Consumer Intent Graph Methodology
Version 1.0.0 · Effective September 21, 2026
Consumer Intent Graph measures observed digital consumer intent and wallet behavior extracted from naturally occurring public signals.
Return to Consumer Intent Pulse · Machine-readable methodology
Construct boundary
Purchase intent, completed purchases, delay, cancellation, affordability constraints, trade-down, deal-seeking, discretionary change, and substitution.
Raw online observations cannot establish population prevalence and do not replace surveys, PCE, retail sales, or company reporting.
Observation pipeline
Structured outputs preserve the consumer-facing brand or retailer and connect it to its parent company and ticker. Duplicate clusters count once in aggregation.
Factors remain independent
| Factor | What it captures |
|---|---|
| Wallet Stress | Affordability, debt, income, or budget pressure affecting choice. |
| Purchase Intent | Intent or reported action to purchase. |
| Purchase Deferral | Delay or cancellation tied to price, uncertainty, promotion timing, or constraints. |
| Trade-Down | Movement toward a lower-cost product, brand, retailer, experience, or category. |
| Deal Sensitivity | Emphasis on discounts, coupons, clearance, or promotion timing. |
| Discretionary Appetite | Expansion or reduction of non-essential spending. |
| Brand Desire | Desire for a named brand, independent of completed purchase. |
| Substitution | Movement from one product, brand, retailer, or category to another. |
No composite is published in v1. Any future aggregate pulse must earn its weights through documented predictive or explanatory validation.
Reading and publication gates
A factor's directional evidence balance is 100 × (supporting confidence mass − opposing confidence mass) ÷ total confidence mass. It summarizes classified observations in the covered sample; it is not a population percentage.
- Minimum 30 unique observations
- Minimum 3 independent sources
- Minimum 5 collection days
- No single source above 60% of the factor sample
If any gate fails, the public status is Insufficient evidence. Confidence describes classification certainty and coverage—not the probability of a financial outcome.
Source and quality policy
Collection is limited to public material where access and terms permit. Source type and permitted attribution are retained. V1 does not invent platform weights: observation confidence is explicit, duplicate clusters count once, and likely spam or promotion is excluded from published aggregation.
Safeguards cover duplicate reposts, bots, brigading, affiliate spam, promotional campaigns, review bombing, and repeated identical language. Licensed sources may be added later without changing the ontology.
Longitudinal versioning
Historical observations and releases are never overwritten. Every release records its methodology and model versions so a past reading can be reconstructed and distinguished from later methods.
Validation
The research program tests Consumer Intent at T0 against later macro, company, and market outcomes. Tests may include correlation, lead/lag, hit rate, stability, category specificity, and confidence intervals where meaningful. Relationships are measured empirically; causality is not assumed.
Known limitations
Online population and platform bias, bot activity, coordinated campaigns, brand fandom, viral events, news-cycle contamination, geographic uncertainty, uneven category coverage, API changes, deletion, and survivorship can all distort the observed sample.
Consumer Intent Graph is an exmxc research system. It does not provide investment recommendations.