Using the AI Insights Widget

This guide explains how to use the AI Insights Widget to generate written analysis from your dashboard data, tailored to your audience and tone, and schedulable to run automatically.

Note

The AI Insights Widget requires Snowflake Managed by Adverity as your data warehouse.

To add an AI Insights Widget to your dashboard:

1. Click + Add widget.

2. In the widget picker, select AI Insights.

The widget opens a configuration panel.

Data

Select which widgets or pages to generate insights from.

• In the Widgets tab, select individual widgets to analyze. The widgets are grouped by dashboard page, so you can select widgets from any page, not only the current one. Use the search field to find a widget by name, and the type filters to show only widgets of a particular type. The counter shows how many widgets you have selected.

• In the Pages tab, select whole dashboard pages to generate insights from. Pages are analyzed from screenshots, so every widget on the selected pages is captured visually.

You can base the insights on widgets or on pages, but not both at once — selecting one clears the other.

When you select widgets, the analysis uses the actual underlying data from those widgets — metrics, dimensions, date ranges, and calculated KPIs.

Note

If the selected data is insufficient for meaningful analysis, the widget states what is missing rather than producing unreliable output.

Question

Select the type of analysis you want to generate and optionally provide additional context.

Summarize

Key takeaways from the selected data — what stands out, what’s important.

What changed?

Period-over-period comparison — ranks changes by magnitude, separates improvements from declines, and filters out noise (changes under 5%). Closes with the biggest story of the period.

Note

Requires comparison periods to be configured on the source widgets.

What’s off?

Anomaly detection — looks for unusual spikes or drops, identifies underperformers compared to peers, and rates severity: cosmetic, notable, or urgent. Will not manufacture concerns if nothing looks off.

What should we do?

Delivers 2–3 data-backed recommendations, each structured as: what to do, why it matters, and expected impact. Only recommends actions the data actually supports.

Additional instructions

Optionally enter context to shape the analysis, for example: “Focus on ROAS trends, compare channels, ignore Display campaigns.”

Voice

Configure who is generating the analysis, who it is written for, and how the findings are framed.

Analyst role

Enter or select the role of the person performing the analysis (for example, Performance Marketing Analyst). This provides context so the AI understands the analytical perspective.

Writing for

Select the target audience for the output:

Audience	Description
My team	Direct language, uses industry jargon freely, focuses on what changed and what needs attention.
My manager	Leads with what matters most, provides enough context to understand without granular detail.
Client	Plain language, explains abbreviations, focuses on outcomes and business impact rather than tactics. When selected, a name field appears (auto-filled from the workspace name) so the AI can address insights appropriately.
Executive	30-second read: a single headline, one or two supporting data points, and what happens next.

Tone

Control how the findings are framed, from an even-handed read of the data through to a positives-only summary. Use it to make sure a client-facing dashboard lands the right way.

Tone	How it frames the insight
Neutral (default)	An objective, even-handed read of the data. Reports the good and the bad plainly — nothing is emphasised or held back.
Positive	Keeps every finding but delivers it diplomatically for client-facing dashboards. Issues are framed as opportunities to watch — softened in wording, never hidden or downplayed.
Celebratory	An upbeat, promotional summary that leads with wins and keeps concerns brief. Minor negatives may be dropped, but genuinely serious problems are still surfaced.
Glowing	Praise only — shows exclusively what went well and leaves out every problem, decline, and risk. Choose this only when you deliberately want a positives-only summary.

The selected tone is saved with the widget and reused every time the insight regenerates, including automatic refreshes, so the framing stays consistent over time.

Note

Tone affects how findings are framed and, for Glowing, which findings appear — it never changes the underlying data. Insights stay grounded in your dashboard data whichever tone you choose. Adverity recommends verifying AI-generated content against the underlying data.

Note

Writing for and Tone are independent settings. Writing for sets the language and level of detail for whoever is reading (for example, plain language for a client). Tone sets how positively the findings are framed. They combine: Writing for: Client with Tone: Positive produces plain-language insights delivered diplomatically.

Format

Adjust the level of detail and the output style independently.

Detail

• Headline — the single most important finding only.

• Concise — key takeaways without granular breakdown.

• Thorough — comprehensive coverage of the selected data.

Style

• Rich — formatted text with emphasis on key figures.

• Bullets — a scannable list of findings.

• Narrative — continuous prose, suitable for client-facing reports.

• Table — structured tabular output.

Schedule

Summaries can be generated automatically at set intervals so that your dashboard always contains up-to-date written analysis without manual effort.

Configure the following settings:

Frequency

Daily, weekly, or monthly.

Time

The time of day at which to generate the analysis.

Timezone

The timezone used for the scheduled time.

Run on (weekly only)

Select the days of the week on which to run.

Note

Regardless of any schedule, an AI Insights summary also regenerates automatically when the data it summarizes changes — for example, when you edit the date range or dimension filters of a summarized widget, or change a dashboard permanent filter that affects the selected data. Changes that do not affect the data, such as adding or removing a metric or dimension, do not trigger a regeneration.

Click Apply to save the widget configuration.

Editing the widget and its schedule

To edit the widget configuration or update its schedule, hover over the widget on your dashboard and click Widget actions, then select Edit widget. Make your changes and click Apply.

To remove the schedule, open the widget for editing, clear the Frequency field, and click Apply.