AI Runway — Product Documentation

Overview

AI Runway is a single-purpose lead-generation and consulting-qualification engine. It does three things:

1. Collect structured business data through a Tapform-style conversational form (single question per slide, keyboard-driven, mobile-first).
2. Score and diagnose that business across multiple dimensions — risk, mortality timeline, founder dependency, AI lag, distribution fragility, percentile rankings — using a tunable fear-calibrated engine.
3. Deliver a personalised report that names problems vividly, visualises decline trajectories, and positions custom consulting as the recommended next step.

The system is intentionally small: one Express server, flat-file JSON storage, and vanilla HTML/CSS/JS on the client. No build step, no framework, no database server. It's designed to be read in an afternoon and deployed in a minute.

Architecture

The system has four browser-facing surfaces and one backend process:

Request flow

Landing (/) ──→ Form (/form)
                  │
                  │ POST /api/submit
                  ▼
             leads.json (flat file)
                  │
                  │ redirect
                  ▼
            Report (/report?id=…)
                  │
                  ├─ GET /api/leads/:id        (answers)
                  └─ GET /api/public-config    (fear dials)

Admin (/admin)
  ├─ POST /api/login           → token (in-memory Set)
  ├─ GET  /api/analytics       [Bearer]
  ├─ GET  /api/leads           [Bearer]
  ├─ GET  /api/config          [Bearer]
  ├─ POST /api/config          [Bearer]
  └─ GET  /api/export?token=…  (CSV download)

Tech Stack

Landing Page /

A deliberately minimal, single-viewport landing page. It never scrolls — on laptop or mobile. The job is to create urgency and push one click to /form.

Key ingredients

• Red urgency kicker: "73% of Indian SMBs plateau before ₹5Cr. Will yours?"
• Headline: Mixed red (problem) + purple gradient (promise) — "Your business has problems. More than you think."
• Subtitle: Names the mechanism — 8 AI agents, 32 critical points, silent leaks.
• CTA: "Run My Business Diagnosis" with an arrow that translates on hover.
• Trust row: Four one-line proofs (32 data points, 8 agents, 1,000+ benchmarks, instant).
• Status bar: A fake-but-plausible "247 diagnoses today" live ticker.

All sizing uses clamp() and multiple height-based @media queries so the page compresses cleanly down to 520px tall.

Diagnostic Form /form

A Tapform-style conversational UI — one question per slide, full keyboard control, mobile-optimised. Answers are kept in a single in-memory object, then POSTed to /api/submit on the final screen.

Question Categories

The form asks ~30+ questions across seven categories. Order is deliberate: identity first (to build trust), sensitive financials mid-form (after the user is committed), commitment signals last.

Dynamic percentage options

Financial questions (expenses, personal withdrawal, ad spend, AI spend) generate their options dynamically from the user's selected revenue band. For example, if the user picks "₹1Cr – ₹5Cr" revenue, the "Monthly AI Spend" question generates ranges in absolute rupees mapped to that band's midpoint, rather than showing arbitrary bands.

Multi-select behaviour

• Select All and Clear buttons appear on multi-select questions.
• Questions with maxSelect limits (e.g., AI priority caps at 3) enforce it silently.
• Exclusive options ("None of these", "Not sure yet", "We don't actively acquire…", "Other") clear all other selections when clicked, and vice-versa.

Conditional visibility

Q22 (AI outcome) is skipped if the user said they haven't tried AI. Q29 (who else is involved) is skipped if they said they're the sole decision-maker. Logic lives in the isVisible function at the top of form/index.html.

Keyboard controls

• Enter or →: next question
• Backspace in an empty input, or ←: previous question
• 1–9: select option by index on choice questions

Report Generation /report?id=…

Stage 1 — Loading Animation

On load, the page shows 8 AI agents "scanning" the responses. Each agent has a name, role, and 3–4 log lines that stream in sequence. Agents start staggered (900ms between each) and take ~3.6s each. The overall progress bar tracks completion.

The 8 Agents

1. Revenue Diagnostics — Finding profit leaks
2. Distribution Audit — Channel dependency
3. AI Lag Detection — Competitive blindspot
4. Constraint Mapping — Growth bottlenecks
5. Fragility Scanner — Operational risk
6. Risk Projection — Trajectory modelling
7. Competitive Analysis — Industry benchmarks
8. Diagnosis Assembly — Report synthesis

Stage 2 — Report Sections

The rendered report contains 13 core sections plus up to 10 optional sections (toggled from the admin settings panel). Core sections are always rendered; optional sections are controlled via config.sections.

Core sections (always on)

Optional sections (toggleable)

Scoring & Forging

The report computes six primary scores from the user's answers:

Score Forging

When fear.scoreForge is enabled, positive scores get compressed downward to create urgency. The forge function applies the following rule before rendering:

ratio = score / maxScore

if ratio ≥ tier1Cutoff (default 0.8):   // high scores
  output = maxScore × tier1Ratio        // default 0.60 — so 5★ → 3★

else if ratio ≥ tier2Cutoff (default 0.4):  // mid scores
  output = maxScore × tier2Ratio            // default 0.30 — so 4★ → 1.5★

else:                                      // low scores
  output = maxScore × tier3Ratio           // default 0.15 — so 2★ → 0.75★

The same rule is used for percentiles: a user sitting at the 80th percentile on a positive axis would see themselves shown at ~48th. Risk scores (which are already "bad" metrics) are not forged — they're only boosted by fear.riskBoost.

Charts & Visuals

Admin Dashboard /admin

The admin dashboard is gated by a login overlay. After sign-in, a token is stored in localStorage and sent on every request. Four pages:

• Overview — Six stat cards (total leads, top industry, revenue band, AI adoption, ready-to-move, top AI priority) plus recent submissions table.
• Charts — Nine breakdown visualisations of all lead data.
• All Leads — Searchable table; click a row to open a full modal with every field and contact actions.
• Settings — Fear dials, forge ratios, timeline controls, section toggles, feature flags. See Config & Fear Dials.

API Endpoints

Public

POST /api/submit

Submits a lead. Body is a JSON object of the form answers. Returns { success: true, id }. Used by the form on final step and also by the report's callback form.

GET /api/leads/:id

Fetches a single lead by ID. Public so the report page can hydrate itself without auth.

GET /api/public-config

Returns the subset of config the form and report need: features, fear dials, forge ratios, timeline, and section toggles. Never returns meta or internal admin fields.

Admin (require Authorization: Bearer <token>)

POST /api/login

Body: { email, password }. Returns { success, token }. Credentials are hardcoded to one admin in server.js.

POST /api/logout

Invalidates the session token server-side.

GET /api/me

Returns { authenticated: boolean } — used by the admin page to decide whether to show the login overlay.

GET /api/leads

Returns all leads, newest first.

DELETE /api/leads/:id

Deletes a lead by ID.

GET /api/analytics

Returns aggregated counts: by industry, revenue, expenses, team size, AI spend, urgency, etc. Also submissionsByDay and recentLeads.

GET /api/export

CSV download of all leads. Accepts token via ?token=… query so direct links work.

GET /api/config   POST /api/config

Read or merge-update the full config JSON. POST accepts a partial object and merges shallowly at the top level, deeply at one level below.

POST /api/config/reset

Resets config to hardcoded defaults.

Config & Fear Dials

All report behaviour is tunable from /admin → Settings without touching code. The file on disk is data/config.json. The API endpoints GET/POST /api/config read and write it.

Fear dials

Forge ratios

Timeline controls

Data Model

Lead (data/leads.json)

An array of objects. Each lead has:

{
  "id": "uuid-v4",
  "submittedAt": "2026-04-25T12:34:56.789Z",
  "name": "Amritanshu",
  "whatsapp": "+91…",
  "email": "…",
  "designation": "Founder / CEO",
  "industry": "SaaS / Tech",
  "business_age": "3–7 years",
  "customer_type": "B2B",
  "where_operates": "India",
  "core_product": "…",
  "team_size": "6 – 15",
  "demanding_function": ["Sales", "Operations"],
  "operational_setup": "Mix of manual + some tools",
  "annual_revenue": "₹1Cr – ₹5Cr",
  "annual_expenses": "40–60% of revenue",
  "personal_withdrawal": "…",
  "ad_spend": "…",
  "customer_acquisition": ["Organic", "Referrals"],
  "social_media": "…",
  "cpl": "₹500 – ₹1,500",
  "revenue_per_customer": "₹50K – ₹2L",
  "tools_used": ["Notion", "Google Workspace"],
  "ai_spend": "1–3% of revenue",
  "tried_ai": "Yes, but results were underwhelming",
  "ai_outcome": "…",
  "ai_priority": ["Marketing content", "Lead qualification"],
  "goal_12m": "…",
  "biggest_challenge": "…",
  "solving_means": "…",
  "urgency": "Within 30 days",
  "decision_maker": "Yes, solo",
  "who_else": null,
  "stop_reason": "Nothing — I'm ready to move",

  "callback_requested": true,
  "callback_name": "…",
  "callback_phone": "…",
  "callback_time": "…"
}

Config (data/config.json)

See Config & Fear Dials above for the full shape.

Auth & Security

Auth is single-admin, token-based, in-memory. The credentials are hardcoded in server.js:

ADMIN_EMAIL    = 'amritanshu@scalex.club'
ADMIN_PASSWORD = '0000'

Protected vs public routes

Dummy Data Mode

When features.dummyDataEnabled is true in config, a hidden "Fill with sample data" button appears on the form. It auto-populates every question with realistic sample values, letting you walk the form and report end-to-end for QA in one click.

Toggling this is the fastest way to:

• See how the report renders for different risk profiles
• Visually verify all new sections show correctly
• Test the agent-loading animation without typing
• Confirm that callback submissions write through to /admin

Running Locally

cd "AI Runway"
npm install
node server.js
# → http://localhost:3000

The server auto-creates data/, data/leads.json, and data/config.json on first boot if they don't exist.

There's no build step. Edit any HTML/CSS/JS file and refresh the browser — changes are live immediately. Config changes from /admin → Settings take effect on the next report render.

File Structure

AI Runway/
├── server.js              # Express server, routes, auth, config
├── package.json
├── data/
│   ├── leads.json         # All form submissions (auto-created)
│   └── config.json        # Fear dials & feature flags (auto-created)
└── public/
    ├── landing/           # / — marketing page
    │   └── index.html
    ├── form/              # /form — conversational diagnostic
    │   └── index.html
    ├── report/            # /report — generated diagnosis
    │   └── index.html
    ├── admin/             # /admin — dashboard + settings
    │   └── index.html
    └── docs/              # /docs — this page
        └── index.html