Ask why. Get the answerwith numbers.

It cannot. A file you drop into the chat is turned into plain text on our side before the model sees anything — never bytes, never an image — and that text is anonymised on the way, so your real location and supplier names are replaced before it leaves and restored in the answer. The file's contents are given to the model as data, never as instructions, so an invoice that happens to contain a sentence like 'approve everything' is read as text and not obeyed. Writing is a separate, human step: the AI can only show you a card that lists what would be written and what would be skipped, and on approval the figures are recomputed on the server from the file itself. Where two locations could match one spreadsheet row, it refuses rather than guesses, and days you already have shift data for are skipped, with the card saying so before you press. A scan with no text layer is marked unread rather than half-read.

Yes. marql AI reads from the same data layer as every other marql module: your POS, ERP, or accounting integration. Nothing is retyped and nothing is synthetic. Where a source has no usable API, its CSV or Excel export is mapped into the same canonical model rather than living beside it — so marql AI still answers from one set of numbers. Answers are grounded in your real transactions, updated on your source system's cadence.

Revenue and margin questions ('Why did margin drop this week?'), store comparisons ('Which store is performing below its usual Tuesday pace?'), plan tracking ('Will we close the plan this month?'), product questions ('Which SKUs are killing margin?'), and operational questions linked to your Loyalty and Dashboard data. marql AI draws context from all connected modules.

Every morning marql AI runs a structured analysis of your overnight and prior-day data: revenue vs plan, margin vs prior period, store-level anomalies, and product-level exceptions. It surfaces the top three to five items that need attention, each with the action it suggests.

The period is carried for you: whatever range your last question used is passed into the next one, so a follow-up stays on the same days instead of quietly falling back to the last seven. A store carries through the thread, and pinning it with @ makes that explicit and durable. Currency and time zone are your own — 'today' is computed in each location's own time zone rather than in UTC, which is what stops a late-evening question from reading yesterday. Weeks run Monday to Sunday everywhere; there is no custom fiscal calendar.

Yes, through Proactive Patterns. marql AI monitors recurring trends across your data: a 3-week revenue decline in a specific daypart, a consistent Monday underperformance, a margin compression appearing in one category. It surfaces these patterns in the Morning Briefing before you think to ask.

ChatGPT has no access to your real data. marql AI runs against your actual POS and accounting records through a structured query layer: the figures are computed by deterministic queries and the model writes the sentence around them rather than inventing the number. Once the answer is written, its numeric claims are checked back against what those queries returned, and the verified ones are marked. It also knows your store names, your plan targets, your product catalogue and your loyalty segments. That is a different kind of answer from a generated one — which is also why the notice above still asks you to check anything you are about to act on.