Skip to content

Outgrowing Excel: when a small business needs a warehouse

Seven signs your Excel reporting file is overloaded, what a data warehouse really does, the cost and risk of each step, and how to move without dropping Excel.

By The DataxLand team· · 6 min read

Excel is the best analysis tool most small businesses have ever had. It is flexible, everyone knows it, and nobody has to ask for a budget. This post will not tell you to drop Excel. It will help you spot the moment Excel turns from an analysis tool into a reporting system — a job it was never designed for — and what to add when that happens.

Seven signs Excel is overloaded

Tick the ones that are true for you:

  • The weekly report takes more than half a day of exporting files from marketplaces, ad platforms and your sales software, then stitching them together.
  • Only one person understands the file. When they are on leave, there is no report.
  • Two files give two different "revenue" numbers, and nobody is sure which one is right.
  • The file is slow or breaks: hundreds of thousands of rows, VLOOKUPs across sheets, minutes just to open.
  • A marketplace changes its export format and your formulas quietly miscalculate for weeks before anyone notices.
  • You cannot answer "how was September calculated?" because the September file has been overwritten.
  • Files containing customer details (names, phone numbers, addresses) get passed around by chat and email.

Three or more ticks, and the problem is no longer Excel skills. The problem is that a spreadsheet is doing the job of an entire system.

One hard limit is worth knowing: an Excel worksheet holds at most 1,048,576 rows (Microsoft — Excel specifications and limits). A shop with a few hundred orders a day, several product lines per order, plus fee rows and daily ad rows, will hit the "everything is slow" stage long before that limit.

What a "data warehouse" actually does

"Data warehouse" sounds like something banks have. Strip away the jargon and it is four jobs your Excel reporting file already does by hand:

JobIn the Excel fileIn a data warehouse
1. Load dataExport each source, paste into a sheetConnect the source or upload the file; keep every load as is
2. Clean and standardizeFix by hand: rename columns, merge order statuses, delete duplicatesRules written once, rerun on every load
3. Join and calculateVLOOKUP, SUMIFS, pivotsA data model: orders ↔ fees ↔ cost of goods ↔ ads
4. Define metricsA formula in each cellNamed, versioned, shared definitions

Plus two things Excel does not do: automatic data checks (duplicate rows, SKUs missing from the catalog, unknown statuses) and history (knowing which data last week's number was calculated from).

Three steps, three levels of cost

Not everyone needs to jump straight to a warehouse. These three steps are ordered by investment:

StepWhat you doFits whenRemaining risk
A. Disciplined ExcelOne "source" file that only holds pasted raw data, one calculation file that references it; a sheet of metric definitions; a saved copy per month1 channel, 1 person on the numbers, under a few thousand orders a monthStill joined by hand; still depends on one person
B. Google Sheets + a reporting toolPull data into Sheets, connect a charting toolYou need to share a dashboard, data is smallFormulas scattered around; Sheets slows down as it grows; still no shared definitions
C. Data warehouse + semantic layerAutomatic loading, modeling, defined metrics, dashboards and AI reading the same numbersMulti-channel, several people need numbers, you want to ask AI about your numbersSomeone has to design the model — or a tool does it for you

Step C used to mean hiring a data engineer and paying for a cloud warehouse. Two things have changed that make step C much cheaper for small businesses:

  1. Lightweight lakehouse warehouses. Data is stored as Parquet files on object storage, the table catalog lives in an ordinary database, and computation runs on an embedded engine. DuckLake is one such format. You do not need a cluster running 24/7 for a few million rows of sales data.
  2. AI proposes the model. The most labor-intensive part of step C — understanding which column is what and which table joins to which — can now be proposed by AI and approved by a person.

How DataxLand does step C

DataxLand is step C, prebuilt for small businesses. You can keep using Excel as a source:

  1. Load. Upload an Excel file (one sheet per table) or connect a source: Shopee, TikTok Shop, Lazada, ad platforms, Google Sheets, databases. The original data is kept as is; personal-data columns are hashed as soon as they load.
  2. Understand. Si analyzes the context: it reads a sample of the data (with personal data masked), proposes which column is cost of goods, which status means a return, which table joins to which, and asks questions when it is unsure. You approve before anything is built.
  3. Build. Data passes through three layers: as is → standardized → modeled. After every build, data checks run automatically; serious failures block publishing, minor ones show a warning. Each workspace has its own warehouse (DuckLake), isolated from other workspaces.
  4. Read. Dashboards, Si, Telegram/Slack reports, Google Sheets for your accountant and agents over MCP all read metrics from the same published warehouse, with the same definitions.

For the detailed architecture, see DataxLand architecture.

Example: what Excel struggles to see

In the DataxLand sample data (a simulated scented-candle shop, three channels, 1 Aug – 29 Oct 2026), the data checks catch three order lines whose SKU is missing from the product catalog and one duplicated fee row. In Excel, the three unknown-SKU lines usually drop silently out of the cost-of-goods VLOOKUP (inflating profit), while the duplicated fee row drags profit down. Neither error is big enough to make anyone suspicious — which is exactly what makes them dangerous.

In the same sample data, the cost of one product changes from 260,000 ₫ to 285,000 ₫ on 1 October. The warehouse stores cost of goods by effective date, so September orders use the old cost and October orders use the new one. In a cost sheet that gets overwritten, September's profit would quietly change too.

Where to keep Excel

A data warehouse does not replace Excel everywhere. Excel is still best for:

  • One-off analysis: price simulations, "what if fees go up 2%" scenarios.
  • Entering small data no system holds: cost of goods, fixed costs, plans.
  • Receiving numbers from the warehouse: DataxLand can push numbers to Google Sheets on a schedule for your accountant.

A simple rule: Excel for thinking, the warehouse for remembering. Numbers that are shared, repeated every week and need to be trusted go through the warehouse; one-off analysis can stay in Excel.

Start small

You do not need to move everything at once. Take the exact Excel file you use for the weekly report, upload it, review Si's proposal, and compare the dashboard numbers with your file. If they differ, find out why — it is usually one of the errors above.

Read next: What is a semantic layer · Profit and loss dashboard: 8 metrics to track

Related posts

Verified dashboards from your own data, this afternoon

Free plan with no time limit · No card required · Sample workspace ready in minutes