Financial due diligence: from trial balance to adjusted EBITDA, refreshed

Financial due diligence is a rebuild problem. The databook is assembled from trial balances, ledger extracts and management accounts; the accounts are mapped to schedules; adjustments are scheduled; the bridge to adjusted EBITDA is drawn. Then the next extract lands, a month later or a restated one, and a large part of that work is done again by hand.

This playbook describes how a team runs the same work in Aqmen so that a new extract re-runs the book instead of the analyst. Four properties hold throughout: every schedule line traces back to its account; a new trial balance refreshes everything downstream; every adjustment carries its reason, in one place; and the schedules are built the same way each time.

Set up

The engagement is a project in the client's workspace, with the scope and the questions in its notes. Collections hold the work by topic: the databook, the mapping, quality of earnings, net debt and working capital, the findings. Every extract the client sends is a source: the file is stored in the workspace and cited by everything built from it.

Workstream 1: the databook

Trial balances, general ledger extracts and management accounts land as datasets. Each keeps the file it came from, and each column carries its meaning: the entity, the period, the sign convention, the currency and the scale.

The rule that keeps this manageable is that datasets arrive clean. A raw export with merged headers, subtotals and a different column order every month is cleaned once, before it is loaded, and the clean shape is what Aqmen keeps. From then on:

• every extract is versioned, with its source file;
• periods are aligned, so twelve monthly balances read as one table;
• the next month is appended, and the history stays.

Nothing about this is exotic. It is the discipline a good databook already has, enforced by the place it lives rather than by whoever built it last.

Workstream 2: trial balance to schedules

Accounts map to your P&L, balance sheet and working-capital schedules through a mapping you own. The mapping is itself a dataset: account, schedule line, and where needed a flag for items that belong in adjustments rather than in the run-rate.

Financial due diligence

Trial balance, mapped to the schedules

Account	Schedule	FY24 · €k
Product revenue	Revenue	48,210
Services revenue	Revenue	6,930
Contract staff	Cost of sales · adjusted	(1,240)
Legal, one-off	Non-recurring	(610)
Rent	Operating costs	(2,050)

The mapping is kept as data: review it once, re-run it on the next extract.

842 of 860 accounts mapped18 to confirm

Your assistant proposes the mapping from the account names and the chart of accounts; you confirm it, and you spend your time on the exceptions. The schedules are then derived from the trial balance and the mapping, with the method recorded, so:

• the mapping is data: reviewable line by line, versioned, and reusable on the next engagement with the same chart of accounts;
• unmapped accounts surface instead of hiding in an "other" line. In the example, 842 of 860 accounts are mapped and 18 are waiting for a decision;
• when the extract changes, the derivation re-runs in one action, and only new or changed accounts need a look.

The proposal is the assistant's, the decision is yours: Aqmen does not map accounts automatically, and it should not.

Workstream 3: quality of earnings

Adjustments are named lines, each with its reason and its evidence, in a spreadsheet model connected to the schedules. "Contract staff reclassified to cost of sales" or "one-off legal settlement" is a row with an amount, a rationale, and a citation to the invoice, the board minute or the management answer that supports it.

Because the model reads the schedules rather than a pasted copy of them:

• reported EBITDA always ties to the trial balance;
• working capital and net debt are built on the same basis, from the same mapped accounts, so a reclassification lands in every schedule it should;
• the bridge from reported to adjusted EBITDA is a chart drawn from the model, never a picture pasted into the report, so it is current every time it is opened.

Checks sit in the model: the balance sheet balances, the bridge sums, the adjusted figure equals reported plus the adjustments. When a check fails, it fails in the workspace.

Workstream 4: findings

Each finding is written in the findings document: the observation, its figure read from the schedule or chart that shows it, and its caveat. A figure in a document is not typed; it is a sourced block that knows where it came from.

Findings · every figure sourced

Where the market sits

Firms of 250+ people are $4.2B of $7.5B¹, 55% of the market. Adoption is a proxy: read the split, not the level.

Southern Europe added the most since 2021, +$0.40B², just ahead of DACH.

¹ Up to date² Stale · the model changed since

1. 1Market model, Summary sheet, 2024.
2. 2Market model, Bridge sheet, 2021 to 2024.

That is what makes the refresh safe. If a later extract moves a number that a finding relies on, the figure is flagged as stale before it reaches the report, and the reviewer sees exactly which paragraph to re-read. The document exports editable for the final report.

Every adjustment, audited

Every number in the book has an origin: the extract, the mapping line that placed it, the adjustment that changed it, and the evidence for that adjustment. All of it is on the record.

Every assumption · audited

What $1.67B rests on

Assumption	Value	How it was derived	Source
Adoption	89.1%	Average of the 2023 and 2025 surveys	Eurostat
Price per seat	$107	Vendor band price × price level	Vendr, Eurostat
Employed	17.5M	As published	Eurostat

Three of three traced to a documentConfidence stated where it is a proxy

• The method that derived a figure is stored with the figure.
• A confidence is stated where an amount is estimated rather than booked.
• Before an extract is replaced, you can see every schedule, adjustment and finding it touches.

When the buyer's lender asks, a month after signing, why adjusted EBITDA is what it is, the answer is the same path the team used, not a reconstruction.

Hand over the workspace, not a file

The report and the databook pack still come out as files, in your house style. The client can also join the workspace as a guest: open the schedules, follow an adjustment back to its evidence, and ask the next question through their own assistant, from the same data. The report links back to the workspace.

Chat, workbook or workspace

What it runs on

Aqmen works inside the assistant your team already uses (Claude, ChatGPT or any MCP-capable assistant) over MCP. The step-by-step practice is in the open Aqmen skills; run the work in the app, or hand it to your assistant with the skills. Models, documents and decks export editable. The workspace is private to your organization and never used to train models.

If your team rebuilds the book every time the extract changes, talk to our team.