Twelve weeks, the implementation plan said. Fourteen countries, one platform, and a green arrow pointing right.
It is not a lie exactly. Twelve weeks is roughly how long it takes to configure a simple country if the data arrives clean and decisions get made on time. Configuration is maybe a fifth of the work. The rest sits on somebody else’s calendar: a tax authority’s registration queue, a works council’s meeting cycle, an incumbent’s extract team.
I once inherited a programme that signed in late August targeting 1 January for eleven countries. By mid-October the German works council had not been contacted, nobody had requested a PAYE reference for the UK entity, and the incumbent had sent four of eleven extracts, two of which did not reconcile. Three countries went live on 1 January. The rest landed in April, in July, and the following January. The business paid to run two payroll models side by side for fifteen months.
That is the normal outcome. Here is how to plan against it.
Your go-live date is a tax year boundary, and there are about six of them
Go live at the start of a tax year and your opening year-to-date balances are zero. Go live in month seven and you migrate YTD figures into a system that must treat them as if it had calculated them, which is where reconciliation goes to die. Cumulative jurisdictions are worse. UK PAYE recalculates the year at every payment; Indian TDS projects an annual liability across the remaining months. A wrong opening balance there does not produce a wrong number once. It produces one every period until year-end, and the employee finds it before you do.
The dates you get to choose from:
- 1 January for most of Europe, the US, Canada, China and Brazil
- 1 March for South Africa
- 6 April for the United Kingdom, where HMRC confirms the current tax year runs 6 April 2026 to 5 April 2027
- 1 April for India’s financial year, New Zealand, Hong Kong and Japan’s labour insurance year, though Japanese income tax runs on the calendar year with a December year-end adjustment
- 1 July for Australia
Miss 6 April and the next honest UK go-live is 6 April 2027. The boundary also leaves your incumbent owing the old year’s output after you have gone: in the UK, the final Full Payment Submission and P60s by 31 May, weeks after your first live run. Put that in the exit terms.
Wave one should be a country that can survive a mistake
Fourteen countries at once means fourteen simultaneous first attempts by people who have never run this system. Group them by complexity and by consequence.
What makes a country hard is predictable. Count the statutory filings per period: France files DSN monthly, Italy Uniemens, Poland ZUS, India Form 24Q quarterly plus monthly challans and state professional tax. Check whether pay rules come from collective agreements rather than statute: German Tarifverträge, Dutch CAOs, Italy’s CCNL with its thirteenth and fourteenth month payments. Then ask whether payslips need localising in language and layout, whether filings must be produced in a mandated format rather than a report, whether local banking imposes its own file specification, and whether tax is cumulative, projected or per period.
A good wave one has real headcount, 300 to 1,500 so the test means something, and a light statutory load: monthly, mostly salaried, one collective agreement or none. Ireland, Singapore and Canada tend to qualify. Germany, France, Italy, Poland, Brazil, China and India belong in wave two or three, once the team knows what the system does when a retroactive increase lands.
Seven workstreams run at once, and data is the one that slips
Statutory and entity setup. Pay element and policy design. Data migration. Integration build. Parallel runs. User training. Cutover and hypercare. Nobody resources the first one properly: tax registrations, employer reference numbers, bank mandates and powers of attorney move at the speed of a government office. Start it in week one.
Data is the one that slips. Every time. PayrollOrg’s Global Payroll Week 2025 survey put poor quality of data inputs at the top of the root causes for reduced payroll accuracy, with four of the six most-cited challenges relating to inbound data rather than calculation.
“Bad data” is not a vague complaint. In practice it is national identifiers missing or failing their check digit, and every country has one: the Dutch BSN eleven-test, the Italian codice fiscale, the Polish PESEL. IBANs that fail the mod-97 check. Addresses that fail local validation because a Japanese prefecture went into a free-text field. Missing date of birth or gender where a filing needs both, which stops a German DEÜV registration cold. Balances that do not reconcile to the incumbent’s own reports. And two records for the same human being, usually a souvenir of an acquisition.
Do the extract and the validation before you sign, or by week two. Push the legacy data through the target system’s validation rules and count the failures. That number, not the methodology deck, sets your timeline.
Germany’s works council has co-determination, not an opinion
Foreign programme managers discover this in month five. It is not a soft consultation.
Section 87(1)(6) of the Betriebsverfassungsgesetz gives the works council co-determination over the “Einführung und Anwendung von technischen Einrichtungen, die dazu bestimmt sind, das Verhalten oder die Leistung der Arbeitnehmer zu überwachen”: the introduction and use of technical systems intended to monitor employee behaviour or performance. The word “intended” does less work than it looks. The Bundesarbeitsgericht held on 25 September 2012 (1 ABR 45/11) that objective suitability for monitoring, judged on technical characteristics and actual deployment, is enough. Your intent is irrelevant, and commentary puts software that stores personal data squarely inside the provision.
Without agreement, section 87(2) sends the matter to the Einigungsstelle, whose award replaces the agreement between the parties, and the works council can seek an injunction against using the system meanwhile. What you need is a Betriebsvereinbarung covering which data is stored, who sees it, when it is deleted, and a ban on using it to monitor performance. Budget two to three months, longer where a Gesamtbetriebsrat is involved.
The Netherlands is comparable and more specific. The Dutch data protection authority states that Article 27(1)(k) of the Wet op de ondernemingsraden requires works council consent for rules on processing personnel data, naming payroll administration explicitly, and that Article 27(1)(l) covers systems capable of registering attendance, behaviour or performance. Article 25(1)(k) gives only an advisory right on a new technological facility, but that becomes consent once the facility monitors staff. A decision taken without consent is void if the works council invokes nullity in writing within one month; the employer’s remedy is to ask the subdistrict court for substitute consent. France adds its own information and consultation duty on the CSE before new technology is introduced, so check the French position with counsel rather than assuming it mirrors Germany.
None of these compress because your steering committee has a date.
The provider you are leaving controls the data you need
Read the exit clause of your current contract before you sign the new one. Notice periods of three to six months are normal, often tied to a year-end. What matters more is what the incumbent owes you: the extract format, whether it carries calculation history or only closing balances, how many extracts you get before they charge, and how long the archive stays reachable.
Then decide who owns historical payslips, because an employee will ask for a 2024 document in 2028. German employers must keep the Lohnkonto until the end of the sixth calendar year following the last wage payment recorded in it, under section 41(1) of the Einkommensteuergesetz. Every other country sets its own period, and they are rarely the same, so pull the retention rule per country before you agree an archive scope. Buy an archive, load historical PDFs into the new self-service, or pay for a read-only legacy tenant. All three cost money. Pick one on purpose.
The awkward part is sequencing. The company that must hand over your data is the company you just fired, and it is often still running your year-end. Get a test extract against an agreed specification while you are still a paying customer in good standing. Give notice after that.
Hypercare ends months before the system is proven
Parallel runs get their own treatment elsewhere; the short version is three cycles, reconciled by pay element and employer cost, not by net pay.
What matters is what happens after. The project team demobilises after two or three stable periods, and two or three stable periods prove one thing: the recurring monthly works. They prove nothing about the annual events, and payroll is mostly annual events wearing a monthly costume. The first bonus cycle. The first termination in a country with statutory severance and untaken leave to pay out. An increase backdated four months. January rate changes. Year-end on a system nobody has taken through a year-end. Your acceptance test is twelve months long, so keep a named person who can still read the configuration in month eleven.
What kills these programmes is rarely the software. It is the absence of a named business owner who can make a decision and be wrong. It is decisions taken by the project team that belonged to the business: whether the Belgian car allowance is taxable is not a configuration question. It is scope added one country at a time, each addition individually reasonable. It is testing compressed because build ran late, which is testing cancelled with extra steps. And it is a go-live date defended for months after everyone in the room privately knew it was gone.
Questions to answer before you sign the statement of work
Take these to the vendor and to your steering committee. Get written answers.
Which tax year boundary is each country’s go-live, and what is the fallback if we miss it? Which countries are in wave one, and why those? Who is the named business owner per country, and which decisions are explicitly theirs rather than the project’s? What did the validation run on real extracted data return, as a failure count by rule? Which countries have works councils, and who owns that consultation? What is the incumbent obliged to hand over, and have we tested it? And who is still on this in month eleven, when the first year-end arrives?
Any question on that list without a written answer is the reason the date will move, and it is telling you so now rather than in month nine.
HR Blizz calculates gross-to-net on its own engine rather than routing to local partner payroll engines, with statutory filings produced inside the run. If you are scoping a multi-country implementation and want to pressure-test the sequencing before you commit to a date, send us the country list.
The co-determination and retention rules summarised here are general information, not legal advice, and works council law in particular rewards local counsel.
FAQ
Q: When should a multi-country payroll implementation go live?
At the start of a tax year, because opening year-to-date balances are then zero and there is nothing to migrate or reconcile. The practical options are 1 January for most countries, 1 March for South Africa, 6 April for the UK, 1 April for India, New Zealand and Hong Kong, and 1 July for Australia. Missing one of those dates usually costs a full quarter or forces a YTD migration.
Q: Does a German works council have to approve a new payroll system?
Yes. Section 87(1)(6) of the Betriebsverfassungsgesetz gives the works council co-determination over the introduction and use of technical systems capable of monitoring employee behaviour or performance, and the Bundesarbeitsgericht held in 1 ABR 45/11 (25 September 2012) that objective suitability is enough, regardless of the employer’s intent. If no agreement is reached, the Einigungsstelle decides and its award replaces the parties’ agreement.
Q: What does the Dutch Works Councils Act require for an HR or payroll system?
The Dutch data protection authority states that Article 27(1)(k) of the Wet op de ondernemingsraden requires works council consent for rules on processing personnel data, including payroll administration, and Article 27(1)(l) covers systems that register attendance, behaviour or performance. A decision taken without consent is void if the works council invokes nullity in writing within one month.
Q: Which workstream most often delays a payroll implementation?
Data migration. PayrollOrg’s Global Payroll Week 2025 survey ranked poor quality of data inputs as the leading root cause of reduced payroll accuracy, with four of the six most-cited challenges relating to inbound data quality. Running the legacy extract through the target system’s validation rules in the first weeks gives you a failure count that predicts the real timeline.