On this page
What does the AI do in a monthly review?
- Variance analysis: actuals against budget, last year and last month, largest differences first.
- Forensic check: the checks an accountant would run, such as margins that shift, working capital that builds up and costs that change category.
- Questions for management: for each variance, a concrete question for the CFO or CEO, with the figure attached.
What does the workflow look like?
- The portfolio company delivers its monthly figures in a fixed format.
- The workflow starts in the morning and runs the analysis and checks as fixed steps.
- Calculations run through tested calculation steps. The language model writes the commentary and the questions.
- A few hours later the report is ready for review.
- The deal team reads, edits and sends the questions to management.
Why not just paste the figures into ChatGPT?
A language model does not calculate by itself. It writes code or it estimates. With a monthly pack of dozens of lines, that goes wrong sooner or later, and nobody sees where.
You also want the same checks to run every month, in the same order. A loose chat does not remember that.
And a portfolio company’s figures do not belong in a personal account. See Deal documents in ChatGPT.
Who reviews the report?
The deal team. The workflow prepares; the investment manager reviews it, picks which questions go to the company and signs off. The review stays with the team that knows the business.
What does it bring besides time?
- Every portfolio company is looked at the same way, every month.
- Variances surface in the month they arise.
- The questions to management and the answers are on record, even when someone leaves the team.
Where do you start?
With one portfolio company and the report you already make. We write down the current steps, rebuild them as a fixed workflow and run it alongside the manual process first. The team switches once the results match.
How such an engagement runs is on How we work. Who builds it is on About Wouter.