Evidence, not AI promises.
Every figure we publish states what was measured, over what period, in which deployments, how it was calculated and what it does not cover.
Human–AI evaluation agreement
Human–AI evaluation agreement
Every AI evaluation can be reviewed by a trained human reviewer who confirms, rejects or overrides it with a recorded rationale. Across completed human reviews, reviewers agreed with the AI-generated evaluation in 99.3% of cases. Reviewers could see the AI assessment when making their decision.
- What is measured
- The proportion of completed human reviews in which the reviewer agreed with the AI-generated evaluation.
- Measurement period
- Live deployments since August 2025, cumulative to the date confirmed (June 2026).
- Deployment scope
- Completed human reviews across live Better Outcomes deployments in UK customer operations.
- Calculation method
- Completed reviews where the reviewer confirmed the AI evaluation, divided by all completed reviews.
- Limitations
- Reviewers could see the AI assessment when deciding, so this is an agreement rate, not an independent or blinded validation. It covers reviews that were completed, not every AI evaluation produced.
Interactions analysed
Interaction coverage
Manual QA typically reviews a small sample of interactions. Better Outcomes assesses connected interactions at scale, in the context of the CRM and journey evidence around them, so organisations move from a sample to substantially broader coverage of what customers actually experience.
- What is measured
- Customer interactions analysed by Better Outcomes across live deployments.
- Measurement period
- Cumulative since the first live deployment (August 2025); figure confirmed June 2026.
- Deployment scope
- Connected, in-scope interactions across live deployments. Coverage within each deployment depends on the channels and systems connected.
- Calculation method
- Count of interactions ingested and assessed by the platform.
- Limitations
- Does not represent every interaction within a customer organisation, unsupported channels or data that was not connected.
Reduction in manual QA effort
Operational efficiency
With AI evaluations and evidence prepared in advance, reviewers spend their time on confirmation, exceptions and coaching rather than listening to whole interactions.
- What is measured
- Reviewer hours spent per 1,000 interactions.
- Measurement period
- Before and after deployment at each organisation.
- Deployment scope
- Money Wellness and LuvCars.
- Calculation method
- Reviewer hours per 1,000 interactions after deployment compared with the pre-deployment baseline.
- Limitations
- “Up to” denotes the largest reduction observed; results vary by deployment and review workflow.
Running inside live UK customer operations.
Working with organisations focused on better customer outcomes
- LuvCars
- Money Wellness
- Think Money Limited
- Ocean Finance
See what Better Outcomes could identify across your customer journeys.
Bring one representative customer journey and we'll show how Better Outcomes could assess it across interactions, CRM evidence and customer outcomes. No customer data is required for the first session.