You askedHow do I know this is right?
Show your workingEvery answer arrives with its receipts.
An analyst who cannot explain a number does not get trusted twice. Datatopple is held to the same standard: every answer shows what it read, what it assumed and how it calculated, and says plainly when the data is not good enough.
Which product sold the most in Salem last month?
- 1
The sentence
Cotton bedsheets, by a long way.
The plain answer comes first, in words a manager would use.
- 2
The number
₹4.2L, 51.2% of Salem revenue
The figure that backs the sentence, with the comparison that makes it mean something.
- 3
The source
Sales_Oct.xlsx, sheet 'Invoices', rows 2-4,812
File, sheet or table, and the exact row range read. Click through to see the rows.
- 4
The filters
branch = Salem · 1-31 Oct · status <> 'cancelled'
Every assumption it made about your words, written out so you can disagree with one.
- 5
The query
SELECT product_name, SUM(net_amount) ...
The actual statement that ran. Copy it into your own tool and you will get the same result.
- 6
The freshness
Synced 14 minutes ago
When the source was last read, so an old number never passes for a new one.
When the data is not good enough, the answer says so first.
Flags sit above the number, not in a footnote. A scheduled email with a partial month leads with the warning, so nobody forwards a half-month total as the real one.
Partial period
This month's data is incomplete after the 20th. The total covers 20 of 31 days.
Missing values
212 invoices have no branch recorded. They are excluded from the branch breakdown and counted separately.
Cannot answer
Your data has no cost column, so I cannot calculate margin. Connect the cost sheet and ask again.
Ambiguous term
"Last quarter" read as July to September, your financial Q2. Switch to calendar Q3?
We would rather tell you than have you find out.
Datatopple is accurate when your data is clear. These are the four situations where it is most likely to need a correction, and what it does about each.
It can misread a column name
If "Amt" means gross in one sheet and net in another, the first answer may pick the wrong one. The mapping is shown on every answer, and once you correct it the correction sticks.
It can take your words too literally
"Customers" might mean billing accounts to finance and outlets to sales. Datatopple states the interpretation it chose, and you can save a definition for your workspace.
It only knows what you connected
Ask about returns without connecting credit notes and it will say the data is missing rather than estimate. If the answer depends on something it cannot see, it will say so.
Garbage in is still garbage out
Duplicate rows, typos in branch names and manual overrides in a workbook flow straight into the answer. Datatopple flags obvious duplicates and outliers, but it cannot know a number was typed wrong.
Read, answered, and left exactly as it was.
Encrypted in transit and at rest
TLS 1.2+ on every connection; AES-256 for anything stored, including cached results.
Never used for training
Your rows, questions and answers are not used to train models for you or for anyone else.
Minimum necessary context
Models receive schema and the rows needed for the question, not your whole workbook.
Deleted on request
Disconnect a source and its cache is purged. Close a workspace and everything is deleted within 30 days.
Type it. Get the answer.
No dashboard to build, no analyst to brief, no 40-tab workbook to open first. Connect a source and your first answer lands in about two minutes.
Want to see it against your own workbook first? Contact us
- Read-only connections
- Source and rows on every answer
- Never used to train models