Digitisation before automation: why the order matters
AI automates well only what it understands, and it understands only what is clean. Digitisation → context → automation, explained on one real case.
Read →Guide · How to digitise
Which processes a company with 20–150 employees can automate, how to start in four weeks with a single process and what it costs, with our 2026 prices.
In short
Updated September 2026 · 11 min
The team of a company with 20–150 employees loses hours every week checking documents, chasing approvals and copying data between systems. Business process automation moves that work into workflows that run on their own, and on our platform the first automated workflow is usually the supplier invoice. This guide covers what can be automated in a company like yours, how to start in four weeks with a single process and what it costs.
Business process automation means that repetitive tasks with clear rules are executed by the system, without anyone doing them by hand. We are not talking about robots taking people's jobs. We are talking about five concrete things:
Many companies confuse digitisation with automation. Digitisation turns paper into files and ledgers into Excel; it is step one, without which there is nothing to automate. Automation is the step in which the files start processing themselves, and only that step brings the real time saving.
The difference shows on the same document. A PDF sent by email instead of fax is digitisation; the same PDF read by the system, assigned to a project and sent for approval on its own is automation. Whoever skips the first step automates the chaos, not the process.
| Digitisation | Automation | |
|---|---|---|
| Paper | Becomes a digital file | The file processes itself |
| Documents | Go out as PDFs by email, not by fax | The system routes them for approval automatically |
| Ledgers | Are kept in Excel, not on paper | The dashboard updates itself from the sources |
| Folders | There is a shared folder | Documents organise, validate and archive themselves |
Documents, approvals, reporting, system integration and customer communication are the five processes with clear rules and steady volume in almost any company. The industry does not matter: the same five appear at an MEP (mechanical, electrical and plumbing) contractor, a certification body and a distributor. Only the names of the documents differ.
Invoices, contracts and forms need reading, data extraction, checking and archiving. Optical character recognition (OCR) reads the document, and an AI Agent (artificial intelligence agent) extracts the amount, the supplier, the date and the line items. The system checks the tax ID, the amount limits and duplicates on its own, then sends the document for approval or straight to the archive.
Since invoices between Romanian companies travel through the national RO e-Factura system, regulated by Emergency Ordinance 120/2021, the data already arrives structured. The reading part gets simpler; the assignment to project, budget and approver remains to be automated.
A contract or a purchase order passes through three to five people, and the tracking happens by email, on WhatsApp and in meetings. Automation starts from written rules: below a threshold the direct manager approves, above it the director approves too. The system sends the case to the right person, tells them when it is their turn, escalates at the deadline and keeps the history: who, when, with what comment.
Every Monday morning someone gathers data from several sources into a report for management. Automated, the dashboard updates itself from all the sources, the report goes out by email at the set time, and an alert appears when an indicator leaves its range. Every manager sees what they need, without asking.
The data sits in the ERP (Enterprise Resource Planning) system, in the accounting software, in Excel and in emails, and someone copies it between them. Connectors through application programming interfaces (APIs) sync it automatically, so there is a single source of truth. Nobody reconciles different figures at month end any more.
The order confirmation, the delivery status, the invoice and the payment reminder are emails written by hand. Automated, the customer gets a notification at every stage, the invoice goes out after delivery and the reminder at the due date. A portal shows them the status without asking.
The same five processes have different names across industries. In construction they are the material approval form (MAF) and the progress statements; at certification bodies, the ISO files with multi-level approvals. In accounting they are supplier invoices and bank statement reconciliation; in retail, orders and stock synchronisation between locations.
Manual
Automated
The first automation project has a single process, four weeks from mapping to launch and a period running in parallel with the old process. Not every process deserves automation. Good candidates are repetitive (daily or weekly), rule-based, hour-consuming, error-prone because of volume and critical for the business, meaning their delays are felt.
The simplest exercise: each manager lists the three repetitive tasks that consume the most of their team's time. You have a list of ten to fifteen candidates within a day. From it you choose by two criteria, impact and complexity, and start with the process that has high impact and low complexity.
Before automating, you document the process as it is, not as it should be. Who performs each step, what they receive and what they deliver, how long it takes, which exceptions occur and where it gets stuck. The workarounds and temporary fixes go into the document, because they are the real process.
You choose the approach by the team. No-code tools work for simple automations, if you have someone with time and a technical bent. An implementation partner builds for you when you have no IT team; enterprise platforms make sense at very high volumes, not at a company with 20–150 employees.
Whatever the choice, the implementation is iterative: one process, four weeks, then the next. The old process runs in parallel for one or two weeks, with the results compared, before you switch it off.
The first process, in four weeks
Each manager lists the three repetitive tasks that consume the most of their team's time.
You pick the process with high impact and low complexity; the rest wait for the second round.
You write down who performs each step, what they receive, what they deliver, which exceptions occur and where it gets stuck.
No-code tools in-house, a partner who builds for you or an enterprise platform.
Week 1 mapping, week 2 design, weeks 3–4 build and launch, then 1–2 weeks in parallel.
The cost depends on the approach: no-code tools paid monthly, a partner project of a few thousand euros or an enterprise platform with annual licences. The first option costs little in money and a lot in someone's time on the team. The third costs more than all the work it would replace at a small company.
For the middle option, our prices, 2026, for a typical project:
| What you buy | Price |
|---|---|
| Assessment and process mapping | €1,000–2,000 |
| Automation of 2–3 simple workflows | €5,000–8,000 |
| Complete platform, 5–7 workflows | €15,000–25,000 |
| Maintenance | 10–15% of the initial cost, per year |
The cost rises with integrations into legacy systems without programming interfaces, with processes full of exceptions, with the need for extensive training and with industry-specific compliance requirements. The cost drops with simple processes with clear rules, modern systems that connect easily, a team open to change and the step-by-step approach. The most expensive process is the one documented wrongly: it gets built twice.
Automation fails for five reasons: a badly designed process, automating everything at once, missing users, underestimated exceptions and no maintenance budget. None of them is about technology. All of them show in the first week, if you know where to look.
A chaotic process full of workarounds is not fixed by automation; the chaos just runs faster. Simplify first, automate after. Large projects fail for the same reason every time: they take too long, the requirements change along the way, management loses interest and the team sees no win.
The team that will use the system joins the project at mapping, not at training; otherwise you get resistance to adoption, not a workflow. “The process is simple, just five steps” holds until the first invoice in the wrong format, the first approver on holiday, the first customer asking for changes. All the exceptions are documented, with a decision for each one: automated or escalated to a person.
Automation is not “set it and forget it”. Systems change, requirements evolve, errors appear, and without a maintenance budget the workflow stops quietly after a few months.
Before you automate
An MEP contractor with 45 employees receives supplier invoices by email and moves them through a single workflow, from arrival to payment. The company is fictitious; the workflow is the one we build most often, because it has clear rules and steady volume. Before, the invoice passed through four pairs of hands and nobody knew where it was:
| Manual step | Hours per week |
|---|---|
| Reading the invoices and typing them into the ERP | 6 |
| Looking up the project and checking the budget | 2 |
| Chasing approvals by email and phone | 3 |
| Reconciling with accounting at month end | 3 |
| Total | 14 |
Fictitious data.
The new workflow starts from the inbox: invoice FF-2291 arrives from Instal Group Ltd with the delivery note attached. The AI Agent extracts the type, supplier, total and project, checks for duplicates and proposes the case. The bookkeeper confirms or corrects with one click, without typing anything.
The case is assigned to the Northgate warehouse project and its budget, and the system flags that the budget is almost spent. The project manager, Robert T., receives the case for approval with all the data in front of him. After approval, the payment is scheduled in the ERP at the due date, and accounting sees everything without emails.
Of the 14 hours per week, what remains are the minutes of confirmation and approval, and the invoice no longer gets lost between emails. You see the same picture in any workflow with clear rules; only the name of the document differs.
Automation makes sense for processes with stable rules and steady volume; in three situations it costs more than the work it replaces.
List the five repetitive processes in your company and write next to each how many hours a week it consumes; that one-hour exercise shows where to start. If you want to understand why the order matters, the guide on digitisation before automation explains what has to exist before the first workflow. This is how we work: one process, four weeks, a fixed price agreed before we start.
A single process with clear rules is automated in four weeks: one for mapping, one for design, two for building and launch. Then come one or two weeks in which the old process runs in parallel, with the results compared. A process with many exceptions or legacy integrations takes longer, so you document it before promising a date.
It depends on the approach. No-code tools are paid monthly and mostly cost the time of someone on your team. With a partner, our 2026 prices are €1,000–2,000 for the assessment, €5,000–8,000 for two or three simple workflows and €15,000–25,000 for a complete platform, plus 10–15% per year for maintenance.
The one with high impact and low complexity. It is usually an approval workflow or supplier invoice processing: it has written rules, steady volume and everyone knows how much time it eats. Ask each manager for the three repetitive tasks that consume their team, put them on an impact–complexity matrix and start from the favourable corner.
They are all documented before the build, with a decision for each one: the system handles it on its own or escalates it to a person. An invoice in the wrong format, an approver on holiday or a customer asking for changes are normal cases, not surprises. A workflow with no path for exceptions stops at the first one.
No, if you work with a partner who builds and maintains the workflow; you need one person who knows the process and can decide. No-code tools need someone on the team with time and a technical bent. Enterprise platforms need a dedicated IT team and make sense at volumes far above a company with 20–150 employees.
One workflow goes live on a real project, with a success criterion set together.
One email a month, only when we publish. Nothing else.
AI automates well only what it understands, and it understands only what is clean. Digitisation → context → automation, explained on one real case.
Read →Performance drops in the first weeks after go-live and climbs above the old level from month three. The stages, the critical moment and seven strategies.
Read →Linear and scalable growth are the two ends of one spectrum. Four questions show where your business sits and how automation moves it to the right.
Read →