Some companies run two sets of books without calling it that. This company had grown organically over five years. Project costs were tracked on a cash basis in Excel — pragmatic, fast, familiar. The actual accounting, meanwhile, lived in the parent group's complex MS Dynamics 365 model. Two worlds that had never been introduced in five years.
With audited statements approaching and asset capitalisations worth millions at stake, one uncomfortably simple question went unanswered: have we captured every cash position? Are there unrecorded liabilities? The gap between the cash view and the books ran into the triple-digit millions — nobody could say which side was right.
Scrapping it was not an option
Most consultants would have recommended a fresh start here: drop Excel, bring in a clean system. That was not an option. Five years of history had to be carried forward, and operations couldn't stop. We took the opposite route — not rebuild, but reconstruct. We replicated the existing Excel logic in a deterministic ETL process, so every figure became traceable.
On an "if this, then that" basis — every number explainable, no black box.
Does this sound familiar?
If your cash view and your accounting show different numbers, let's look together in a free first conversation at how both worlds can be brought together in an audit-proof way.
Reverse-engineering an unknown data model from scratch
In parallel, we rebuilt the Dynamics data model directly at raw-data level. No one in the group could tell us which tables mattered — so we found out ourselves. Newly introduced financial tags let us attach a cost centre to every posting retroactively, without re-booking and without holding up payments.
The trickiest part: closed financial years must not be altered. So we placed a management adjustment layer directly in the database.
Corrections, without distorting the statutory accounts.
The moment both worlds meet
Today all figures — actuals, management adjustments, forecast and payments — meet in one central model. The cumulative reporting deviation across all periods is 0.23%. Reporting stays consistent month over month, and the banks see one reliable number. In the process, around 13,000 posting lines were reclassified and made unambiguously attributable for the first time.
Auditable by anyone who can read a table
The platform runs on standard SQL and Python — maintainable and readable by anyone on the team, with no dependency on a single expert's head. Operations were never interrupted; only the forecast feed and the cost-centre tagging changed. Ongoing software cost: €250–400 a month.
Three things you can take away
- Reconstruct instead of scrapping. Five years of history and a well-rehearsed Excel model aren't a reason to start over — they're the starting point.
- An adjustment layer instead of re-booking. Closed financial years stay untouched; corrections live in their own, traceable layer.
- Standard SQL instead of a black box. Auditability doesn't come from an expensive tool — it comes from every number staying explainable.