Skip to main content

Trust, Methodology & Evidence

Current evidence for AudiAInce® model-generated audience estimates, with sample sizes, exclusions, assumptions, and limitations attached. No accuracy claim is published until enough eligible real-world outcomes exist and a report is approved.

AudiAInce® outputs are directional decision aids, not human respondents or guarantees of market performance.

Directional hit rate
Insufficient data
n < 30
Published status
Collecting evidence
No metric is shown until n >= 30 and approval is complete.
Methodology
trust-v1
AAPOR/ESOMAR-inspired transparency disclosure.

Published Metrics

CTR MAPE
Insufficient data
n < 30
CVR MAPE
Insufficient data
n < 30
ROAS MAPE
Insufficient data
n < 30
Brier score
Insufficient data
n < 30
Expected calibration error
Insufficient data
n < 30
Directional hit rate
Insufficient data
n < 30

Methodology & Limitations

Data sources

  • outcome_records

Assumptions

  • Metrics require matching predicted and actual outcome fields.

Limitations

  • Public evidence is still collecting until approved snapshots reach sufficient sample size.

ESOMAR Transparency FAQ

Who operates this evidence program?

AudiAInce® aggregates eligible campaign outcome records submitted or connected by participating customers.

What methodology is used?

Model-generated audience estimates are compared against later campaign outcomes using predeclared calibration metrics.

What data informs the metrics?

Only outcome records with both predicted and actual values are eligible, after tenant opt-outs and missing-field exclusions.

What is the minimum sample size?

A metric is not published until it has at least 30 eligible records.

Are customers identified?

No account-level content or result is displayed here; this page publishes aggregate metrics only.

Can customers opt out?

Yes. Enterprise policy settings can exclude tenant data from public trust aggregation.

How are exclusions handled?

Exclusions are stored with every snapshot and shown as part of the methodology.

What is MAPE?

Mean absolute percentage error measures average prediction error as a share of actual performance.

What is directional hit rate?

The share of comparison groups where the predicted winner also won in observed campaign outcomes.

What is a Brier score?

A probability calibration score for binary predictions; lower is better.

What is ECE?

Expected calibration error compares predicted confidence bins against observed frequency.

Are confidence intervals shown?

When enough eligible outcome data exists, published metrics include a 95 percent interval.

How often is the page updated?

Snapshots are generated on a controlled cadence and require human approval before publication.

Can a snapshot be retracted?

Yes. Admins can retract published snapshots while preserving audit history.

Is publication automatic?

No. Drafts must be approved and then explicitly published by an admin.

What are the limitations?

The data reflects users who share outcomes, and ad-platform attribution may shift over time.

Is the result cherry-picked?

No. The aggregation uses the selected date window and enforces minimum sample sizes in code.

Are channel breakdowns published?

Only channels with enough eligible data are published.

How is methodology versioned?

Each snapshot stores the methodology version used to compute it.

What happens before there is enough data?

The page shows methodology and an insufficient-data state instead of unsupported claims.