Neha Mayacharya

Clarity · Self-initiated concept

An AI copilot for the mortgage lifecycle

Borrower app · Loan officer copilot · Regulated fintech

AI guides. Humans decide.

Product & AI Experience Designer · Self-initiated concept

Task analysis to hi-fi, designed end to end

See the screens

  • ROLE

    Product Designer

  • SCOPE

    Borrower app · Loan officer copilot

  • METHODS

    Task analysis · Service blueprints · Prototyping

  • CONTEXT

    ECOA · Reg B · TRID

The problem

The mortgage lifecycle spans two disconnected worlds.

Borrowers move through an opaque process with almost no real-time visibility. Loan officers reconcile information that already exists somewhere, just not in front of them.

LB

Laura Bennett, 34

First-time homebuyer

“I can handle complexity, I can’t handle silence.”

  • Visibility blackouts at the highest-stakes moments: underwriting and closing

  • Generic rejections (“document not valid”) with no reason or next step

  • Surprises land late: a $1,200 fee change two days before signing

Core problem: visibility
AW

Adam Webb, 41

Senior loan officer · 12 years · NMLS-licensed

“I didn’t get into this job to be a data-entry clerk between five systems.”

  • 8–12 systems: origination, compliance checklists, pricing engines, CRMs

  • Manual cross-referencing of data that already exists elsewhere in the org

  • Requests and updates scattered across email and phone

Core problem: fragmentation

CONSTRAINT

Fair lending law (ECOA, Regulation B) requires a documented, human-owned rationale behind every credit decision. “AI decides” isn’t a design option. It’s a compliance non-starter.

Current-state journey

Where it breaks, stage by stage

From the task analysis across five lifecycle stages

Pre-approval
Application
Underwriting
Closing
Servicing

Laura

borrower

Unsure what pre-approval actually commits her to

“Document not valid.” No reason, no next step

Nine days of silence. No idea who has her file or why

$1,200 fee surprise two days before signing

Handed to a servicer she’s never heard of

Adam

loan officer

Re-keys the same data into pricing engine and origination system

Chases missing pages by email and phone

Reconciles income by hand across application, pay stubs, W-2s

Fee changes arrive from the title company with no alert

Loses visibility once the loan funds

Laura’s

confidence

PATTERN 1

Manual reconciliation across disconnected systems

PATTERN 2

Opacity at the borrower’s most anxious moments

PATTERN 3

No structured, two-way communication

How might we

How might we design a connected AI experience, spanning borrower and loan officer, that reduces fragmentation and cognitive load through guidance and analysis, while keeping the human as the sole decision-maker at every consequential step?

FROM PATTERN 1 · RECONCILIATION

…let AI do the cross-checking, so Adam starts from the discrepancy instead of the raw data?

FROM PATTERN 2 · OPACITY

…replace silence with continuous, honest visibility at Laura’s most anxious moments?

FROM PATTERN 3 · COMMUNICATION

…give both sides one shared source of truth, where nothing reaches Laura without Adam’s review?

Success looks like: fewer systems in Adam’s day · no silent stretches for Laura · every decision traceable to a named human

Comparative analysis

Not a midpoint. A different architecture.

PRODUCT

WHAT IT SOLVES

AI IN THE DECISION

Wells Fargo

Your Loan Tracker

Visibility: status, to-do lists, secure messaging

None

Bank of America

Home Loan Navigator

Visibility: status, to-do lists, secure messaging

None

Rocket Mortgage

Automated income and asset verification; underwriting judgment stays human, an emergent pattern, not a stated principle

Partial, upstream

Better.com

Tinman

Conversational AI that underwrites end to end, in as little as 47 seconds; underwriters join only on AI-flagged variances

AI is the decision

Clarity

AI on every file, every time, with a flag-or-clear moment reviewed by a named, accountable person

Guides only

MARKET POSITIONING

Ideation · three directions

Where should the AI sit?

Each direction was tested against the same two questions: does it solve both Laura’s visibility and Adam’s fragmentation, and who owns the decision?

DIRECTION A

Two point solutions

Tracker?LO tool

A borrower status tracker plus a separate efficiency tool for loan officers, the pattern most banks ship today.

SET ASIDE

Laura’s visibility depends on Adam’s reconciliation. Solving them separately leaves the tracker showing stale status.

DIRECTION B

AI decides, humans handle exceptions

AI underwriteshuman if flagged

Full automation end to end; a licensed underwriter joins only when the AI itself raises a variance.

SET ASIDE

No human-owned rationale on most files: a hard problem under ECOA / Reg B at a large regulated lender.

DIRECTION C · CHOSEN

Connected copilot: guide, don’t decide

LauraRecordAdamAI

AI on every file, every time: surfacing, flagging, drafting. One shared record feeds both apps. Every consequential step ends with a named human.

WHY

Solves visibility and fragmentation with one system, and makes the compliance constraint part of the structure.

Design principle

Guide, don’t decide, at every stage

The AI’s role never changes category: it surfaces, flags, drafts and cross-checks. A person confirms, approves, decides and sends.

Pre-approval

Pulls income and credit data; drafts loan scenarios

Chooses the program; issues the letter

What pre-approval means, in plain words

Application

Checks each upload within minutes; drafts specific requests

Reviews and edits every message before it sends

The exact reason, and what’s still outstanding

Underwriting

Cross-checks income and assets; flags anomalies with citations

Recalculates, requests or clears, with rationale logged

Calm status, who’s handling it, how long it takes

Closing

Compares disclosure versions; detects fee changes on arrival

Verifies with title; approves the explanation

Changes flagged with timestamps, days before signing

Servicing

Watches for escrow and payment changes; drafts notices

Confirms before anything reaches the borrower

An early heads-up instead of a surprise bill

RULE

The left column never moves right. Not by stage, not by persona, not by confidence score.

UX flow · information architecture

Two apps, one shared case record

The borrower app and the loan officer copilot never talk to each other directly. Both read from and write to a single source of truth.

SOURCE SYSTEMS

Origination · Pricing · Compliance · Credit · Title

Shared case record

single source of truth

Loan data & documents

Flags & decision log

Messages (approved only)

Stage timeline

Named owner per decision

Borrower app

Laura · mobile

Home dashboard

Documents & rejections

Stage status

Closing Disclosure

Messages with Adam

Loan officer copilot

Adam · desktop

Pipeline (triage-first)

Customer tabs

Risk review

Document request drafts

Closing verification

AI layer

Reads the record · cross-checks · flags · drafts

Writes nothing without Adam’s approval

WHY THIS SHAPE

The AI that fixes Adam’s reconciliation problem is the same AI that gives Laura her visibility, but only through a record Adam has signed off on.

One underwriting flag, end to end

Every branch routes through Adam first

Every file gets a flag-or-clear moment reviewed by a named person, however confident the AI is. Laura hears only what Adam has approved.

UX, wireframes to hi-fi

Lo-fi, mid-fi, then hi-fi. One screen at a time.

Structure was settled in grayscale before any color, copy or data went in.

Adam · Risk review

LO-FI

MID-FI

HI-FI

Laura · Home

LO-FI

MID-FI

HI-FI

Key decisions

Four calls that keep a human accountable

Guide, don’t decide

Clarity flags, explains and cites. Adam picks the outcome and writes the rationale.

Nothing reaches Laura unreviewed

Every AI draft waits for Adam to edit and send. The AI’s path never reaches Laura directly.

Every decision has a name on it

Rationales are logged under the person who decided, ready for fair-lending review (ECOA / Reg B).

Reassurance for Laura, triage for Adam

One flag, two views: a calm status with a named owner for Laura, a ranked queue for Adam.

Hi-fi

The full loop, on both sides

Laura’s side is reassurance-first. Adam’s is triage-first. The same flag moves through both without ever skipping Adam.

Laura · Borrower app

Home dashboard
Document rejection
Underwriting status
Closing Disclosure

Adam · Loan officer copilot

Pipeline dashboard
Document request draft
Risk review
Closing verification

COMPONENT

Customer tab bar

Four states, so an open flag is seen without clicking in: default, hover, flagged, and active + flagged.

Outcome

One AI, two completely different jobs, and one accountable human.

9 days → 0

silent days in underwriting, replaced by a live status with a named owner and an expected window

11 days early

the $1,200 fee change is caught and explained before the disclosure is final, not at the signing table

4 minutes

from upload to a specific “page 2 of 2 missing” message, drafted by AI and sent by Adam

Outcomes from the designed scenario, not measured results.

WHAT THIS PROJECT SHOWS

Service & systems thinking

Task analysis, blueprints and a shared-record architecture

A defensible AI stance

Enforced by structure, consistent across five stages

Regulated-industry fluency

ECOA, Reg B, TRID and NMLS shape the constraint

Range across two users

Same reasoning, translated for a first-time buyer and a 12-year loan officer

Clarity is a self-initiated design concept. Laura Bennett and Adam Webb are fictional personas; the product is not affiliated with any lender, and nothing here is legal or lending advice.

MORE PROJECTS

All work