Consolidating several ERPs: the step nobody puts a price on, and the one that decides everything

A group of four companies, four instances of the same ERP, and a management team that wants a consolidated revenue figure. On paper, that is four connections and a sum.

In practice, the customer Dupont is coded CLI-0042 in France, C00042 in Belgium, 42-CLIENT in Switzerland and IT_42 in Italy. Nobody ever decided this: each subsidiary kept its own conventions when it went live, and the question never had to be asked as long as each one steered on its own.

It is this step, aligning the master data, that is missing from most quotes. And it regularly accounts for a third of the project workload.

Three sets of master data to align, almost always the same ones

Third parties

Customers and suppliers. The same group appears in several companies under different codes, sometimes under different legal names. A customer in France can be a supplier in Italy.

The question to settle is: at which level do we consolidate? The group's head office, the invoiced legal entity, or the delivery site? All three answers are defensible and give three different figures. A choice has to be made, and it belongs to the finance department, not to us.

Items

The same product carries different references depending on the company, sometimes with different units: per kilo here, per piece there. Family classifications almost always diverge.

The trap is partial grouping: ten references aligned out of twelve give a report that looks right and quietly understates that family by twenty per cent.

The chart of accounts

Each company has its own chart, often inherited from its country. A French account 706 has no direct equivalent in a Belgian or Swiss chart. Consolidation goes through a consolidation mapping: a correspondence table that says which local account feeds which consolidated line.

This table rarely exists in a usable format. It lives in a workbook maintained by management control, with exceptions noted in comments.

What we do, in order

First, we produce the gap table. For each set of master data, the list of distinct values per company, and the matches we propose. This is a deliverable, not an internal step: the client must be able to read it.

We get a decision, in writing. At which level do we consolidate third parties, which reference is authoritative for an item, which account feeds which line. One person must have the authority to decide, otherwise the project stops at the first divergence, and there is always one.

We materialise it in an intermediate layer. The correspondence tables live in the platform, versioned, with an effective date. A mapping that changes during the year must not rewrite history: the past keeps its rules, the future takes the new ones.

We reconcile company by company before consolidating. The consolidated total means nothing until each company is right on its own. Two errors of opposite sign cancel each other out perfectly in a total, and show up immediately in the detail.

What it costs, honestly

On our projects, alignment represents between a quarter and a third of the workload when the companies share the same ERP, and more when the systems differ. That is more than extraction and more than building the reports.

It is also the part that produces the most lasting value, because these correspondence tables then serve everything: reporting of course, but also intercompany exchanges, reconciliations and the next migration.

The technical detail that changes the scale

With multi-instance Business Central, extraction through the API quickly reaches its limits: call quotas are used up fast, and a full daily reload becomes impractical.

We use bc2adls, an open source component published by Microsoft, which exports Business Central tables directly in Delta format into a data lake, incrementally. The ERP's quotas stay available for what they are meant for, day-to-day operations, and loading the history stops being a problem.

What to remember

If a multi-ERP consolidation quote mentions neither the alignment of third parties, nor the accounting consolidation mapping, nor validation by a named person, it is not that the quote is optimistic: it prices half of the project.

We have run up to seven Business Central instances and seven CRMs on a single model. Each time, the difficulty was the same, and it was never the connection.