---
title: "How JAM! Horeca recovered 55 hours a month in HR back-office time "
description: "JAM! Horeca is the payroll and HR specialist for Dutch hospitality, with roughly 8,000 employees and 350+ client businesses. Every placement starts with an…"
canonical: "https://www.whatsnext-ai.com/cases/how-jam-horeca-recovered-55-hours-a-month-in-hr-back-office-time"
updated: "2026-08-14T10:06:49.898Z"
client: "JAM! Horeca"
---

# How JAM! Horeca recovered 55 hours a month in HR back-office time 

JAM! Horeca is the payroll and HR specialist for Dutch hospitality, with roughly 8,000 employees and 350+ client businesses. Every placement starts with an…

![Bartender behind the bar at a Dutch hospitality venue staffed by JAM! Horeca](https://fgzcpjbyiakhjifaciaj.supabase.co/storage/v1/object/public/media/Man%20Making%20Drink%20at%20Bar.jpg)

## Overview

JAM! Horeca is the payroll and HR specialist for the Dutch hospitality sector, with roughly 8,000 employees and 350+ client businesses. Every placement starts with an onboarding submitted through the MyJAM form by both the client business and the worker. HR back-office exports the submissions to Excel and checks them before anything is committed in MyJAM.

## Details

- Industry: HR & Payroll
- Region: Nederland
- Company size: 100+
- Timeline: 2 weken

## Challenge

Back-office staff checked every onboarding line by line against a written procedure manual, then keyed the approved fields into MyJAM by hand and cross-checked them against the worker's passport.

At 70 onboardings a day averaging 4:11 each, that was close to five hours of daily review before a single placement moved forward.

Consistency depended on which reviewer had the file, and a miscalculation that got through ended up in a worker's reported annual salary.

## Approach

Discovery started with the procedure manual back-office already worked from. The rules were not missing; they just lived in a document instead of a system, which is why consistency varied by reviewer.

Two decisions shaped the build. The first was to write the rules as deterministic code rather than run them through a language model. Onboarding data is BSN numbers, passports and bank details, so keeping it out of a model was a privacy requirement before it was an engineering preference. A minimum-wage (WML) check or a BSN checksum has exactly one correct answer, has to return the same verdict on every run, and has to be explainable to an auditor. Roughly 15 coded rules backed by around 139 automated tests do that; a probabilistic system does not.

The second was to leave HR's way of working alone. Changing it would have meant integrating directly into MyJAM, which holds live worker and payroll data, and that would have added months. Instead the fastest route to value ran alongside it: a scheduled job reads the SharePoint folder the team already uses and drops the validated Excel file back for collection. Same form, same folder, same people.

That got the system into production two weeks after the first conversation, followed by four rounds of improvements over the next four weeks to settle it into a stable process.

## Solution

A validation service picks up each exported onboarding file and applies roughly 15 rules to every row: Dutch BSN checksum, IBAN format and account-holder name match, minimum wage (WML) by age bracket, salary formula, contract length, ID expiry against contract dates, AOW state-pension routing, and a mandatory second review for non-Dutch nationality.

It returns a colour-coded five-sheet Excel file. Green rows are ready to process. Red, orange and yellow rows carry a plain-language explanation of what failed and a suggested route: back to the worker, back to the client, or escalate to HR.

The system does not approve anyone into payroll. It decides what a human needs to look at.

## Outcome

Across 70 onboardings a day, the shift from 4:11 to 1:52 per check returns roughly 2.7 hours a day and about 55 hours a month of back-office capacity.

Every submission now passes the same 15 checks in the same order, with a written record of what failed and why.

MetricBefore After ImpactTime per onboarding check4:11 minutes avg1:52 minutes avg55% faster (2.2×)Daily back-office review (70 cases)~4.9 hrs~2.2 hrs~2.7 hrs/day recoveredRules applied per submissionManual, varied by reviewer~15, identical every runReproducible and audit-ready

In short: 2.2× faster onboarding checks (4:11 → 1:52); ~55 hours of back-office time recovered per month; ~15 compliance rules applied identically on every run, backed by ~139 automated tests; no onboarding data stored and none of it passed through a language model; in production two weeks from the first conversation, stable at six weeks; and a human still makes the final call on every placement.

## Results

- **~55 hrs** back-office time recovered per month: Across 70 onboarding checks a day
- **4:11 → 1:52** average time per onboarding check: The same checks, 55% faster
- **0** onboarding records through a language model: Deterministic code, ~139 automated tests

## Quote

> With this solution we brought our average onboarding time down from 4:11 to 1:52. That gives our team a significant daily time saving and makes the onboarding process faster, more efficient and less error-prone.
> — Nikki, Manager Proces & Compliance HR/Backoffice, JAM! Horeca