Product
Ask it, save it, or schedule it: three homes for one question
Where a question should live has almost nothing to do with how hard it is — and almost everything to do with how often the answer is needed, and by whom.
Three surfaces, one question — the difference is recurrence, not difficulty
Most people pick where to ask by how hard the question feels. A quick one goes in the chat; a serious one gets built as a report. That instinct sorts questions by difficulty, and difficulty is the wrong axis — the same engine answers both.
The axis that matters is recurrence: how often the answer is needed, and by whom. Assisty gives a question three homes, and they line up with that axis rather than with effort.
The AI Assistant is where Atlas answers in plain English from your live store data — stock counts, sales performance, what to reorder, and why — then hands you the action that follows. The AI Report Builder is where you type what you want to know and Atlas builds the report from that same live data; you refine it by talking and save it next to the rest of your reports. Scheduled Reports is where a report you already have runs on your cadence and gets delivered where you work: email, Google Sheets, Google Drive, or a Shopify Flow.
All three read the same live store data, and a schedule captures its report exactly — breakdown, time range, filters, columns. So you're not picking a more or less capable answer. You're picking recurrence and audience.
| AI Assistant | AI Report Builder → Save as Custom Report | Scheduled Reports | |
|---|---|---|---|
| How often you need the answer | Once, or once in a while, while you're still investigating | Again — and probably with a tweak each time | On a clock you set: daily, weekly, or monthly |
| Who reads it | You, right now, inside the conversation | You again — from My Reports, next time you want it | People who may never open Assisty — multiple recipients at once |
| Where the answer lands | In the chat; bigger answers arrive as sortable, expandable tables you can pop out and export to CSV | In My Reports, beside your custom reports | Email, Google Sheets, Google Drive, or a Shopify Flow |
| What it costs when you pick wrong | Nothing — until you find yourself typing the same question again | One more saved thing to scan in a list you'll never re-run | A standing send to someone's inbox for a question asked exactly once |
Ask once, save when you'll refine it, schedule when someone else is waiting
Three questions, in order. Stop at the first "no".
Will I need this again? If not, ask Atlas and take the answer — or take the action. When Atlas finds something worth doing, it offers it as a one-tap chip: add items to a purchase order, expedite what's about to run out, cap a quantity to your max stock. A question asked once should end in an answer or an action, not an artifact.
Will I re-run it with different filters, or do I want a chart on it? Then save it. This is the boundary BI has always drawn between ad hoc and standard reporting: ad hoc work is created on the fly for one specific, often one-time question and prizes flexibility; standard reporting follows a definition that stays put and prizes consistency. Ad hoc is investigating; standard is monitoring.
Does someone who isn't me need it on a clock, somewhere that isn't Assisty? Then schedule it. Note that the test has two clauses — a cadence and an audience. If both hold, a schedule earns its keep. If only one does, a saved report you open when you want it is usually enough.
The rule exists because both failure modes are quiet ones. Scheduling a question asked exactly once is how report sprawl starts: creation is cheap, management isn't, and old reports stay live by default. The opposite failure is quieter still — re-typing the same question every Monday, retrieving from scratch what a saved report would already have been holding for you.
A question graduates; it doesn't get rebuilt
Getting the rule wrong is cheap because promotion is a real product path, not a polite word for starting over. Each step captures what came before rather than replacing it.
The one move the path doesn't offer is a jump from a chat answer straight to a schedule. Saving sits in between, and that's a property of the rule rather than a gap in it: saving is where a question stops being exploratory and becomes a definition you're willing to stand behind, and a schedule is only ever as trustworthy as the definition underneath it. A chat answer you liked once is not yet that.
The side branch matters as much as the main line, for the opposite reason. A promoted question isn't a frozen one — refinement stays conversational. "Add gross profit." "Show the last 90 days." "Just the top 20." Atlas changes only what you asked for and keeps the rest of the report intact, every filter showing as a chip you can remove. Promotion adds a home; it doesn't close the others.
| Step | What carries forward |
|---|---|
| Ask → Save | An AI-built report isn't a dead end — "Save as Custom Report" and it joins My Reports beside your custom reports |
| Save → Schedule | Quick Actions → Schedule Report captures that report exactly — breakdown, time range, filters, columns — including the reports you built by asking Atlas |
| After scheduling | It's still a report: open it in the full report editor to add charts, and export it as CSV, PDF, or Google Sheets like any other report |
| Every run | Kept in history — with the file it sent, re-downloadable any time |
What each promotion costs you — and how to demote without deleting
Asking and saving are nearly free. A schedule is not, because it's the only one of the three that keeps acting while nobody is thinking about it — a standing promise to someone else's inbox, made once and kept indefinitely.
Two settings decide how heavily that promise sits. Send when empty is the setting for quiet weeks: a schedule with nothing to report can stay silent or keep announcing itself, and which you want depends on whether the recipient is watching for exceptions or reconciling a routine. Reusable schedule profiles keep your favorite cadences one click away, which makes the cheap half of this cheaper still — a cadence becomes a setting you reach for, not a decision you relitigate.
When the schedule is too much but the report isn't, shrink the schedule rather than the question behind it: a schedule's columns and filters can be edited without touching the source report, so a daily send stays narrow while the report someone opens on demand stays wide. Run now sends a test delivery the moment you save, and every run is kept in history with the file it sent, re-downloadable any time.
None of that touches the question. That's what keeping three homes distinct buys you — a question can change how often and how widely it arrives without ever being asked again from scratch.