.

Alignment drives outcomes

Design leadership is aligning 14 people around trade-offs, constraints, and shared success metrics, not just making interfaces.

Intentional automation

AI doesn't make products ethical by default. Designing safeguards and override paths required deliberate choices that became company standards.

Project 02 of 04

Project 02 · Prosperr.io · Design systems

One Design System, Mapped for Every Team and Any AI Model

A single design system that anyone in the organisation can map to the AI tool they use, so designers, PMs and engineers all ship consistent, on-brand UI without waiting on a designer.

Role: Product Design Manager
Team: Design, Product, Engineering
Duration: 6 months
Ownership: System architecture, AI mapping, governance

3×Faster prototypingPMs built mockups without designer dependency

↓42%Design reworkFewer off-brand outputs

89%Non-designer adoptionPMs and engineers using the system weekly

18Teams using the system45+ users across the product org

Measured 8 weeks post-rollout across 18 product teams.

The short version

1Problem

Designers had become bottlenecks. Every mockup, variant or iteration waited on a designer, slowing product velocity.

2Insight

A component library doesn't teach intent, and AI tools without guardrails produced fast but off-brand UI.

3What I did

Built one system in three layers (tokens, components and AI mapping templates) that anyone can plug into the AI model they use.

4Result

3× faster prototyping, 42% less rework and 89% non-designer adoption across 18 teams. Designers moved from gatekeepers to enablers.

ProblemWhy it mattered

Designers as bottlenecks, not multipliers

Core issue: product managers and engineers couldn't move without designer approval. Every mockup, variant or iteration required designer involvement, creating queues and slowing product velocity. It wasn't just a UI consistency problem; it was a cross-team collaboration breakdown, where designers became execution blockers instead of strategic multipliers.

What we observed

PMs built prototypes in slides because Figma felt inaccessible.
Engineers made UI decisions in code, creating drift from design intent.
Designers spent 60% of their time on repetitive component variations instead of strategy.
AI tools (ChatGPT, Midjourney) generated inconsistent outputs with no system guardrails.

Strategic betTurn the design system into a shared product for every discipline, with a mapping layer that guides non-designers to correct outputs without blocking their velocity or needing 1:1 designer oversight.

My roleDecisions I owned

System architecture and cross-team enablement

Owned the system architecture, component strategy and governance model connecting design, product and engineering, through structured AI-guided workflows that balanced automation with creative control.

Decision authority: Mapping structure, component rules, system governance
Alignment: Design, Engineering, Product, Marketing
Scope: 18 teams, 45+ users, 6 months

UsersOne system, three ways in

Three user types, one system

Pick a role to see what the same system gives each team.

Job

Need

What the system gave them

Before and afterA real handoff

From two hours of back-and-forth to fifteen minutes

These metrics represent a shift in how teams collaborate: designers evolved from gatekeepers to enablers, PMs gained design autonomy without sacrificing quality, and engineers contributed to the system instead of working around it.

Before

PMWrites a request✕ Ambiguous requirements

5+ messages

DESDesigner clarifies✕ Constant clarifications

3+ messages

ENGEngineer builds✕ Inconsistent results

~2 hrsper handoff

After · mapped system

PMUses a structured template✓ Clear intent from the start

AI applies the system rules

DESWeekly quality review✓ Not blocking

Direct to implementation

ENGEngineer extends components✓ Predictable, on-brand output

~15 minper handoff88% faster

Real team workflow: PM prototypes with templates, engineer extends components, designer reviews weekly.

StrategyWhy this approach

Why a mapped system, not just a component library

Components alone don't teach intent

A button library doesn't tell you when to use primary versus secondary. PMs and engineers need decision-making guidance, not just assets. The mapping layer encodes design thinking (when to emphasize hierarchy, how to balance density, which patterns match user intent), turning components into a teaching layer.

AI tools need constraints, not freedom

ChatGPT and Figma AI plugins generate UI fast, but inconsistently. Mapping structured templates to our component vocabulary constrained AI outputs to match brand rules, typography scales and spacing tokens automatically. Speed without drift.

The system as a shared product, not a designer tool

Treated as infrastructure for all teams, not a design artifact. PMs got templates for common flows (onboarding, dashboards, settings). Engineers got extensible component APIs. Designers focused on governance and edge cases, not every screen.

ArchitectureHow it maps

One system, plugged into any AI model

Three layers sit between the brand and the output. Because the rules live in the system rather than in one tool, anyone can take the same system and map it to the AI model they already use.

TokensColors, spacing, type

ComponentsButtons, cards, inputs

AI mappingTemplates and rules

OutputsOn-brand UI screens

Works with the AI tools teams already useChatGPTFigma AIAny other model

Layer 1

Token foundation

Semantic tokens for color (primary, secondary, danger), spacing (compact, comfortable, spacious) and type scales. Non-designers reference tokens, not hex values, ensuring consistency across AI-generated and hand-coded UI.

Layer 2

Component library

Reusable patterns mapped to jobs: CTAs for conversion, cards for content hierarchy, modals for decisions. Each component includes usage guidance, accessibility defaults and responsive behavior, eliminating guesswork.

Layer 3

AI mapping templates

Structured templates for common scenarios. “Create a dashboard for finance users prioritizing data density” automatically suggests grid layouts, compact spacing and neutral colors. PMs and engineers get design thinking without designer involvement.

The systemIn practice

Design system overview: color system, spacing, radius, CTA hierarchy, typography, icon system, and sample B2C, B2B and admin surfaces — One foundation across the individual filer (B2C), business portal (B2B) and tax professional dashboard (admin).

Implementation guide for the default text input, with typography, colors, spacing, focus state and Tailwind classes — Every component ships with plain-language guidance and exact values, so developers, PMs and AI tools read the same rules.

Key decisionsTrade-offs I made

Speed vs. precision, automation vs. control

Every design system balances flexibility and constraint. These decisions defined how we preserved quality while enabling non-designer autonomy.

Constrained templates over open-ended AI

Open AI requests generate unpredictable results: beautiful but off-brand. We chose structured templates that map user goals (e.g. “high-urgency action”) to specific components and tokens. This reduced creative expression but eliminated drift, crucial for trust in a financial product.

Chose consistencyover creative freedom

Self-service enabled, quality reviewed

We let PMs and engineers ship without design approval, accepting that 10–15% of outputs would need refinement. Designers reviewed weekly, not per ticket, shifting from gatekeepers to curators.

Chose speedover occasional inconsistency

3× faster iteration cycles while maintaining an 85%+ quality threshold.

Refinement through usage data

Tracked which templates led to rework versus approval. High-failure requests (e.g. “modern dashboard”) were refined with additional constraints (“data-dense dashboard with neutral colors”).

Chose real usageover assumptions

ConstraintsDesign boundaries

What the system had to work within

Variable AI quality

ChatGPT and Figma AI outputs were inconsistent. We needed guardrails to constrain AI without blocking velocity.

Non-designer skill gaps

PMs and engineers lacked design fundamentals (hierarchy, spacing, accessibility). The system had to teach, not assume knowledge.

Startup velocity pressure

We couldn't slow product teams for designer approval. The system had to scale without adding headcount.

Multi-surface consistency

Web, mobile and internal tools needed a unified brand without custom solutions per platform.

ImpactBeyond the numbers

What changed for each group

Product

Faster hypothesis testing
Consistent multi-platform UX
Reduced design debt
Higher iteration velocity

Designers

60% less repetitive work
Focus on strategy, not execution
Governance without bottlenecks
Scaled without headcount growth

Organisation

Cross-functional autonomy
Unified design language
Shared vocabulary across teams
Scalable quality standards

ReflectionWhat I carry forward

System thinking, not screen thinking

This system redefined design as enablement infrastructure, not output production. Designers aren't bottlenecks when non-designers have the right guardrails, and the mapping layer turned design thinking into a repeatable process, accessible to all teams.

What worked

Templates as a teaching layer, not just an asset library
Governance shifted from pre-approval to quality curation
AI constraints reduced drift without blocking velocity
Cross-functional ownership, not a designer-only tool

What I'd evolve

Automated quality scoring for AI-generated outputs
Real-time feedback on template usage patterns
A component contribution model for engineers
Integration with code generation tools (Cursor, v0)

Design systems fail when they're designer tools, not team infrastructure. This one succeeded because it empowered everyone: designers governed, PMs prototyped, engineers extended.

Project 03 of 04

Project 03 · Prosperr.io · B2B automation

Carryforward Automation

Reducing repetitive work through transparent automation in corporate tax workflows, without compromising verification or auditability.

Role: Product Design Manager
Team: 8-person cross-functional
Users: CSRs and employees
Rollout: Pilot with 5 CSRs, then the full team of 45

↑18%Task completionAutomation handled the predictable transfers

↓32%Filing errorsVerification caught variance in manual transfers

↓26%Support queriesException queues gave CSRs clarity

4.2/5CSR confidenceAfter automated transfers, thanks to audit trails

Measured 2–3 weeks post-launch across live CSR workflows.

The short version

1Problem

Manual tax data transfers consumed 40% of CSR capacity in peak season: hundreds of entities, identical workflows, high error risk.

2Insight

CSRs rejected “black box” automation. They needed to see exactly what was auto-applied and what was flagged.

3What I did

Designed a three-tier verification system with exception queues and full audit trails, plus an employee app view that explains every automated deduction.

4Result

↑18% task completion and ↓32% filing errors, with 40% of peak-season CSR capacity reclaimed.

ContextWhy this mattered

40% of the team's peak season, spent copying data

Manual tax data transfers consumed 40% of CSR capacity during peak season: hundreds of entities, identical workflows and high error risk. The constraint was that automation couldn't compromise verification standards or auditability.

Before

Manual transfers · 40%Everything else

After

Exceptions and clientsEverything else

Peak-season CSR capacity. The reclaimed 40% moved to exception handling and client relationships.

ResearchConstraints

Constraints that shaped the strategy

40%

CSR capacity bottleneck

40% of time went to repetitive transfers. Automation needed to free capacity for exception handling.

Verification cannot be compromised

CSRs rejected “black box” automation and needed full visibility into what was auto-applied versus flagged.

Year-over-year variance is the edge case

Changed values required manual review. The system had to surface exceptions, not auto-apply blindly.

Compliance requires audit trails

Every automated decision had to be traceable, with override controls essential for regulatory confidence.

Options weighedWhy three tiers

Two obvious answers, and why neither worked

There were two simpler ways to solve this. Each failed one of the constraints, which is how the three-tier system took shape.

Option A · Auto-apply everything

Fast, but a black box

Copying every value forward would save the most time. But CSRs rejected automation they couldn't see into, values that changed year over year would slip through, and compliance needs every decision traceable.

Option B · Review every record

Safe, but no time saved

Keeping a CSR check on every entity protects accuracy, but that manual transfer work was already eating 40% of CSR capacity in peak season. It solves nothing.

What I designed · Three tiers

Automate the predictable, surface the rest

Unchanged data is auto-verified, small variance gets a quick review, and anything over 2% or new is flagged and never auto-submitted. Every decision is logged with an override.

ApproachStrategy

Automation as augmentation, not replacement

Frame automation as augmentation

Positioned the system to handle predictable transfers, freeing CSRs for exception handling.

Trust over speed.

Build transparency into the system

Showed CSRs exactly what was auto-applied versus flagged, and made overrides penalty-free so they kept control.

Design for exceptions first

The exception queue prioritizes edge cases, and confidence thresholds surface uncertainty rather than hiding it.

Try itThree-tier verification

How a record gets sorted

Every entity's data is compared with the previous year. Move the slider to change how much a value differs, or mark it as a new entity, and see which tier it lands in.

Change from previous year0%

0%2%5%10%

New entity (no previous-year data)

Auto-verifiedUnchanged and confidentData matches the previous year with 99%+ confidence.No action required

Needs reviewPotential changesMinor discrepancies detected that need verification.Quick review

Manual entryNew or changedVariance above 2%, or a new entity. Flagged for review and never auto-submitted.Full review

An illustration of the rules, not real client data.

Key decisionsWhy they worked

Critical choices

Three-tier verification system

Auto-verified (99%+ match), needs review (variance detected), manual required (new or changed). Clear status at a glance.

Why it worked: it focused CSR attention on exceptions, not predictable transfers.

Exception queues over auto-apply

When year-over-year variance exceeded 2%, the record was flagged for manual review. Never auto-submitted.

Chose flaggingover auto-applying

Why it worked: it built trust by showing that the system knows its limitations.

Audit trails as the foundation

Every automated decision is logged with confidence scores, source data and override controls.

Why it worked: CSRs kept compliance confidence and regulatory accountability.

Transparency and controlEvery automated decision is explainable. CSRs can view the comparison logic, override automation when needed, and keep full audit trails for compliance.

InterfaceEmployee app

Making automated deductions legible

Designed to reduce employee confusion and increase confidence in automated tax adjustments.

History screen listing claims with Approved, Rejected, Partially Approved and Carry Forward tags, and claimed, approved and deducted amounts — Status breakdown made automation legible.

Carryforward breakdown

Status-coded breakdown made automated deductions legible at a glance.
Reduced “Where did my money go?” support queries by making automation explainable.

Success confirmation

A clear confirmation state reinforced trust in automated outcomes.
Lowered post-submission anxiety by explicitly showing what changed and why.

Claim details screen explaining that part of the amount will be adjusted next month, with deduction progress and monthly breakdown — Confirmation reduced post-task anxiety.

RolloutPilot to full team

Validated small, then scaled

Pilot5 CSRs across 50 entities, to validate confidence thresholds and exception handling.
RefineAutomation rules refined based on the edge cases the pilot surfaced.
ScaleDeployed to the full team of 45 CSRs.

Operational leverageReclaimed 40% of CSR capacity during peak tax season, shifting the team from repetitive transfers to exception handling and client relationships, with more predictable workflows for enterprise accounts.

ReflectionWhat I carry forward

Leadership principles

Transparency drives trust

CSRs adopted automation when they could verify outcomes. Audit trails preserved verification standards while reducing manual effort.

Design for edge cases first

Exception queues determined adoption success. Clear variance flagging built CSR confidence without compliance risk.

Augmentation over replacement

Freed CSRs from predictable work so they could focus on exceptions and client relationships. Automation positioned as a capacity enabler.

Systems create organisational scale

Reclaiming 40% of peak capacity created measurable team leverage. Systems thinking compounds across quarters, not features.

Project 04 of 04

Project 04 · Lollypop Design Studio · Enterprise platform

End-to-End Performance Management Platform

One platform that unified reviews, normalization and compensation for 2,500+ employees across 4 campuses, cutting manual HR effort by 45% while building governance into the system itself.

Role: Senior UX Designer, Lollypop Design Studio
Team: PM, 4 engineers, design
Duration: 16 weeks
Ownership: End-to-end design across every workflow

↓45%Manual HR effortEvent-driven automation removed coordination overhead

↓38%HR-driven follow-upsStructured workflows replaced manual status tracking

↓52%Appraisal setup timeConfiguration-first design sped up cycle kickoff

4Campuses scaledMulti-tenant architecture enabled org-wide adoption

The short version

1Problem

Reviews for 2,500+ employees ran on spreadsheets, email and inconsistent criteria, creating compliance risk and eroding trust.

2Insight

Ratings drive pay, tenure and promotions. People trusted the process only when decisions were visible and traceable.

3What I did

Designed configuration-first setup, guided reviews, normalization as decision support, event-driven reminders and structured escalation.

4Result

↓45% manual HR effort, ↓38% follow-ups and ↓52% setup time, scaled across 4 campuses.

ContextWhy it mattered

High stakes, held together by spreadsheets

Scale: 2,500+ employees across 4 campuses, each with different HR practices and approval hierarchies.
Stakes: Performance ratings tied directly to compensation, tenure decisions and promotions, so they needed to be auditable and consistent.
Breakdown: Workflows lived in spreadsheets and email. Manual follow-ups led to missed deadlines, ratings varied widely by campus and manager, and exceptions were handled outside the system.

My Reportees dashboard showing employee review status, campus and observation progress for a year-end review — One dashboard with end-to-end workflow visibility for every reportee.

ResearchWhat we learned

From what we heard to what we built

We learned

Managers defaulted to subjective ratings without structured criteria

So I designed

Predefined criteria with mandatory justification fields, reducing cross-campus variation while preserving flexibility.

We learned

Employees distrusted decisions made behind closed doors

So I designed

Real-time visibility into normalization and compensation logic, making fairness transparent without exposing individual data.

We learned

HR spent 40% of its time on manual follow-ups and status tracking

So I designed

Event-driven notifications and role-based workflows, shifting coordination from manual to automated.

We learned

Exceptions bypassed the system, leaving no audit trail

So I designed

Structured escalation paths with approval chains and justification requirements, bringing governance inside the system.

ConfigurationSet up once, right

Configuration-first design

Design decision: treat configuration as a first-class product experience, not an admin screen. Letting HR define expectations upfront cut mid-cycle confusion and manual corrections by 52%.

Cycle-specific weightages for academic and non-academic structures
Role and campus scoping to match the organisational hierarchy
Version control for year-over-year comparison

Academic

KRA 50%Comp. 20%TD 20%

Non-academic

KRA 70%Comp. 20%

Same platform, different structures. Each campus and role gets the weightage that fits it.

Appraisal configuration screen with academic and non-academic weightages, appraisal cycle details and KRA section settings — Flexible configuration for appraisal cycles, weightages and rating scales.

ReviewsTwo audiences

Guided reviews with accountability

Employees: clarity and confidence

Clear section weightages and inline guidance
Real-time status visibility

Managers: flexibility with guardrails

Structured feedback against predefined criteria
Promotion recommendations with mandatory justification

Try itNormalization

Decision support, not automation

Ratings varied across campuses, so I designed normalization as decision support: it surfaces fairness gaps and cost impact, and leadership makes the final call. Drag the slider to see a campus that rates more generously than the ideal curve.

How generously this campus ratesBalanced

Ideal distributionThis campus

Distribution gap0%

Cost impact preview+0.0%

What happensNothing to flag

An illustration of the pattern, not real employee data.

Campus normalization screen with cost summary and a bell curve report comparing ideal, actual and normalized curves — Campus-wise rating distribution with actual vs. ideal bell curves, and the cost impact alongside.

AutomationCoordination

Event-driven communication

Design decision: shift coordination from manual to event-driven to remove follow-up overhead.

3 days before dueEmployee, manager and HR

1 day before dueEmployee, manager and HR

On the due dateEmployee, manager and HR

Still pending afterEscalates to HR and manager

Role-targeted notifications reduce noise
Editable templates preserve the organisation's tone
Workflow context is embedded in every notification

Create new notification screen with trigger conditions for goal setting completion — Configuring event-based notifications and role targeting.

GovernanceExceptions inside

Structured escalation

Design decision: bring exceptions into the system instead of letting them bypass governance. Rating disputes and overrides follow structured paths with mandatory approvals, justification and a full audit trail.

Goal setting escalation screen with trigger conditions and notification time period — Escalation triggers configured, with who gets notified at each stage.

Key decisionsTrade-offs I made

Key decisions and trade-offs

Normalization: configurable, not automatic

Leadership sees distribution patterns and cost impact, but makes the final call, preserving judgment while surfacing fairness gaps.

Chose visibilityover enforcement

Justification: mandatory, not blocking

Managers can override ratings or promotions, but must give a written rationale. Accountability without bottlenecks.

Chose accountabilityover gatekeeping

Escalation: structured and auditable

Rather than blocking workflows, exceptions trigger approval chains with full context, keeping decisions inside the system and traceable.

Chose traceabilityover side channels

Configuration: first-class, not administrative

Setup was treated as a product experience. Versioning, scoping and weightage design cut mid-cycle confusion and corrections by over 50%.

↓52% appraisal setup time.

ReflectionWhat I carry forward

The system shifted performance management from manual coordination to a trusted decision framework

Systems scale better than policies

Governance must be embedded in workflows, not documented outside them.

Guardrails enable trust, not friction

Mandatory justification creates accountability without blocking progress.

Fairness needs visibility, not enforcement

Showing distribution patterns lets leadership make informed decisions instead of algorithmic ones.

Leadership is deciding what not to automate

We automated coordination, not judgment. And configuration is a product, not a feature.

About me

“I design and lead systems that make complex products clear and scalable.”

I'm a Product Design Manager with 10+ years in product design (13 years in design overall), designing and scaling fintech and SaaS products.

10+ yearsProduct design · 13 in design overall

B2B, B2C, FintechDomains

Systems & TeamsPrimary strength

My storyWhere I am now

Design systems, UX strategy and team leadership

My work sits at the intersection of design systems, UX strategy and team leadership. I enjoy solving complex problems and turning them into simple, usable experiences that work for both users and businesses.

At Prosperr, I lead design for AI-assisted tax products across B2C and B2B platforms, partnering closely with product and engineering teams to build scalable, user-focused solutions. My work has contributed to improved funnel completion, better advisor efficiency, and reduced manual processing through well-designed workflows and automation.

Beyond delivery, I focus on building design maturity: setting up design systems, research practices and team rituals that make design consistent, predictable and easier to scale. I also enjoy mentoring designers and helping teams grow their design thinking and craft.

My approachHow I work

Four things I bring to every product

User-centred design

Users struggle with unclear interfaces. I validate assumptions through research and testing, which drives higher adoption and lower support costs.

Strategic design decisions

Great design must drive measurable outcomes. I align design decisions with revenue, retention and efficiency metrics, which proves ROI to stakeholders.

Systems thinking

Inconsistent experiences slow teams and confuse users. I build scalable design systems that ensure consistency, which accelerates development velocity.

Team and cross-functional alignment

Silos create misalignment and delays. I co-create solutions with product, engineering and data teams, so we solve the right problems.

How I leadPhilosophy

Creating the conditions for great work

Leadership philosophy

Great design leadership creates conditions for teams to do their best work. I set clear direction, remove blockers and foster psychological safety, and I build systems of collaboration, feedback and accountability that scale.

Collaboration style

I work best when design has a strategic seat at the table. I partner with product, engineering and data to solve the right problems. My approach: share work early, invite critique, iterate transparently.

Beyond workWhat keeps me curious

Outside the day job

Creative exploration

Experimenting with motion design, 3D rendering and AI-assisted tools.

Lifelong learning

Staying current with emerging design systems and AI technologies.

Community

Mentoring designers and contributing to design community knowledge.