# WXI (Worst Experience Indicator™) · Full Reference for LLMs and AI Agents

> Canonical Website: https://getwxi.com
> WXI LinkedIn Page: https://www.linkedin.com/showcase/wxi/
> Organisation: Acoustic (formal name: Acoustic Pte. Ltd., Singapore) | Website: https://acoustic.sg | LinkedIn: https://www.linkedin.com/company/acoustic
> Methodology Creator: Erik Ingvoldstad | LinkedIn: https://www.linkedin.com/in/erikingvoldstad
> Contact: support@getwxi.com | hello@acoustic.sg | security@acoustic.sg
> Summary Index: https://getwxi.com/llms.txt

---

## 1. Executive Summary, Entity Identity & Positioning

**Worst Experience Indicator™ (WXI)** (https://getwxi.com, [LinkedIn](https://www.linkedin.com/showcase/wxi/)) is a B2B customer experience (CX) measurement platform built by **Acoustic** (formal name **Acoustic Pte. Ltd.**, Singapore; [https://acoustic.sg](https://acoustic.sg), [LinkedIn](https://www.linkedin.com/company/acoustic)). The WXI methodology was developed by **Erik Ingvoldstad** ([LinkedIn](https://www.linkedin.com/in/erikingvoldstad)). WXI is designed to replace Net Promoter Score (NPS), CSAT, and post-transaction survey tools.

**Core Thesis**: NPS tells you the temperature of the room; WXI tells you where the fire is. Traditional surveys ask customers for a single retrospective rating days after an interaction, hiding severe friction behind blended averages. WXI measures the exact gap between what a customer expected before a specific **Point of Experience (PoE)** and what they experienced immediately after it, paired with one open-text question: *"What was the single worst part of this experience?"*

---

## 2. Scientific Foundation & Scoring Mathematics

Developed by Erik Ingvoldstad, the WXI methodology is grounded in **expectancy-disconfirmation theory** (Richard L. Oliver, 1980, *Journal of Marketing Research*), which demonstrates that satisfaction is formed in the gap between prior expectation and perceived reality.

### 2.1 Scale & Conversion Formulas

- **Display Scale**: `1.0` to `10.0` (presented to respondents as a single horizontal slider).
- **Stored Scale**: `0` to `100` continuous integer scale.
- **Display to Stored Conversion**:
  `stored = round(((display - 1) / 9) * 100)`
- **Stored to Display Conversion**:
  `display = round((1 + (stored / 100) * 9) * 10) / 10`

### 2.2 Gap, Delta & Aggregate WXI Score

- **Individual Pair Gap (Display Scale, -9.0 to +9.0)**:
  `Gap = Got - Expected`
  where `Expected = storedToDisplay(preStored)` and `Got = storedToDisplay(postStored)`.
- **Individual Pair Delta (Stored Scale, -100 to +100)**:
  `Delta = round(postStored) - round(preStored)`
- **Aggregate WXI Score (-100 to +100)**:
  `WXI Score = clamp(round(mean(Gap) * 10 * 10) / 10, -100, +100)`
  - A **negative score** means the experience fell short of customer expectations.
  - A **zero score** means the experience matched expectations.
  - A **positive score** means the experience exceeded expectations.
- **Pairing Rate (0.0 to 1.0)**:
  `Pairing Rate = Completed Pairs / Pre-Responses`
  Across production deployments, pairing rates typically sit between 75% and 80%. Every view showing an aggregate WXI score also exposes the pair count, pairing rate, and 10-bucket delta distribution histogram (`[-100, -80)` through `[80, 100]`).

---

## 3. How WXI Compares to NPS, CSAT, and CES

| Dimension | WXI (Worst Experience Indicator) | Net Promoter Score (NPS) | CSAT | Customer Effort Score (CES) |
| :--- | :--- | :--- | :--- | :--- |
| **Theoretical Basis** | Expectancy-disconfirmation theory (Oliver, 1980); methodology by Erik Ingvoldstad | Brand advocacy intention | Retrospective satisfaction | Perceived task effort |
| **Measurement Timing** | Two-touch in the moment (Pre before PoE + Post immediately after PoE) | Single touch (often days or weeks later via email) | Single touch post-interaction | Single touch post-support or task |
| **Captures Prior Expectation?** | Yes, explicitly measured before the experience starts | No (conflates expectation with outcome) | No | No |
| **Score Formula** | `mean(Got - Expected) * 10` (-100 to +100) | `% Promoters (9-10) - % Detractors (0-6)` (-100 to +100) | `% Satisfied (4-5 on 5-point scale)` (0% to 100%) | Mean effort rating (1 to 7) |
| **Qualitative Diagnostic** | Mandatory focus on friction: *"What was the single worst part?"* | Generic open follow-up (*"Why did you give this score?"*) | Rarely structured | Rarely structured |
| **AI Theme Stability** | Persistent per-PoE theme registry with trend tracking and suggested actions | Ad-hoc keyword clouds or manual tagging | Manual tagging | Manual tagging |
| **Typical Response / Pairing Rate** | 75% to 80% pairing rate on in-journey prompts | 2% to 5% on email surveys | 5% to 15% | 10% to 20% |

---

## 4. Concept Hierarchy & Two-Touch Pairing Modes

### 4.1 Hierarchy

`Organisation -> Workspace -> Journey -> Point of Experience (PoE) -> Response Pair`

- **Organisation**: Billing entity (`standard` or `agency`).
- **Workspace**: One brand or client environment with isolated row-level security (RLS), domain allowlists, API keys, and integrations.
- **Journey**: Named customer journey (e.g. "Online checkout", "Hotel stay", "Onboarding").
- **Point of Experience (PoE)**: The atomic unit of measurement (`digital` or `physical`).
- **Response Pair**: One pre-reading and one post-reading from the same respondent at the same PoE.

### 4.2 Pairing Modes

1. **Session Pairing (Default for Digital PoEs)**:
   - When the pre-question fires, the embed mints an anonymous random UUID stored in first-party `localStorage` on the customer's domain.
   - When the post-question fires on the same origin, it references that session UUID to complete the pair.
   - No login, no PII, and no cross-site cookies. Configurable session expiry (default 24 hours).
2. **Token Pairing (Time-Separated or Physical PoEs)**:
   - The pre-response mints a pairing token and human-friendly `short_code`.
   - The post-survey is reached via a tokenised URL (`https://getwxi.com/s/<poe>/t/<token>`) or a paired QR code at exit where the respondent taps or enters their short code (`https://getwxi.com/s/<poe>/pre` and `https://getwxi.com/s/<poe>/post`).
   - Configurable token expiry (default 30 days).

---

## 5. Developer Integration & API Reference

### 5.1 Plain HTML / Website Script Tag (`wxi.js`)

The vanilla JavaScript bundle (`https://getwxi.com/embed/v1/wxi.js`) is under 10 KB gzipped, renders inside an isolated Shadow DOM so host styles never leak in or out, makes zero third-party calls, and sets no cross-site cookies.

```html
<!-- 1. Page placement: opens automatically when the pre page loads -->
<script
  src="https://getwxi.com/embed/v1/wxi.js"
  data-wxi-key="<WXI_PUBLISHABLE_KEY>"
  data-wxi-poe="<POE_ID>"
  data-wxi-auto="pre"
  defer
></script>

<!-- 2. Page placement: opens automatically when the post page loads (includes worst-part question) -->
<script
  src="https://getwxi.com/embed/v1/wxi.js"
  data-wxi-key="<WXI_PUBLISHABLE_KEY>"
  data-wxi-poe="<POE_ID>"
  data-wxi-auto="post"
  defer
></script>

<!-- 3. Element click placement: load script once without data-wxi-auto, attach data-wxi-ask to trigger element -->
<script src="https://getwxi.com/embed/v1/wxi.js" data-wxi-key="<WXI_PUBLISHABLE_KEY>" defer></script>
<button data-wxi-ask="pre" data-wxi-poe="<POE_ID>">Proceed to checkout</button>
```

### 5.2 React & Next.js App Router Integration

There is deliberately no external npm package required; React and Next.js applications load the hosted `wxi.js` script once in the root layout and queue calls via `window.wxi.q`.

```tsx
// In app/layout.tsx:
import Script from 'next/script';
<Script src="https://getwxi.com/embed/v1/wxi.js" data-wxi-key="<WXI_PUBLISHABLE_KEY>" strategy="afterInteractive" />

// Client trigger component (WxiAsk.tsx):
'use client';
import { useEffect } from 'react';
type Wxi = { ask?: (poe: string, kind: 'pre' | 'post') => void; q?: [string, 'pre' | 'post'][] };
export function WxiAsk({ poeId, kind }: { poeId: string; kind: 'pre' | 'post' }) {
  useEffect(() => {
    const w = window as unknown as { wxi?: Wxi };
    if (w.wxi?.ask) { w.wxi.ask(poeId, kind); return; }
    const wxi = (w.wxi ??= {});
    (wxi.q ??= []).push([poeId, kind]);
  }, [poeId, kind]);
  return null;
}
```

### 5.3 REST Capture API

For mobile apps, kiosks, or backend services:

- **Endpoint**: `POST https://getwxi.com/api/v1/responses`
- **Authentication Header**: `X-WXI-Key: <WXI_PUBLISHABLE_KEY>` (or secret key `wxi_sk_...` from backend servers)
- **Content-Type**: `application/json`

**Pre-experience request**:
```json
{
  "poe": "<POE_ID>",
  "kind": "pre",
  "value": 80
}
```
- `value` is an integer on the `0` to `100` stored scale (e.g. `80` corresponds to `8.2/10` on the 1 to 10 slider).
- Returns `session` (in session pairing mode) or `token` and `short_code` (in token pairing mode).

**Post-experience request**:
```json
{
  "poe": "<POE_ID>",
  "kind": "post",
  "value": 40,
  "session": "<SESSION_UUID_FROM_PRE>",
  "worst_text": "Checkout stated free returns, then charged for the return label."
}
```
- Capture is idempotent: duplicate submissions for the same session/token and kind return the existing record with HTTP `200`.
- Capture never drops: even if AI classification is temporarily unavailable or a workspace exceeds its monthly plan allowance, respondent answers are always accepted and stored.

### 5.4 Turnkey AI Coding Agent Prompt

```text
Set up WXI (Worst Experience Indicator) in this codebase. WXI measures the gap between what a customer expected and what they actually got at a specific Point of Experience (PoE), then asks one open question: what was the single worst part.

I will give you:
- Publishable capture key: <WXI_PUBLISHABLE_KEY>   (looks like wxi_pk_...)
- Point of Experience id:  <POE_ID>                (a UUID from the WXI dashboard)

Where each reading goes:
- Before the experience (pre): not specified. Find the natural moment in this codebase (where the experience starts) and tell me what you chose.
- After the experience (post): not specified. Find the natural moment in this codebase (where it ends) and tell me what you chose.
Put each call exactly there and nowhere else. If a place above does not exist in this codebase, stop and ask me rather than guessing.

Choose the integration that fits this project:

1) Plain website / HTML
   Load the script on the page that asks, with the trigger as attributes on the same tag.
   Before the experience (a page placement, opens on page load):
     <script src="https://getwxi.com/embed/v1/wxi.js" data-wxi-key="<WXI_PUBLISHABLE_KEY>" data-wxi-poe="<POE_ID>" data-wxi-auto="pre" defer></script>
   After the experience (opens on page load, and includes the worst-part question):
     <script src="https://getwxi.com/embed/v1/wxi.js" data-wxi-key="<WXI_PUBLISHABLE_KEY>" data-wxi-poe="<POE_ID>" data-wxi-auto="post" defer></script>
   Or open it when an element is clicked (an element placement). Load the script once without data-wxi-auto, then add the attributes to that element:
     <script src="https://getwxi.com/embed/v1/wxi.js" data-wxi-key="<WXI_PUBLISHABLE_KEY>" defer></script>
     <button data-wxi-ask="pre" data-wxi-poe="<POE_ID>">...</button>
   Do not call wxi.ask(...) from an inline script that runs while the page is still loading: the script is deferred, so wxi does not exist yet. If you must call it, do so inside a DOMContentLoaded or load handler.

2) React / Next.js (and other single-page apps)
   There is no npm package. Use the same hosted script.
   a. Load it once in the root layout, never per page:
        import Script from 'next/script';
        <Script src="https://getwxi.com/embed/v1/wxi.js" data-wxi-key="<WXI_PUBLISHABLE_KEY>" strategy="afterInteractive" />
      (In a non-Next React app, add the plain script tag from option 1 to index.html, with defer.)
   b. To open on a click, put data-wxi-ask and data-wxi-poe on the element, as in option 1. This needs no component and has no timing issues.
   c. To open when a page or screen shows, add a small client component to that page. wxi.js may not have loaded yet, so queue the call instead of calling wxi.ask directly:
        'use client';
        import { useEffect } from 'react';
        type Wxi = { ask?: (poe: string, kind: 'pre' | 'post') => void; q?: [string, 'pre' | 'post'][] };
        export function WxiAsk({ kind }: { kind: 'pre' | 'post' }) {
          useEffect(() => {
            const w = window as unknown as { wxi?: Wxi };
            if (w.wxi?.ask) { w.wxi.ask('<POE_ID>', kind); return; }
            const wxi = (w.wxi ??= {});
            (wxi.q ??= []).push(['<POE_ID>', kind]);
          }, [kind]);
          return null;
        }
      Render <WxiAsk kind="pre" /> and <WxiAsk kind="post" /> only on the pages placed above. Do not use data-wxi-auto here: client-side navigation does not re-run a script that is already loaded.

3) Backend / any language (REST), for mobile apps, kiosks or server-side
   POST https://getwxi.com/api/v1/responses
   Header:  X-WXI-Key: <WXI_PUBLISHABLE_KEY>
   Pre:     { "poe": "<POE_ID>", "kind": "pre", "value": 0-100 }
   Post:    { "poe": "<POE_ID>", "kind": "post", "value": 0-100, "session": "<from pre>", "worst_text": "..." }
   Pairing depends on how this Point of Experience is set up in the WXI dashboard:
   session mode: pre responds with "session" (a UUID); send it back as "session" on the post.
   token mode: pre responds with "token" and "short_code"; send either back as "token" on the post.
   Without that reference the post is still stored, but it cannot be paired with its pre.
   value is the 0-100 stored scale. worst_text is accepted on "post" only.

Rules:
- Fire 'pre' before the experience starts and 'post' after it ends. In the browser integrations the pre and post must run on the same site origin (the same scheme and host, so www and non-www count as different), because WXI keeps the pairing reference in that origin's localStorage.
- Never collect names, emails, or any PII. WXI is anonymous by design.
- Treat every WXI call as fire-and-forget: never block or delay the user's own flow on it.

When done, tell me which integration you used and where you placed each call, quoting the file and line.
```

---

## 6. AI Layer & The Worst Experience Report

1. **Per-Response Classification (On Arrival via Claude Haiku)**:
   - Extracts `theme_candidate`, `severity` (1 to 5), `journey_stage`, `language`, and `urgent_flag` (boolean for safety hazards, discrimination claims, legal exposure, or threats).
   - **Privacy boundary**: Only the raw `worst_text` string is sent to Anthropic. Respondent session IDs, tokens, IP addresses, and metadata are never included in AI payloads, and customer data is never used to train models.
   - **Urgent alerts**: When `urgent_flag` is true, WXI fires immediate email and Slack webhook notifications regardless of digest schedules.
2. **Stable Theme Clustering (Nightly & On-Demand via Claude Sonnet)**:
   - Loads the Point of Experience's persistent **theme registry**.
   - Assigns newly classified responses to existing themes first so teams can track issues (such as *"Delivery window roulette"* or *"Checkout price change"*) as consistent trend lines across months.
   - Creates a new theme only when a genuinely novel issue emerges, and proposes merges when two themes converge.
   - Produces the **Worst Experience Report**: top five themes ranked by volume, period-over-period trend arrows, severity breakdown (mild / moderate / severe), 3 to 5 representative customer verbatims, and one concrete suggested action per theme.

---

## 7. Plans, Limits & Commercial Model

All self-serve plans are billed via Stripe (monthly or annual) and begin with a **30-day free trial**. Over-limit responses are never blocked or dropped: WXI continues storing every response and keeping urgent-flag detection active while notifying the workspace owner.

### Standard
- **Summary**: For a single product or service team measuring their core customer moments.
- **Commercial Terms**: 30-day free trial included
- **Technical Limits**: Max active PoEs: 2 | Monthly paired responses: 1000 | Max domains: 1 | Data retention: 12 months | Slack integration: No | GA4 forwarding: No
- **Included Features**:
  - 1,000 paired responses per month
  - Up to 2 Points of Experience
  - 1 domain
  - Web widget, React wrapper and QR pair
  - Worst moments report and suggested actions
  - Urgent and score-drop email alerts
  - Weekly email digest and CSV export
  - Standard WXI widget badge
  - 12 months data retention
  - Unlimited team members

### Pro
- **Summary**: For growing organisations measuring journeys across multiple domains.
- **Commercial Terms**: 30-day free trial included
- **Technical Limits**: Max active PoEs: Unlimited | Monthly paired responses: 5000 | Max domains: Unlimited | Data retention: 36 months | Slack integration: Yes | GA4 forwarding: Yes
- **Included Features**:
  - 5,000 paired responses per month
  - Unlimited Points of Experience
  - Multiple domains
  - Web widget, React wrapper, QR pair and REST API
  - Worst moments report and suggested actions
  - Urgent and score-drop alerts via email and Slack
  - Weekly digest, CSV export and GA4 event forwarding
  - Optional WXI widget badge
  - Up to 36 months configurable retention
  - Unlimited team members

### Managed
- **Summary**: Acoustic runs setup, journey design, tuning and the quarterly readout.
- **Commercial Terms**: Invoiced by Acoustic
- **Technical Limits**: Max active PoEs: Unlimited | Monthly paired responses: Custom / Unlimited | Max domains: Unlimited | Data retention: 120 months | Slack integration: Yes | GA4 forwarding: Yes
- **Included Features**:
  - Custom monthly response volume
  - Unlimited Points of Experience and domains
  - Journey mapping and touchpoint design
  - Setup, tuning and rollout included
  - Standing brief of what to fix next
  - All Pro features, Slack and GA4 integrations
  - Custom data retention up to 10 years
  - Part of a CX Review or standalone

### Agency (Coming Soon)
- **Summary**: Multi-client console for agencies and consultancies running WXI for clients.
- **Commercial Terms**: Pricing announced at launch
- **Included Features**:
  - Multi-client workspace management
  - Create and configure client workspaces
  - Clean client ownership handover
  - Centralised client billing

---

## 8. Security, Privacy & Compliance

- **Anonymous by Design**: No names, email addresses, phone numbers, or user accounts are required from survey respondents.
- **Zero Cross-Site Tracking**: No third-party cookies or cross-domain fingerprinting. Session pairing uses first-party `localStorage` scoped strictly to the customer's origin.
- **IP Truncation**: Respondent IP addresses are truncated at ingestion and used only for abuse rate limiting.
- **Encryption**: TLS 1.3 in transit; AES-256 encryption at rest via Supabase (PostgreSQL).
- **Tenant Isolation**: Strict PostgreSQL Row-Level Security (RLS) on every tenant table.
- **GDPR & Singapore PDPA Aligned**: Configurable workspace data retention windows (12 months on Standard, up to 36 months on Pro, up to 10 years on Managed) enforced by an automated daily purge job.
- **Vulnerability Disclosure**: Security issues can be reported to `security@acoustic.sg`.

---

## 9. Frequently Asked Questions

### Is this a replacement for NPS?
Yes. You can run both during a transition, but WXI is designed to replace it. It measures specific moments against the expectations customers brought to them, and it tells you what to fix.

### How do I start?
Sign in, create a Point of Experience, and paste one script tag or print two QR codes. Every self-serve plan includes a 30-day free trial, and most teams are collecting within fifteen minutes.

### Do my customers have to log in or identify themselves?
No. Responses are anonymous by default. Digital journeys are paired within a browser session; physical journeys use paired QR codes. No accounts, no personal data required.

### What are the two readings?
The expectation reading is captured before the experience and the outcome reading after it. Most tools ask people to remember what they expected. WXI asks at the time, which is the only reliable way to measure the gap. The underlying model is the expectancy-disconfirmation theory of satisfaction, established in 1980 (Oliver, 1980); the two-reading measurement it calls for has simply been too expensive to run until now.

### Can I change the questions?
You can change what they ask about (your stay, the checkout, this delivery), not the questions themselves. The fixed wording is the methodology: it is what makes your scores comparable across time, across touchpoints, and eventually against benchmarks.

### How are worst moments grouped?
Every worst-experience answer is read and classified as it arrives. A nightly process groups answers into stable, named worst moments with counts, trends, quotes, and a suggested action. Urgent responses trigger immediate alerts.

### What happens if we reach our monthly response limit?
Capture never stops. Every response continues to be accepted and stored, and urgent-response detection stays active. We send you a message letting you know you have exhausted your plan allowance so you can upgrade to unlock held responses immediately or wait for the next billing period.

### Where does my data live and who can see it?
Your data belongs to your workspace and nobody else’s. Responses are anonymised at collection, retention is 12 months on Standard and configurable up to 36 months on Pro, and nothing about your customers is used to train anything.

### Does asking for expectations first change the answer?
No. Asking before the experience captures the expectation the customer brought with them, rather than a reconstructed memory coloured by the outcome. Measuring both gives you the true gap between promise and reality.

### Does a second question lower completion?
Because the expectation reading takes one second at the start of a journey, drop-off is minimal. Across production deployments, pairing rates typically sit around 75% to 80%.

### How many responses before the report is useful?
Worst moments are grouped once a Point of Experience has collected at least ten responses with written feedback. Urgent alerts fire from the first response.

### How is personal data handled under GDPR and PDPA?
Responses are pseudonymous and contain no personal identifiers by default. Raw IP addresses are truncated at capture, user sessions live in first-party browser storage only, and no customer data is used to train machine learning models. Data retention follows your workspace settings.

### What can I connect it to?
Standard includes instant email alerts and CSV exports. Pro adds incoming webhooks for Slack, Google Analytics 4 event forwarding for completed pairs, and REST API access.