AI Features

Dataset metric suggestions

When you create a new dataset, Lighthouse automatically runs an AI job that reads your column structure and proposes metrics worth monitoring. The job runs in the background — suggestions appear on the dataset detail page when it completes.

What the AI looks at

The AI reads the dataset snapshot including column names, types, and the roles you assigned (metric, filter, segment, datetime). It uses this structure to propose metrics that are commonly useful for tables with similar shapes — for example, a transactions table with an amount column and a created_at column will receive suggestions like "Total revenue per day" or "Transaction volume last 3 hours."

Working with suggestions

Each suggestion card shows a title and a description of what to monitor and why. From the card you can:

• Create Metric — Opens the metric creation form pre-filled with the suggestion's recommended configuration. You can edit anything before saving.
• Reject — Removes the suggestion from the list if it's not relevant to your use case.

Metric AI suggestions

For metrics that are already live, Lighthouse can run an AI analysis that suggests improvements — better thresholds, new segment dimensions to add, or changes to the time window that would reduce false positives.

Triggering a suggestion run

Start a run from the metrics table or from the action menu on a metric's detail page. Lighthouse runs the analysis and surfaces the results as suggestion items.

What the AI looks at

The AI reads the metric's full configuration — aggregation, time window, threshold settings, and segment configuration — along with the last 30 days of metric run history. It uses this to identify patterns: thresholds that are too tight or too loose, segments that consistently fire but are never acknowledged, or time windows that may be missing late-arriving data.

Acting on suggestions

Each suggestion item has three actions:

• Implement — Opens the Metric Assistant with the suggestion pre-loaded as context. Review the recommended changes and apply them interactively — nothing is auto-applied.
• Resolve — Marks the suggestion as addressed without implementing it — use this if you've already made the change manually.
• Reject — Removes the suggestion as not applicable.

Metric Assistant

The Metric Assistant is a chat interface that appears in the side panel when you're creating or editing a Business metric. It lets you describe what you want to monitor in plain English and will configure the metric form for you.

What it can do

• Configure the metric — Set the aggregation, column, time window, filters, and segments based on your description.
• Run test queries — Execute a sample query against your warehouse to preview what the metric would return. You approve each query before it runs.
• Explain options — Answer questions about what each configuration field does and why you might choose one option over another.

Example prompts

You can describe metrics conversationally. For example: