Product
Your First Answer Is Rarely the Report You Want
The first answer is a draft. The order you refine it in decides whether you can trust the last one.
A big question gets you an answer you can't check
When a report builder takes plain English, it's tempting to put everything in the first sentence: every measure, a filter or two, a date range and a ranking. It feels efficient. The cost shows up the first time a number looks wrong. With six clauses in one question, you can't tell which clause produced which part of the answer. The window might be off, a filter might be too tight, or the ranking might have run on the wrong set. All you have is a finished table you can't audit.
A small first question gives you something you can check. You already know roughly who your biggest vendors are and what a normal month looks like, so a plain first cut is easy to hold up against what you know. If it matches, you have a baseline. Every refinement after that is a change you can see against it.
This is the ordinary discipline of anyone validating an analysis. Change one thing at a time, so that when a result moves it has one cause. Keep the change or back it out, then make the next one.
| One enormous question | Small question, one refinement per turn | |
|---|---|---|
| Can you verify the first answer? | Hard — too many clauses to check at once | Yes — against what you already know |
| When a number looks wrong, which clause caused it? | Any of them | The one you just added |
| What changes between answers? | Everything at once | One measure, one window, or one limit |
| Can you tell when it's finished? | Hard to say | Yes — when nothing is left to change |
Scope first: what you're reporting on, cut by what
Two decisions come before anything else: which report you're asking for, and what it's broken down by. Atlas answers sales, inventory, forecasting, replenishment, lost-revenue and stock-availability questions, by product, variant, category, vendor, location, customer, or over time. There are no measure pickers or filter trees to set up first. You say it, and Atlas queries your synced store data and answers with a live table.
Scope comes first because it decides what a row is. A measure or a window changes the values in a row. Scope changes which rows exist. If you get it wrong, every later refinement polishes the wrong table. So make your first turn only the report and the dimension, such as sales by category, and check that the rows are the ones you expected before adding anything else.
Then add one measure or one window per turn
Next, add what you want to see and the period you want to see it over, one per turn. "Add gross profit." adds a column. "Show the last 90 days." changes the window, and the window shows up as a chip you can read at a glance: 30 days → 90 days. Atlas changes only what you asked for and keeps the rest of the report intact. Every filter shows as a chip you can remove, and every change shows where it came from.
You can put both in one sentence, and Atlas will handle it. Splitting them is about reading the result. A measure turn leaves the rows and the existing values alone, so the only new thing is the column. You can check it against the baseline. A window turn changes every value in the table, and that's expected. Do both at once and gross profit comes back different from what you expected, and you can't tell whether margin is the story or the extra 60 days are. One change per turn keeps the answer to that question obvious.
Windows run from the last 7 days to year-over-year, so the same discipline applies at every scale.
| Step | Example turn | What Atlas does |
|---|---|---|
| Scope | "Sales by category" | Builds the report from your live store data. Sales, inventory, forecasting, replenishment, lost-revenue or stock-availability, by product, variant, category, vendor, location, customer, or over time |
| Measure | "Add gross profit" | Adds a gross profit column. Changes only what you asked for and keeps the rest of the report intact |
| Window | "Show the last 90 days" | Changes the time range, shown as 30 days → 90 days |
| Limit | "Just the top 20" | Narrows to the top 20, shown as a chip ("Top 20 by quantity sold") |
| Save | Save as Custom Report | Joins My Reports beside your custom reports, ready to edit, schedule, or export |
Limit and sort last
"Just the top 20." goes after the scope and window are right, never before. A ranking only sees the set that already exists. If you take the top 20 before the window is right, you've ranked the wrong set: a product that only climbs at 90 days never makes a list cut at 30. Cutting a list early also hides the rows you'd otherwise have noticed were wrong.
When you do limit, read the chip. On the report builder's own example it reads "Top 20 by quantity sold," so it tells you what the ranking is by as well as how many rows it keeps. If that isn't the ranking you meant, say so in the next turn.
When to stop asking and save
Stop when the report answers the question you started with and you have nothing left to change. The step-by-step build is what makes that point easy to spot. Every column, window and limit went in one at a time, and you checked each one as it landed.
That's also where asking and saving part ways. Your recently asked reports are one click from running again, which is fine for a question you'll glance at once more. Saving is what turns it into a report like the rest of yours.
Choose Save as Custom Report and it joins My Reports beside your custom reports. From there you can open it in the full report editor to add charts, schedule it for email delivery (daily, weekly, or monthly), and export it as CSV, PDF, or Google Sheets.