Why Choose This Topic?

Shared living in India is daily, messy, and under-served.
People get matched on rent & location, but real life runs on habits, tolerance, fairness, safety, and local rules. Current tools do math (bills) and lists (chores); they don’t do compatibility, mental load, or early prevention.

This space lets me show empathy + measurable outcomes + responsible AI — exactly the combination Deloitte cares about: leadership, integrity, care, inclusion, measurable impact.

Problem Identification

Key Questions

• Value: Which pains recur daily across PGs, hostels, coliving and flat-shares in India?

• Compatibility: What truly predicts a good match — habits & tolerance (veg/non-veg, quiet hours, guests, language, night shifts) rather than demographics?

• Fairness: How do we make mental/emotional load (maid/cook, gas, RWA forms, chasing UPI) visible and adjustable?

• Prevention: Which leading signals (streaks, tone drift, tolerance breaches) let us prevent conflict rather than react?

• Adaptation: How should the home auto-rebalance for travel, illness, exams, night shifts, outages?

• Governance: How do we reflect RWA/PG rules (CCTV/guest/curfew) and provider SLAs transparently?

• Trust: What can be processed on-device, what needs consent, and how do we explain decisions (“Why this match?”) without bias?

• Scale: Can this generalize from households → coliving operators → student/corporate housing?

Customer Pain Points

(From 2 Deloitte Employees)

1. Hygiene & cleanliness

2. Non-friendly/avoidant behaviour (privacy wall, low collaboration)

3. Work/quiet hours clashes (weekends, exams, night shifts, late calls)

4. Budget & job distribution (food timings/types, washing/cooking)

5. Restrictions & safety rules (weekends, guest policy, CCTV/curfew)

6. “Smell of conflict” (low conversation, messages ignored)

7. Provider delays (PG/room owners slow to fix)

8. Ragging / friends disturbing others

9. Chore allocation not followed

10. Habits (smoking/drinking) + no cleanup after

11. Unplanned food sharing / menu disputes

12. Unethical practices / boundary breaches

13. Need agreement, tracking, consent before staying

Personas Identified

Meera – Early-career PG/Coliving

• Include: PG/hostel/coliving in last 12 months; quiet after 10 pm; veg/mixed kitchen; cares about guest/CCTV clarity.

• Exclude: Lives alone/family home; no PG/RWA exposure.

Rohit – Night-shift IT/BPO

• Include: Day-sleep window; cohabits with non-night-shift roommates; real clashes around chores/noise.

• Exclude: Whole house on same schedule.

Aman – Pet Parent

• Include: Dog/cat in shared housing; society pet rules; allergy interplay if possible.

• Exclude: Pet-free building with full consensus.

Ananya – Household “Ops Lead”

• Include: Owns ≥3 admin lanes (maid/cook, gas, RWA, broadband, UPI rent).

• Exclude: Fully managed coliving with concierge.

Imran – Non-veg in mixed home

• Include: Uses separate utensils/zone; recurring hygiene/menu conversations.

• Exclude: Fully veg or fully non-veg homes with no contention.

Operator – RWA/PG/Coliving

• Include: Manages properties, sets rules, tracks SLAs/churn.

• Exclude: Broker/landlord with single tenant only.

Empathy Mapping

Meera (PG/coliving):

• Thinks/Says: “I need quiet after 10; I hate surprise guests.”

• Feels/Does: Cleans silently; avoids conflict; anxious about CCTV/guest norms.

• Pain: #1, #3, #5, #6, #11

• Gain: Cleanliness standards, quiet-hour guardrails, predictable guest rules.

Rohit (night shift):

• Thinks/Says: “Morning chores kill my sleep; I’m not lazy.”

• Feels/Does: Uses headphones; misses chores after shift; guilt → silence.

• Pain: #3, #6, #9

• Gain: Day-sleep quiet block; chore swaps by schedule; non-judgmental nudges.

User Journey Mapping

5 Whys (Root-Cause Analysis)

Hygiene/Cleanliness (#1):

1. No one cleans after use

2. No shared definition/standard

3. No proof/feedback loop

4. Rules not codified

5. Tools track tasks, not standards/acceptance
   Roots: rules gap, verification gap, incentive gap

Quiet hours / night shifts / exams (#3):

1. Routines clash

2. Schedules not captured

3. No policy engine to rebalance

4. No early escalation

5. No quiet-window visibility/enforcement
   Roots: orchestration/policy gap, visibility gap

Chores ignored (#9):

1. Assigned ≠ done

2. One person nags

3. No effortless proof

4. No consequence or swaps

5. Perceived unfairness
   Roots: automation gap, fairness gap

Silence/avoidance (#6):

1. People dodge awkward talks

2. No neutral nudges/mediator

3. Tone drift unseen

4. Past nudges felt accusatory

5. No household-level early warning
   Roots: early-warning gap, safe-mediation gap

Provider delays (#7):

1. Tickets vanish

2. No SLA transparency

3. No shared view

4. Ad-hoc re-raising

5. No single source of truth
   Roots: accountability & transparency gap

Market Research

Role-play interviews (2 users) — Directional Data Points:

• Hygiene/cleanliness friction: 2/2

• Quiet hours/routine conflicts: 2/2

• Chores not followed / mental-load concentration: 2/2

• Guest policy/CCTV confusion: 1/2

• Food/veg zoning/menu disputes: 2/2

• Silence/ignored messages: 2/2

(Use these as seeds, not population stats.)

Hypothesis Benchmarks to Validate:

• ≥ 70% rate habits/tolerance as top-3 match criteria

• ≥ 60% report one person doing most household coordination

• Streaks + tone drift precede conflicts in ≥ 65% of cases

• Auto-rebalance proposals accepted ≥ 60–70% if explained and reviewable

• Agreement + consent reduces disputes in ≥ 30% of households within 1–2 months

Problem Statement

P1 — Habit & Tolerance Matching (India):
Prospective roommates are matched on rent/location, not habits/tolerance, resulting in mismatched homes and 90-day churn.
(#1, #3, #5, #10, #11, #12)

P2 — Fairness beyond Money (Mental Load):
Invisible mental/emotional load isn’t recognized, so people feel unfairness even with equal splits → resentment & non-compliance.
(#4, #9)

P3 — Micro-Conflict Prevention:
Homes react after fights; there’s no early-warning combining missed-chore/UPI streaks, tolerance breaches, and chat tone drift → blowups, silence, exits.
(#6, #8, #9, #10, #11)

P4 — Adaptive Exceptions & Orchestration:
Static plans don’t adapt to night shifts, exams, travel, outages; one person becomes the “house PM” → burnout and rule avoidance.
(#3, #4, #9, #11)

P5 — Safety & Governance (RWA/PG):
Guest/CCTV/curfew rules are unclear; ragging/friend disturbances occur; PG/room providers delay fixes; there’s no SLA transparency → distrust and escalation.
(#5, #7, #8)

P6 — Agreement & Consent:
No standard, privacy-respecting roommate agreement with consent, rule tracking, and grievance workflow → ambiguity and repeated disputes.
(#13)

Ethics, Privacy, and Deloitte Values

• Privacy-first: per-signal consent; on-device NLP/CV where feasible; data minimization; deletion on request.

• Bias-safe matching: never optimize on protected traits. Language only for communication/accessibility.

• Explainability: “Why this match?” + “Why this suggestion?” + appeal flow.

• Household-level risk: no labeling individuals; focus on situation.

• Safety: anti-harassment/ragging workflows; guest/CCTV compliance tracking; escalation routes.

• Deloitte alignment: Lead the way, Integrity, Care, Inclusion, Measurable impact.

Prioritization (Value × Confidence × Ease)

1. P1 Habit/Tolerance Matching — High Value, High Confidence, Medium Ease → Top 1

2. P2 Fairness beyond Money — High Value, Med-High Confidence, Medium Ease → Top 2

3. P3 Micro-Conflict Prevention — High Value, Medium Confidence, Medium Ease → Top 3

4. P4 Adaptive Exceptions — Medium-High Value, Medium Confidence, Medium Ease

5. P5 Safety & Governance — High Value, Medium Confidence, Lower Ease

6. P6 Agreement & Consent — Medium Value, High Ease, High Confidence → can ship alongside P1/P2

Analytics/AI Enablers

• HCTS (Habit/Tolerance Match Score): Structured habits + tolerance vectors → embeddings → constraint-aware ranking. “Why this match?” explainer; ID/phone verification; fairness audits.

• Fairness Index (beyond money): Behavioral analytics; optional on-device CV proof; gen-AI suggests offsets/swaps.

• Harmony Risk (prevention): Local NLP for Hinglish/regional sentiment + time-series on missed-chore/UPI streaks → household-level risks; gentle nudges.

• Policy Engine (adaptive): Constraint solver + RL suggests auto-rebalances; human-approved, auditable.

• Orchestration Copilot: Context reminders; alternative proofs; no raw images stored.

• Pets/Medical/Accessibility: Care zoning; quiet-hour guardrails; sensor fusion for pet routines; low-stimulus UI; speech↔text; multi-language.

Phase 2

Focus

Prioritize P1 and P2 for initial measurement:

• P1 — Habit & Tolerance Matching (India): Rent/location matching misses habits & tolerance → wrong fit & 90-day churn.

• P2 — Fairness Beyond Money (Mental Load): House admin is invisible → resentment, non-compliance.

Support next (measured, but 2nd wave):

• P3 — Micro-Conflict Prevention

• P4 — Adaptive Exceptions & Orchestration

• P5 — Safety & Governance

• P6 — Agreement & Consent

(This order follows your Value–Confidence–Ease stack from Phase 1.)

OKRs

Objective O1 — Get Matching Right, Bias-Safe (Compatibility)

• KR1: ≥ 75% of new matches have Habit/Tolerance Match Score (HCTS) ≥ 70

• KR2: “Why this match?” explainer viewed on ≥ 70% of match decisions

• KR3: 90-day satisfaction ≥ 75% for new matches

• KR4 (guardrail): 0 confirmed bias incidents; 100% fairness audits passed

Objective O2 — Make Fairness Visible & Reduce Resentment

• KR1: Fairness suggestions acceptance ≥ 60%; decline reasons logged ≥ 90%

• KR2: UPI late-payment streaks ↓ 40% within 8 weeks

• KR3: Missed-chore streaks ↓ 35%

• KR4: Top reminder-burden share ≤ 35%

Objective O3 — Prevent Escalation (Early Warning & Mediation)

• KR1: Harmony risk alerts precision ≥ 70%, false positives ≤ 10%

• KR2: Dispute median resolution time ↓ 40%

• KR3: Nudge follow-through ≥ 60% on risk alerts

Objective O4 — Respect Real Life with Adaptive Orchestration

• KR1: Auto-rebalance accept rate ≥ 65%; overrides ≤ 20%

• KR2: Assign → complete (chores) median time ↓ 30%

• KR3: Bill → UPI paid median time ↓ 25%

Objective O5 — Build Trust on Safety & Governance

• KR1: Provider ticket on-time SLA ≥ 90%; escalations ↓ 50%

• KR2: Guest/CCTV/curfew rule ack ≥ 95%; compliance ≥ 95%

• KR3 (guardrail): Ragging/harassment incidents ↓ 50%

From OKRs to a Single North Star Metric

Chosen NSM:
% of households with a conflict-free month

• Why: Captures the promise of shared living (calm, fair, compatible) and ties directly to churn and referrals.

• Definition: A household is “conflict-free” for a calendar month if all are true:

  1. No high-severity disputes logged (household-level).

  2. Harmony risk (household score) stays below threshold for ≥ 28/30 days.

  3. End-of-month check-in is “Agree” or “Strongly agree” to: “This month felt fair and respectful at home.”

  4. No safety escalations and no provider SLA breach older than 7 days.

Key Variables of NSM

• Activation (fit & clarity):

  • Habits/Tolerance profiles completed ≥ 80% of new users

  • “Why this match?” explainer viewed on ≥ 70% of match decisions

  • House rules set in ≥ 85% households week 1

• Reliability (follow-through & fairness):

  • UPI late-payment streaks ↓ 40% within 8 weeks

  • Auto-rebalance proposals accepted ≥ 65%

  • Reminder burden concentration ≤ 35%

  • Provider ticket on-time SLA ≥ 90%

• Efficiency (less friction, faster closure):

  • Assign → complete (chores) median time ↓ 30%

  • Bill → UPI paid median time ↓ 25%

  • Dispute median resolution time ↓ 40%

• Retention (staying power):

  • 90-day post-match satisfaction ≥ 75%

  • Renewal rate ↑ +15 pp in 6 months

  • Churn due to incompatibility ↓ 30%

• Guardrails (non-negotiable):

  • Privacy: ≥ 95% of sensitive signals processed on-device; 100% of deletion requests honored

  • Fairness: 0 optimization on protected traits; 100% quarterly bias audits passed

  • Safety: Ragging/harassment reports ↓ 50%; guest/CCTV compliance ≥ 95%

  • Accessibility: Low-stimulus/multi-language mode satisfaction ≥ 80% positive

Hypotheses (To Test)

1. Compatibility truth (India): Habits × tolerance predict 90-day satisfaction better than rent/location.

2. Fairness truth: Recognizing house admin with optional suggestions increases felt fairness and compliance.

3. Prevention beats reaction: Streaks + tone drift reliably precede conflict; household-level nudges reduce escalation.

4. Adaptation builds trust: Transparent, reviewable auto-rebalance during exceptions gets ≥ 60–70% acceptance.

5. **Agreement & consent reduce disputes ≥ 30% within 1–2 months in Indian households.

Experiments to be Planned

• E1 — HCTS + “Why this match?” (P1):

  • Design: Show HCTS and a transparent, plain-language explainer.

  • Measure: Match acceptance, explainer viewed %, perceived fairness, 90-day satisfaction.

  • Success: HCTS≥70 cohort shows +10–15 pp higher acceptance and ≥75% 90-day satisfaction.

• E2 — Fairness suggestions (P2):

  • Design: Recognize invisible work and propose offsets/swaps; user can accept/decline with a reason.

  • Measure: Acceptance %, decline reasons coverage, resentment proxy, streak reduction.

  • Success: ≥60% accept; streaks drop 35–40%; pulse improves by ≥0.3 on 5-pt scale.

• E3 — Harmony sentinel (P3):

  • Design: Local NLP on chat + streak tracker → household-level early alerts; nudges are optional and respectful.

  • Measure: Precision ≥70%, FP ≤10%, nudge follow-through %, escalations ↓.

  • Success: Reduced escalations ≥25% month-over-month in pilot homes.

• E4 — Adaptive policy engine (P4):

  • Design: For exceptions, propose a rebalanced plan; human review; audit log.

  • Measure: Accept rate ≥65%, overrides ≤20%, assign→complete time ↓ 30%.

  • Success: Less friction; higher felt fairness in exception weeks.

• E5 — Provider SLA transparency (P5):

  • Design: Shared, visible ticket SLA timeline; simple escalation path.

  • Measure: On-time ≥90%, escalations ↓50%, trust pulse +0.3.

  • Success: Fewer “they’re ignoring us” complaints.

• E6 — Agreement + consent starter (P6):

  • Design: India-specific agreement template; consent log.

  • Measure: Disputes ↓30%, acknowledged rules ≥90%, improved clarity scores.

  • Success: Clearer expectations; simpler mediation.

Data & Instrumentation

• profile_habits_completed {veg_zone, quiet_hours, guests_rule, language, night_shift, smoking, snoring, pets}

• tolerance_set {per-factor 1-5}

• HCTS_generated {score, factors, version}

• rules_set {quiet_hours, guests_policy, veg_zone, maid_slots, CCTV_ack}

• society_rule_acknowledged {rule_id, ack_time}

• task_* {type, assigned_to, due, completed_at, verified_by, on_device}

• reminder_* {channel, recipient_count}

• reminder_burden_share {top_user_pct}

• bill_created / bill_paid_UPI {amount, split, payers, paid_at}

• payment_late_streak {count, window}

• streak_missed_task {count, window}

• harmony_risk_alert {score, drivers, acted}

• tolerance_breach {type, time}

• quiet_hours_violation {time, severity}

• pet_care_event_logged {type}

• allergy_zone_breach {zone}

• ticket_created {provider, severity}