Technical Whitepaper

Methodology & Scientific Framework

How echotest constructs, simulates, and calibrates synthetic audience responses across 195 markets using census-anchored archetypes and multi-round debate dynamics.

Version 2.0 · April 2026 · echotest Research

1. The 10-Layer Intelligence Stack

Every echotest simulation passes through a 10-layer pipeline. Each layer adds a dimension of behavioral realism that generic AI tools cannot replicate. The layers are cumulative — each builds on the output of the previous one.

Layers 1–6 are pre-computed for all 2.5 million archetypes. Layers 7–10 are applied at simulation time based on the specific content, product, or policy being tested. This means every simulation is contextually modulated — the same archetype reacts differently to a luxury product vs. a budget SaaS tool.

2. Archetype Construction Methodology

2.1 Population-Proportional Sampling

echotest maintains a database of 2,521,382 synthetic archetypes distributed across 195 UN member states. Each country's archetype count is proportional to its population, ensuring that India (1.4B people) has proportionally more archetypes than Luxembourg (650K).

Within each country, archetypes are stratified across 5 demographic dimensions:

• Age group — 7 buckets (13-17 through 65+), weighted by national age distribution
• Gender — Male/Female, weighted by national ratio
• Income bracket — 5 tiers (low through high), calibrated to GDP per capita
• Education level — 6 levels (primary through doctorate)
• Urban/Rural — Weighted by national urbanization rate

Source: UN World Population Prospects 2024, World Bank World Development Indicators

2.1b Why we go down to the neighborhood

A country average hides more than it reveals. The person browsing on Palm Jumeirah doesn't buy like the person in Deira. A Riyadh executive doesn't shop like a Hofuf factory worker. A Mumbai Bandra resident doesn't make the same decisions as someone in Dharavi. Treating them all as “the average Indian / Saudi / Emirati consumer” throws away the very variation that decides whether your product lands or stalls.

In our key markets we build separate audiences for distinct neighborhoods — their own income mix, religious composition, language, brand reach, and cultural texture. So when you test a luxury launch, the panel that responds is shaped by the streets it would actually sell on, not a country-wide mean that smooths the signal away.

2.2 Cultural Dimensions

Each archetype inherits cultural dimensions from its country, sourced from the Hofstede Institute's 6-Dimension model and the World Values Survey. These dimensions shape how archetypes process information, respond to authority, handle uncertainty, and make purchasing decisions.

Source: Hofstede Insights (hofstede-insights.com), World Values Survey Wave 7 (2017-2022)

2.3 Transaction DNA

Beyond demographics and culture, each archetype carries 35+ behavioral fields that model purchasing psychology. These are computed from the intersection of income, education, cultural dimensions, and sector-specific modifiers.

2.4 OCEAN Personality (Big Five)

Each archetype's Big Five personality profile is computed mathematically from Hofstede cultural dimensions + Transaction DNA fields. Zero LLM calls are used in personality computation — it is a pure mathematical derivation that runs deterministically.

3. Simulation Pipeline

When a user submits content for simulation, the following pipeline executes:

Total simulation time for 200 agents: ~60-90 seconds. For 2,000 agents: ~5-8 minutes. Results are streamed in real-time via WebSocket as each stage completes.

4. Calibration Framework

echotest does not claim oracle-level accuracy. Instead, every prediction is accompanied by explicit confidence metrics, reliability grades, and calibration level indicators. The system is designed to be honestly uncertain and to improve over time.

4.1 Confidence Scoring

Confidence scores (0–100%) are computed from four weighted components:

4.2 Reliability Grades

Every report receives a reliability grade based on sample size, using Wilson score confidence intervals:

Based on Wilson score interval for binomial proportions (Wilson, 1927). Margin of error computed at 95% confidence level.

4.3 Calibration Levels (The Learning Loop)

The system improves per customer through a closed-loop feedback mechanism. Users who feed real campaign results back receive progressively more accurate predictions.

4.4 Lifecycle Calibration Loop (How it works in practice)

The Lifecycle Simulator (Engine 5) uses a market-bucket calibration system that tracks systematic prediction deltas per (country, category, dampening mode). Each completed loop improves the bucket's rolling stats and sharpens future predictions for the same market.

5. Validation: Prediction vs. Reality

echotest predictions are continuously backtested against actual campaign outcomes. When users submit real-world results through the Decision Intelligence system, the comparison engine computes:

• Sentiment accuracy — Predicted sentiment distribution vs. actual (mean absolute error)
• Conversion accuracy — Predicted conversion rate vs. actual (relative error)
• Segment delta analysis — Per-segment predicted vs. actual, identifying systematic biases
• Country delta analysis — Per-country prediction accuracy
• Bias detection — Systematic over/underestimates flagged across 3+ campaigns

6. Comparison to Traditional Research

echotest is not a replacement for all market research. It is a complement that excels at speed, breadth, and early-stage hypothesis testing. Here is an honest comparison:

When to use echotest: Before you commit budget. Test 10 ideas in the time it takes to brief one focus group. Narrow your options, then validate the winner with traditional research if the stakes justify it.

7. Limitations & Honest Uncertainty

Scientific honesty is a core principle of echotest. We explicitly acknowledge the following limitations:

Every echotest report includes a disclaimer: “Results are directional indicators based on synthetic agent responses. Not a statistical prediction. Real-world outcomes may differ. Use results as one input among many in your decision-making process.”

8. Data Sources & Citations