Your accounting package is fine. Your ERP is fine. The problem is the twenty spreadsheets in between, where the month-end, the order list and the report to head office actually get made. This page is about which of those to replace first, what that costs, and how to do it without a year-long software project.
By Tom Joseph · Last updated: 8 September 2026
Replace the spreadsheet that carries the most hand-typed data between two systems, not the biggest one. Every operation we visit has one workbook that a director or CFO personally maintains, that pulls numbers from Exact, AFAS or an ERP, and that someone re-types into a report for a head office, a bank or a board. That one is worth automating first: it has a clear owner, a clear deadline and a clear error cost. At Crux Digits a fixed-price scan of €2,500 finds it, a €20,000 Production-ready MVP replaces it with a working flow on your own data, and production from €50,000 covers the connections, the checks and the handover.
Where the spreadsheet lives
None of them is a spreadsheet problem. Each is a process that never got a system.
Exported from the accounting package, re-typed into the group template, checked by the CFO on a Sunday. Same numbers, four hands.
Email in, spreadsheet out, warehouse copies it again. Three versions by noon and the customer gets the oldest.
Colour-coded, macro-driven, understood by one person. When they are on holiday the plan is on holiday.
Purchase invoices matched to orders in a lookup that breaks when a supplier renames a column.
A workbook with eleven tabs is not documentation. It is the process, and it leaves when its owner does.
Excel is where a process goes when no system was built for it. That is not a criticism of the people who built the workbook; they solved a real problem with the tool they had. It becomes a problem when the workbook is the only place the process exists: the rules live in formulas nobody reads, the data lives in copies nobody reconciles, and the knowledge lives in one person.
The owners and directors who contact us describe it the same way every time. "Too much manual work." "Reporting to the head office is not automated." "The CFO spends a week a month on it." The phrase they do not use is "we need a system", because they already have three. What they need is the connection between them, and the handful of rules that currently run in someone's head.
The public numbers say the same thing from the other side. Statistics Netherlands reports that 29.8 percent of Dutch companies with 10 to 249 staff used at least one AI technology in 2025, and that among micro-enterprises the fastest-growing technique was text mining (from 3.1 to 9.8 percent in three years) while robotic process automation stayed flat at 2.3 percent. Firms are automating, but through reading documents and connecting systems, not through bots that click on spreadsheets.
Score every workbook in the building against these four. The one that hits three of them is your first project.
Almost never a new ERP. The replacement is usually three small things: a connection that reads the source system directly (Exact Online, AFAS, e-Boekhouden, Twinfield and most webshop and WMS platforms have APIs), a set of rules written as code with a test for each one, and a place where the result is checked and approved before it goes out. That last part matters more than the automation: the four-eyes step stays, it just stops being the whole job.
For the report to head office that means the numbers are pulled at 06:00 on the second working day, the mapping to the group template runs as rules, the exceptions land in a short list for the CFO, and the file goes out after one click instead of a weekend. For the order list it means email and webshop orders land in one place, the warehouse sees the same list, and the customer gets the version that exists.
Where AI enters is the unstructured part: the supplier invoice in a new layout, the order that arrives as a sentence in an email, the contract with the delivery date on page four. A language model reads that into the structured fields, and the deterministic flow takes it from there. Our AI automation work is built on exactly that split, and the API integration page explains what the connections cost and where they break.
Anonymised, but real. Each of these started as a director describing a spreadsheet.
| The situation | What was replaced | Result |
|---|---|---|
| Energy company, subsidiary of a foreign group. Monthly reporting to HQ assembled by the CFO from the accounting package and four operational sheets. | A read connection to the accounting package, mapping rules for the group template, an exception list, one approval step. | Reporting from five working days to one morning. The CFO reviews exceptions, not cells. |
| Medical publisher and agency. Sponsor bookings, ad placements and webinar registrations tracked in separate sheets by sales, editorial and events. | One record per sponsor deal, fed by the CRM and the webinar platform, with the invoice run generated from it. | No more reconciling three lists before invoicing. Sponsor statements produced from the record, not from memory. |
| Ingredients plant. Incoming drums logged on paper, put-away decided by the warehouse team, stock position kept in a workbook updated at end of shift. | Goods receipt on a handheld, a put-away rule per product family, stock position live in the existing ERP. | Put-away decided by the system, stock visible during the shift, one fewer end-of-day reconciliation. |
Three prices, all fixed, all excluding VAT.
Three pieces that go into the specific jobs this page covers in general:
Almost never. The packages Dutch companies already run (Exact Online, AFAS, e-Boekhouden, Twinfield, most ERPs) have APIs. The spreadsheet exists because nobody connected them and wrote the rules down. That is the work, and it is a fraction of a migration.
A single report built from one or two source systems is typically the €20,000 Production-ready MVP: read connection, mapping rules, exception list, approval step, parallel run. If the report draws on five systems and three subsidiaries, it moves into a production scope priced against that.
The scan is one or two days. A first flow is four to six weeks, most of which is the parallel run. Access to systems (credentials, a test environment, a named contact at your accountant or IT provider) is what sets the schedule, not the build.
They gain it. The rules that lived in formulas become readable, tested and versioned, and the approval step stays with the same person. What disappears is the re-typing, not the review.
Yes, and that is usually where AI belongs in this work: turning an emailed order, a supplier invoice in a new layout or a contract clause into structured fields. The rest of the flow stays deterministic, so every number can be traced to its source.
Nieuwegein, in the Utrecht region, and yes. The scan is done on site by default, because the process lives with the people who run it, not in a document. We work across the Netherlands and the Benelux in Dutch and English.
A €2,500 process scan on site names it, writes down its rules and prices the replacement. If your existing package can do it with a setting, the plan says so.
Book a conversation