An AI dashboard is a live reporting page that pulls numbers straight from the systems you run on, such as QuickBooks, Shopify and Amazon Seller Central, and lets you ask follow-up questions about what you see. Instead of exporting three reports every Monday and pasting them into a spreadsheet, the numbers are already there, current, and laid out the same way every time.
We build daily sales reports and dashboards across several e-commerce brands selling on Shopify and Amazon, and financial views from QuickBooks. This guide covers how the data gets in, how to lay the page out so it reads quickly, and where AI actually helps versus where it gets in the way.
What makes a dashboard "AI"
There are two separate things people mean.
The first is AI in the build. A non-developer can describe the dashboard they want, and Claude writes the code, connects the data sources and deploys the page. We have built internal dashboards this way and deployed them to Cloudflare. That lowers the cost of a custom dashboard from a development project to an afternoon.
The second is AI in the use. With the underlying systems connected to Claude through MCP connectors, you can look at the dashboard, see that Amazon revenue dropped on Tuesday, and ask "why?" Claude can then pull the order data, ad spend and inventory levels to look for the cause. The dashboard tells you what happened. The AI helps you find out why.
The numbers on the dashboard itself should not be AI-generated. They should come straight from the source systems by fixed queries, so the same question always gives the same number. AI is for building the page and investigating what it shows, not for doing the arithmetic.
Getting the data in
Each platform exposes its data through an API. The practical questions are what to pull, how often, and what quirks to plan for.
QuickBooks Online
QuickBooks exposes its standard reports, including Profit and Loss and Balance Sheet, through its accounting API, along with the underlying transactions. For a dashboard, the common pulls are revenue, gross margin, operating expenses, cash, receivables and payables.
What to watch:
- Cash versus accrual. Pick one basis and label it. Mixing them is the most common reason a dashboard and the accountant disagree.
- Timing. Books are often not closed until a few weeks after month end. Mark the current and prior month as preliminary.
- Categories. The dashboard is only as clean as the chart of accounts. If similar expenses land in different accounts, fix that at the source.
Shopify
Shopify's Admin API provides orders, refunds, products, customers and inventory. For a daily sales view, pull orders by created date with gross sales, discounts, refunds, shipping and taxes broken out.
What to watch:
- Define revenue once. Gross sales, net sales after discounts and returns, and total including shipping and tax are all different numbers. Write the definition on the dashboard.
- Refunds land later. A refund issued today may belong to an order from last month. Decide whether refunds hit the day they happen or the day of the original order, and be consistent.
- Time zone. Use the store's time zone so the dashboard matches what the team sees in Shopify.
Amazon Seller Central
Amazon's Selling Partner API offers reports including the Sales and Traffic report, which covers ordered product sales, units, page views and sessions by date and by product. Order and settlement data are available separately, and Amazon Ads data comes from the Ads API.
What to watch:
- Lag. Amazon's sales and traffic data typically arrives with a delay of at least a day and can be revised afterward. Re-pull the last several days on each refresh rather than only yesterday.
- Ordered sales versus settled cash. What customers ordered and what Amazon pays you after fees and refunds are different. Show ordered sales for momentum and settlement data for cash.
- Multiple accounts. If you run several seller accounts, keep each connected and reported separately, then roll up. Blending them at the source makes problems hard to trace.
Layout: make time read left to right
The single most useful layout rule we apply is this: time runs left to right, oldest on the left and newest on the right. Everywhere. Charts, tiles and table columns.
It sounds obvious, but many dashboards break it. A tile row shows "Today, Yesterday, Last 7 days" from left to right, while the chart below plots history with today on the right edge. The eye has to flip direction between the two. Tables often list the newest month in the first column, so a reader scanning across a row reads the trend backwards.
When everything runs the same direction, a reader can scan any row and see whether the line is going up or down without thinking. The most recent number is always in the same place: the right edge.
Trailing 7, 30 and 365-day comparisons
Calendar periods are misleading for a business that sells every day. "This month so far" on the 4th compares four days to a full prior month. Weekly reports swing with how many weekends fell in the period.
Trailing windows fix this. A trailing 7-day number always covers the last seven full days, so it always contains exactly one of each weekday. A trailing 30-day number smooths out weekly noise. A trailing 365-day number shows the underlying size and direction of the business, with seasonality already averaged in.
We lay these out as three columns in that order, 7, 30 and 365 days, from left to right. Then, under each, show the comparison to the same window one period earlier, and for the 7 and 30-day windows, the same window a year ago.
| Metric | Trailing 7 days | Trailing 30 days | Trailing 365 days |
|---|---|---|---|
| Shopify net sales | value and change vs prior 7 | value and change vs prior 30 | value and change vs prior 365 |
| Amazon ordered sales | value and change vs prior 7 | value and change vs prior 30 | value and change vs prior 365 |
| Gross margin (QuickBooks) | not shown, books not closed | value and change vs prior 30 | value and change vs prior 365 |
The bottom row illustrates a point worth repeating: do not show a number at a granularity the source cannot support. Accounting data that is reconciled monthly should not appear as a 7-day figure.
A related rule: do not judge a change on too short a window. A 7-day number moving 15 percent after an ad change is a reason to keep watching, not a conclusion. Use the 30-day view, and wait for enough data, before calling something a win or a loss.
What to put on the page
Most owners need less than they think. A good starting dashboard for a multi-channel e-commerce business fits on one screen:
- A tile row: trailing 7, 30 and 365-day revenue by channel, with change versus the prior period.
- A daily revenue chart for the last 90 days, one line per channel, oldest on the left.
- A margin row from QuickBooks: gross margin and operating expenses by month, oldest month on the left.
- An inventory watch list: products with fewer than a set number of days of stock left, based on recent sales velocity.
- Ad efficiency: spend and return by channel on the same trailing windows.
Leave out anything nobody has made a decision with in the last quarter.
How it runs
The reliable pattern separates collection from display. A scheduled workflow pulls fresh data from each source on a fixed timetable, stores it, and the dashboard reads from that store. We use n8n for this kind of scheduled pull, and explain why in n8n vs Claude for workflow automation. The result is fast page loads, a history you control, and dashboards that still show yesterday's numbers if a platform API has a bad morning.
Then connect the same sources to Claude through MCP so you can investigate, not just observe. We cover the setup in what MCP servers are.
Access control matters too. A dashboard showing margins and cash should sit behind a login, not on a public URL. Platforms like Cloudflare can put an access check in front of an internal page without changing the page itself.
Common mistakes
- Different definitions of revenue on different tiles.
- Mixing time directions between charts and tables.
- Showing today as a full day when it is half over. Show complete days, or label today as partial.
- Rolling up multiple stores or seller accounts before checking each one separately.
- Trusting the AI to compute totals in a conversation instead of reading them from fixed queries.
Our AI dashboards and BI work covers the full stack: connecting the sources, defining the metrics with you, building the page and setting up AI investigation on top.
Frequently asked questions
Do I need a data warehouse to build an AI dashboard?
Not at first. A small business can store scheduled pulls in a simple database or even a spreadsheet and build the dashboard on top. A warehouse becomes worth it when you have many sources, years of history or several people writing their own queries.
How often should the dashboard refresh?
Match the source. Shopify orders can refresh hourly or more often. Amazon sales data lags, so daily with a re-pull of recent days is sensible. QuickBooks figures change mostly at close, so daily is plenty.
Why do my Amazon numbers not match Seller Central exactly?
Usually timing, time zone or definition. Amazon revises recent data, reports in a set time zone, and distinguishes ordered sales from settled payments. Re-pulling recent days and labeling which measure you show resolves most gaps.
Can AI tell me why a number changed?
It can investigate. With your systems connected, Claude can pull the related data, such as ad spend, inventory, pricing and order mix, and suggest likely causes. Treat its answer as a lead to verify, not a final conclusion.
Can someone non-technical maintain a dashboard built with AI?
Yes, if it was built with that in mind. Keep the metric definitions written down, keep the data pulls in a visible workflow tool, and use Claude to make changes. We have non-developer executives maintaining their own internal dashboards this way.