AI Data Enhancement

AI-powered enhancement of company data across 80 million domains and 50+ attributes, built so that more coverage did not mean less trust.

100%
data coverage
Under 2%
hallucination, down from around 80% when we started using AI
80 million
domains, 50+ attributes

Coverage

Before

50%

After

100%

Share of the data filled. Higher is better.

Hallucination

Start

~80%

After

<2%

Share of the AI output that was wrong. Lower is better.

How a value earned its way to customers

3 models
Llama, GPT, Claude

One evaluation set
quality, cost, coverage

Manual validation
humans review samples

Next rollout step
only if under 2%

Rollout in 3 steps: revenue and employee count, then the top 5, then all 50+.

Jun to Nov 2024

Two risks, and what happened

Risk	How it was handled	Result
Value. Wrong or stale data reaching customers.	Quality defined per attribute. Samples validated by hand before each rollout step.	Hallucination from around 80% to under 2%.
Business. Model cost at 80 million domains.	An AI call for everything was too expensive and too slow. Embeddings replaced it.	Budget and time saved. Coverage reached 100%.

My role. Owned the product strategy, the validations, the prioritization and the rules, all based on customer feedback.

With engineering. Engineering and data science built the models, the embeddings and the pipelines. The validation results decided when each rollout step was ready.

Product leader with a data scientist past · Zapier, Clearbit, HubSpot

Product doc: ale-cabrera.com/portfolio/ai-data-enhancement