Understand your business.
Act on the data.

A CSV, TSV, or Excel export with one row per order or transaction. It needs a date and an amount; everything else is optional and adds detail. We handle the usual export problems — title rows above the headers, duplicate column names, currency symbols, mixed date formats, and unusual encodings — and tell you plainly when something cannot be read.

No, and this is the most important thing about the product. Every figure is computed in Python by a deterministic engine that is tested against datasets with hand-verified answers. The model receives those computed findings and writes them up. Anything it produces is then checked back against them, and a response citing a figure the analysis never produced is discarded rather than shown to you.

Aggregated metrics, segment names such as product and region labels, period labels, the calculation used for each figure, and your question. Individual rows of your file are never sent. With the default local model nothing leaves the machine at all, and the Settings page shows the active provider and exactly what is transmitted to it.

The product keeps working. Analysis, metrics, trends, anomalies, investigations, and reports are all produced without a model; the explanations are written by the engine instead, in plainer prose. The entire test suite runs this way, which is how we know the claim holds.

It will tell you what changed, where, by how much, and what else moved at the same time — and it will label that as coincidence, because co-occurrence is what the data establishes. If sales fell during days when inventory read zero, it reports both facts, says they coincide, and gives you a confidence level. It will not tell you one caused the other, because your spreadsheet cannot show that.

Every insight, anomaly, and answer carries a View evidence control. Opening it shows the underlying figure, the column it came from, the period, the row count, and the named calculation that produced it.

Each workspace is isolated, and access is enforced in the backend on every request rather than hidden in the interface. You can delete a dataset and its analyses at any time, and the files go with it. We make no compliance certification claims: InsightFlow AI holds no SOC 2, ISO 27001, HIPAA, or GDPR certification, and says so rather than implying otherwise.

Seconds for a typical export — the dashboard shows the real stages as they complete rather than an invented percentage. Written explanations take longer when a model runs locally on a CPU, so the evidence is shown as soon as it exists and the prose arrives when it is ready.