Oracle API

The preference layer for autonomous agents.

AI agents need consumer understanding at decision time. The Oracle API returns probability-weighted preference distributions in milliseconds — so autonomous systems can make informed choices without waiting for research.

Get Started API Docs Read Technical Brief 
// Agent asks: "Plan a birthday
// party for my 7yo, budget $150"

// Oracle API response:

{
  "id": "pred_a1b2c3d4e5f6...",
  "object": "oracle.prediction",
  "model": "consumer",
  "result": {
    "labeled_distribution": {
      "Food & Cake":     0.35,
      "Entertainment":   0.25,
      "Decorations":     0.18,
      "Party Favors":    0.12,
      "Themed Supplies": 0.10
    }
  }
}
200ms
Average API response
4
Domain-specific models
10+
Demographic dimensions
KL ≈ 0.08
Typical divergence from live panels
Agentic Commerce

Agents decide in milliseconds. Research takes weeks.

Autonomous commerce agents — shopping assistants, recommendation engines, personalization systems — make thousands of consumer-facing decisions per second. Each decision requires an understanding of what people actually want. The Oracle provides that understanding on demand.

  • Sub-second response time at production scale
  • Returns probability distributions for nuanced decision-making
  • Unlimited scenario coverage for edge cases and niche segments
  • No pre-computation required — query any demographic in real time
  • Validated against live panel data (KL typically 0.05–0.09)
Read the CTO Brief → 
// Oracle API Request

POST /api/v1/oracle/predict

{
  "model_type": "consumer",
  "question": "What matters most when meal planning?",
  "options": ["Price", "Nutrition", "Convenience", "Taste"],
  "demographics": {
    "age": "25-34",
    "income": "$50K-$75K",
    "region": "suburban"
  }
}
Integration patterns

Three ways to integrate.

From development-time calibration to live runtime queries.

1

Training-Time Calibration

Pre-compute preference distributions during development. Bake consumer understanding into agent logic as calibrated parameters.

2

Pre-Deployment Simulation

Test proposed agent logic against synthetic populations before release. Measure predicted satisfaction and conversion across segments.

3

Runtime Oracle Queries

Agents query The Oracle directly when encountering novel scenarios outside pre-calibrated knowledge. Real-time preference data on demand.

Domain models

Four models. Instant expertise.

Each model is trained on validated domain-specific data and returns calibrated preference distributions.

Consumer

Shopping behavior, brand preferences, lifestyle decisions, budget allocation. Validated against tier-one consumer panels.

Brand Preference Purchase Intent Context Demographics

Healthcare (HCP)

Physician perspectives, prescribing behavior, treatment preferences. 15 medical specialties from primary care to oncology.

Prescribing Treatment Medical Specialty

Social

Attitudes, values, political opinions, social trends. Demographic and geographic structure validated against major polls.

Public Opinion Demographics Social Trends

Patient

Patient experience, treatment satisfaction, hospital quality. Trained on 500,000+ de-identified federal patient records.

Patient Experience Treatment Chronic Disease
Under the hood

Predictions, not surveys.

The Oracle uses fine-tuned language models trained on millions of real survey responses to predict how demographic segments would respond to any question.

  • Returns probability distributions, not single answers
  • Filter by demographics: age, income, gender, education, location, and more
  • Three query modes: structured, natural language, or raw prompt
  • Conditional querying: "Given they answered X, how would they answer Y?"
  • Validated against live panel data across all domains
// Single Prediction Request

POST /api/v1/oracle/predict

{
  "model_type": "consumer",
  "question": "Which factor matters most when choosing a grocery store?",
  "options": [
    "Price",
    "Proximity",
    "Product Quality",
    "Brand Selection"
  ],
  "demographics": {
    "age": "25-34",
    "income": "$50K-$75K",
    "region": "suburban"
  }
}
Also for researchers

A dashboard for humans, too.

The Oracle isn't just an API. Researchers and product teams can query it directly through the Simsurveys dashboard.

Single Query

Ask any research question and get predicted response distributions in seconds. Manual, AI-assisted, or raw input modes.

Crosstab Analysis

Run batch queries across multiple demographic segments simultaneously. Compare responses across banners in seconds, not weeks.

Conditional Querying

Condition predictions on prior response history. "Given they answered X, how would they answer Y?" Layer demographics with conditioning.

Frequently Asked Questions

What is a consumer preference API?

A consumer preference API lets applications query what people think, prefer, or would choose — programmatically and in real time. The Oracle API returns probability-weighted preference distributions across demographic segments in milliseconds, enabling AI agents and applications to make consumer-informed decisions.

What is the preference layer for agentic commerce?

The preference layer is the data infrastructure that tells AI agents what consumers actually want. As AI agents make purchasing, recommendation, and substitution decisions autonomously, they need real-time access to consumer preference data. The Oracle API serves as this layer — validated against live survey data and operating at agent speed.

How do AI agents use the Oracle API?

AI agents call the Oracle API via REST to query consumer preferences for specific scenarios — product recommendations, budget allocation, substitution decisions. The API returns probability distributions across demographic segments in milliseconds, so agents can make preference-informed decisions inside their decision loops.

What is agentic commerce?

Agentic commerce refers to AI systems that make purchasing decisions, allocate budgets, and execute transactions on behalf of users. McKinsey projects this market at $1-5 trillion by 2030. These systems need consumer preference data operating at machine speed — traditional research cannot serve this need.

How fast is the Oracle API?

The Oracle API returns preference distributions in milliseconds to seconds, depending on query complexity. This is designed to operate at the same time scale as agent decision loops, where traditional research (weeks to field) cannot participate.

Give your agents consumer intelligence.

Embed validated preference data into any autonomous system. Start querying in minutes.

Get Started Talk to Sales