Analyse spreadsheets and explain the results

Key takeaways

• Docsi calculates figures exactly from the cells; the AI only explains.
• Check columns, units and row count before the analysis.
• With outliers, the median is often more meaningful than the mean.
• Percentages need a clearly stated base.
• Separate data basis, figures and interpretation in the report.

Check the data first

Import XLSX or CSV where possible rather than a photograph of a table. Check sheet names, column headings and data types. Numbers with leading zeros can be identifiers; text and dates should not be treated as measurements.

Formula cells need particular care: stored results may be outdated. Docsi does not replace a spreadsheet program with full formula recalculation.

Request a specific analysis

1. Name the sheet, column and period. Explain which rows should be included.
2. Request suitable key figures: count, sum, mean, minimum or maximum.
3. For group comparisons, name the grouping column, such as month or category, and the numeric series.
4. Compare samples against the source file and check units in the report and chart.

Separate numbers from interpretation

A chart should show a real comparison or trend. Ask for labelled axes and an explanation of missing values. A correct sum is not a reliable explanation of causes or a forecast; ask for assumptions to be labelled.

For binding financial calculations, check rounding and decimal precision in the appropriate program. The table tools are not a specialised financial calculation engine.

What Docsi calculates exactly

For tables, Docsi has its own tool that the AI calls instead of estimating numbers itself. It describes each column with type, count, sum, mean, median, minimum and maximum, aggregates values per group and returns filtered or sorted rows. A second tool calculates individual expressions exactly and understands decimal commas such as “12,4” and grouped numbers such as “1.234,56”.

Docsi does not reinterpret text, dates or booleans as numbers. By default he does not count formula cells, because their stored result may be outdated; if a stored value is used after all, Docsi points this out. That keeps it clear which cells a figure is based on.

Mean or median?

The mean is the sum divided by the count; a single outlier can shift it strongly. The median is the middle value of the sorted series and stays stable with outliers. For incomes, processing times or prices, the median therefore often describes the “typical” result better.

Ask Docsi for both values if you are unsure, plus minimum and maximum. If mean and median are far apart, look at the largest values: often a typo, a different unit or a special case is behind it that deserves its own explanation.

Comparing shares and groups properly

A share always needs a clear base: “30% of orders” is different from “30% of revenue”. State in your request what percentages should refer to. In group comparisons, check whether the sum of all groups adds up to the total again; otherwise rows are missing or a group is empty.

Groups of very different size are hard to compare in absolute numbers. Then ask for values per unit – such as revenue per branch or errors per 1,000 units. Avoid conclusions about causes: a table shows relationships in your data, but not why they exist.

How large can a table be?

Docsi reads Excel workbooks with up to 50 sheets of 10,000 rows and 100 columns each and 200,000 cells in total. For comparison: Excel itself allows over a million rows per sheet and calculates with 15 digits of precision. So filter or trim very large exports beforehand to the period and columns your question concerns.

Docsi rejects macros, external links and embedded programs. Hyperlinks in cells are read as labels only. If a workbook does not load for that reason, save a copy without macros as a normal XLSX file.

From result to report

A good analysis report separates three levels: the data used (file, sheet, period, row count), the calculated figures and the interpretation. Ask Docsi explicitly for this structure. Then every reader can follow which statement is a calculation and which is an assessment.

As an Excel file you get the result table with native charts you can keep editing in Excel; as a Word or PDF report the same figures with text. Keep the source file together with the report so you can repeat the calculation later.

Data quality before the analysis

The most common errors arise before the calculation: duplicate rows, inconsistent spellings (“Berlin”, “berlin”, “Berlin ”), numbers stored as text and subtotals in the middle of the table. Ask Docsi first for a description of the columns with value counts and types; anomalies show up there immediately.

Fix such issues in the source file or have Docsi output a cleaned copy as Excel and check it before you calculate. Note what you changed so the analysis stays traceable later.

A worked scenario

An example: you have an Excel list of 2,400 orders from one year and want to know how revenue developed per quarter and region. You first ask Docsi for a column description. It turns out the “Amount” column contains 2,396 numbers and four text entries – rows reading “n/a”.

You decide to exclude these four rows and tell Docsi so explicitly. He then calculates the totals per quarter and region exactly, adds a bar chart and writes three sentences of assessment, separate from the figures. You verify one total with a filter in Excel; it matches, and the report can go out.

Key terms explained

Mean

The sum of all values divided by their count. Sensitive to outliers.

Median

The middle value of a sorted series. Robust against outliers.

Grouping

Combining rows by a feature, such as the sum per region.

Base

The total a percentage refers to.

Formula cell

A cell containing a formula. Docsi does not count it by default because its stored result may be outdated.