Turned a consultancy's know-how into a product it sells

The result

Global HR now owns a product, Launch Saviour, built on its own method and listed on its website. It can sell it alongside its consultancy, and its growth is no longer limited by how many experienced people it has to check spreadsheets.

Weeks of manual checking become a repeatable process with a record of every decision. It has been tested at the size of a 300,000-employee payroll. AI does the sorting, and employees' personal details are never shown to it.

The problem

Global HR moves companies from one payroll system to another, often across several countries. The riskiest moment comes before go-live, when the old and new payrolls run side by side and someone has to prove every employee will be paid exactly the same.

That proof was done by hand in spreadsheets. It took weeks, depended on a few experienced people, and one missed difference meant someone paid wrongly on day one.

Global HR wanted to turn that know-how into something it could sell. The data involved is some of the most sensitive a business holds: names, addresses, dates of birth, bank details and pay.

What we did

We designed and delivered Launch Saviour in four fixed-scope stages between January and July 2026, with daily check-ins with the client throughout. It compares the two payrolls line by line, uses AI to sort each difference into "explainable" or "needs investigating", and records who decided what and when.

For the technical reader:

• Reconciliation across systems. Source files from any payroll platform (SAP, ADP, Workday and others), in CSV or Excel, are normalised to one canonical format and mapped wage type by wage type. The comparison runs field by field with configurable tolerances.
• AI inside a hard privacy boundary. Before anything reaches a model, real employee IDs become anonymous sequential IDs, and names, addresses, dates of birth, bank details and National Insurance numbers are removed. The model sees only codes, amounts and differences, and suggests a likely cause with a confidence score. The AI provider can be swapped without a rebuild.
• Evidence for every call. Payslip verification checks each AI diagnosis against real payslips. Correction reports go to the implementation team, corrected data comes back as a new round, and the match rate is tracked from round to round.
• Isolation and assurance. Each migration's employee data sits in its own database. Upload, parsing and comparison were benchmarked at 100,000, 200,000 and 300,000 employees. The platform carries 37 end-to-end browser tests and structured audits of security, correctness, performance and dependencies, each tracked to closure.

If your business depends on moving sensitive data between systems without a single mistake, talk to us.