AI Summaries & Q&A
Ask a report a question in plain language, or have it write the summary for you.
Summarize a Visual
AI Summarize on a visual's menu produces a written description of what that chart or table is showing — the shape of the data, what stands out, and where the notable values sit. It reads the visual's current result, so it reflects whatever filters are applied.
Ask a Question of a Report
Ask in plain language and get an answer drawn from the report's own data. The conversation is kept, so you can return to a report and pick up where the questioning left off rather than starting over.
Your last 50 conversations are listed under History on the Ask PersivX page, each titled from its first question. To remove one, hover over it and click the bin icon, then Yes, delete.
Agentic Answers
Where a question needs more than one look at the data, the agentic mode works through it in steps — querying, reading the result, and querying again — and streams its reasoning as it goes, so you can see how it reached an answer rather than just being handed one.
Deep Reasoning Mode
When deep reasoning is switched on for the instance, the agent investigates a question the way an analyst would. It first writes a plan, a few sub-questions and competing explanations, and the thinking steps show each explanation being marked supported, refuted or inconclusive as evidence comes in. It answers once every one of them is settled.
- "Why" questions. It splits a change between two periods by region, product, customer or other fields, and names the ones that drove it, each with its share of the change.
- Trends. It flags a first or last period that has much less data than usual, so a part-month or part-year isn't reported as a fall.
- Unusual values. It points out customers, products or other groups far from the typical value.
- Every figure is checked. Before an answer is shown, each number in it is matched against the query results. A figure that can't be traced is sent back to be recalculated. If it still can't be traced, it is listed in the answer's caveats.
Answers You Can See
A deep reasoning answer opens with the answer in one sentence and the two to four figures worth quoting, each with its change. Each finding that follows has a chart where the data has a shape: a line for a trend, a sorted bar for a ranking, a stacked bar or pie for a share of a whole, and a waterfall for what drove a change between two periods. Every chart is drawn from the data the agent queried, never typed in, and names its source underneath. A part-year or part-month is faded and pinned on the chart. Ask for something "as a chart" and you will get one. Three suggested follow-up questions sit under the answer; click one to ask it.
Figures follow the instance's number format, so an Indian-format instance shows ₹45.16 Cr in the text, the same as on the chart's axis.
The instance can also allow it to look past the report into the report's underlying datasources, when the charts are too summarised to answer. It only reads datasources you have access to, and your data governance rules apply to it exactly as they do to visuals.
Across Every Report
Cross-report Q&A takes a question when you don't know which report holds the answer. It routes to the right report first, then analyses within it.
Executive Deep Summaries
The most involved of these produces a full executive summary as a presentation, with generated observations and rendered visuals. See Executive Deep Summaries.
Guiding the Agent
Where an AI step runs as part of a scheduled delivery, you can supply written instructions for the agent — focus areas, boundaries for exploration, exceptions to ignore, and business logic it should take into account.
Comments on Rows
Not AI, but adjacent, and often used alongside it: rows in a table can carry comments, so the human interpretation of a number lives next to the number rather than in a separate thread. See Working with Tables.
