How to fill Excel with figures from financial reports

What this is about

Valuation, credit analysis, annual accounts: almost every piece of financial work starts with the same handful of figures. Revenue, EBITDA, total assets, liabilities, cash flow. They are buried in annual reports several hundred pages long, sometimes only as a scan, sometimes in a language you do not read every day.

There are essentially four ways to get them into a workbook. What separates them is not so much speed as whether you can still tell, afterwards, where each figure came from.

Copying by hand

The classic: open the report, find the figure, type it into the cell, move on to the next. Careful people rarely get it wrong, because nothing stands between them and the source.

The price is hours per report, and the typical mistakes are easy to miss: two digits swapped, the prior-year column by accident, thousands instead of millions. Which figure came from which page is known only to whoever wrote it down.

PDF converters

A converter pulls the tables out of a PDF and drops them into Excel. It is fast, and the figures themselves come through untouched.

What a converter does not know is your model. Which row is revenue, which column is the current year and which cell the figure belongs in is still yours to work out. On top of that, tables that run over two pages, tightly set columns and scanned pages often come out scrambled, and you only find out when you compare with the report.

Autopilot AI in Excel

AI now sits right inside Excel: the big chatbots as add-ins, plus specialised tools that read a report on their own and fill your model. One click, and a few seconds later the cells are full.

That is exactly where the risk lies. A language model does not look a figure up, it writes the most likely answer. Usually that matches the report, sometimes it does not. Then the prior-year figure is suddenly in your model, thousands have become millions, one segment’s revenue has been taken for the group’s, or the figure does not appear in the report at all. Because everything is already in the cells before anyone has looked, a wrong figure looks exactly like a right one. The only way to be sure is to check every cell against the report, which is precisely the work the tool was meant to take off your hands.

Why this weighs so heavily in finance is the subject of a separate article.

modelflow

modelflow starts from a sober assumption: no AI reads a report without mistakes, ours included. So the checking sits at the centre, not the AI. modelflow reads the report, finds the figures for the fields in your model and shows each one on the page it came from, with the figure highlighted. If it is right, you accept it. If it is wrong, you click the right figure. Nothing is accepted automatically, and nothing reaches the workbook until you have confirmed it.

You still need to pay attention, and that is the point. You just no longer hunt for a figure in a 300-page report, you look at one highlighted spot. And every approved value keeps its source page, so years later you can still show where a figure came from.

Side by side

Which one fits

For one figure from a short report, no tool is worth it, typing is quicker. If all you want is a rough overview, an autopilot will do, as long as you treat everything it wrote as unchecked.

Once others work with your model, base decisions on it or audit it, something else counts: that every figure was checked against the source, and that you can prove it. That is exactly the case we built modelflow for.