Notification System Design

Purpose

Design notification systems that surface important buying signals while respecting cognitive limits. Core philosophy: "Never miss something important" without "constant interruption anxiety."

Related Skills

• •adhd-interface-design — For cognitive load and batching patterns
• •action-oriented-ux — For toast notifications and undo patterns
• •b2b-visualisation — For alert severity badges
• •uk-police-design-system — For semantic colour system

The Core Problem

Standard notification architectures prioritise real-time delivery and uniform visual urgency, inadvertently triggering:

• •Executive dysfunction
• •Alert fatigue
• •Decision paralysis
• •Platform abandonment

The Solution: Neuro-Adaptive Signal Engine

Shift from "Push-First" to a system that:

• •Treats attention as a finite, non-renewable resource
• •Adapts the "Quiet Cockpit" philosophy from aviation
• •Distinguishes between immediate intervention, ambient awareness, and scheduled consumption
• •Decouples information availability from interruption

Two Goals in Tension (Resolved)

Multi-Axis Alert Classification

The Five Classification Axes

Priority Score Calculation

Priority Score = (0.45 × Urgency) + (0.35 × Impact) + (0.20 × Confidence)

Urgency Scores:

• •Hiring surge: 0.9 (24-48h window)
• •Follow-up overdue: 0.9
• •Follow-up due today: 0.7
• •Job posting: 0.6 (≤7d window)
• •Market trend: 0.2

Impact Scores:

• •Key account (strong relationship): 1.0
• •Target account (some relationship): 0.7
• •General account: 0.4

Lane Assignment:

• •Lane A (Interruptive): Score ≥ 0.80 AND Actionable = true
• •Lane B (Digest): Score 0.40-0.79
• •Lane C (Ambient): Score < 0.40

Four-Tier Severity Model

Alert Fatigue Prevention

Causes of Alert Fatigue

ADHD-Specific Challenges

• •Compromised filtering — Difficulty distinguishing relevant from irrelevant
• •Executive function load — Manual categorisation drains limited resources
• •Time blindness — Critical alerts "meant to get to" but never acted upon
• •Object permanence — Dismissed notifications cease to exist mentally

Prevention Strategies

From Clinical Alarm Management:

• •Reduce non-actionable alarms at source
• •Tune thresholds and require confidence levels
• •Make interruptive alerts rare and highly trustworthy

From Aviation:

• •Reserve "Red" strictly for immediate danger
• •Suppress alerts during high workload
• •Indicate if alerts are suppressed

Batched Notification Delivery

Research Evidence

Predictable batches 3×/day improve well-being outcomes versus "endless stream."

Critical: Turning notifications off entirely can increase anxiety/FOMO. Batching provides safety net without interruption.

The Three-Batch Model

The Rule of Three

Each digest highlights maximum three critical items at top. Additional items accessible via expansion.

Two-Lane Delivery Model

INCOMING SIGNAL
      │
      ▼
┌─────────────────┐
│ Priority Calc   │
└────────┬────────┘
         │
    ┌────┴────┐
    │         │
    ▼         ▼
INTERRUPTIVE  DIGEST
(Score ≥0.80) (Score <0.80)
    │         │
    ▼         ▼
Push + Badge  Batched 3×/day

Debouncing

Wait for Debounce Window (e.g., 15 minutes) to group related events:

Bad:  "Microsoft posted 1 job" → "posted 2 jobs" → ... (15 notifications)
Good: [Wait 15 min] → "Microsoft: 15 New Roles (3 Engineering, 2 Sales...)"

Ambient Awareness vs Active Notification

Nagging Protocol for Critical Alerts

For Tier 1 alerts that remain unacknowledged:

User Controls:

• •"I've handled this externally" → Resolves without logging
• •"Not relevant" → Dismisses with feedback

Noise Control Rules

Visual Design System

Semantic Colour System

Dual-Coding Requirement

All colour cues must be accompanied by iconography:

• •Warning Triangle for critical
• •Arrow Up/Down for changes
• •Info Circle for informational

User Preference Framework

Mode Presets

Default Configuration

"Quiet Cockpit" Default: Low noise. Only Tier 1 interrupts. User can opt into more.

Success Metrics

Summary Principles

1. •Attention is finite — Defend it, don't exploit it
2. •Quiet Cockpit — Only anomalies generate alarms
3. •Predictability beats cleverness — 3 batches/day > ML-optimised timing
4. •Persistence over pinging — Items stay visible until resolved
5. •Opportunity framing — Never punitive, always supportive
6. •One-click actions — Bridge knowing and doing
7. •Dual coding — Colour + icon for every state
8. •Rare and trustworthy — Interruptive alerts must earn credibility
9. •Context-aware snoozing — "When I view this account" > "In 2 hours"
10. •Cross-device sanity — One source of truth, no duplicate badges