From Excel and email to a workflow: choose the first step
LAST UPDATED ·
Why data entry consumes the hours
In a business with 10 to 100 employees, information arrives by email, PDF, Excel, and phone. It has to end up in the ERP, the accounting package, the TMS, or the planning. The part in between is almost always a person retyping.
That is not strange; it is how those businesses grew. But it adds up. A February 2025 Visma study found that 41% of Dutch business owners spend more than half their week on administration. And according to Statistics Netherlands (CBS), only 3% of companies use process automation (RPA), while administration is the second-largest area of AI use. The gap sits exactly there: everyone knows the admin is too heavy, and almost nobody has automated the part in between.
Three signs of a good first process
Not every process is a good starting point. We look at three things:
1. Repeated. It happens dozens of times a week or more. Preparing annual accounts once a quarter is not a candidate; 180 orders a day is.
2. Regular. The steps are roughly the same every time. If an employee can explain in one sentence what they do (“I open the PDF, look up the client, type in the lines”), that is a good sign.
3. Structured source. The information is already digital and has a recognizable form: a PDF order, an Excel price list, an email with fixed fields. Handwritten notes and phone calls are harder; those come later.
Does a process score on all three? Then it can usually become a working prototype in 10 business days.
From source to destination: email, PDF, Excel, and ERP
Almost every first automation follows the same four blocks. As a diagram: Source (email, PDF, Excel) → Read and recognize (AI step) → Human review → Destination (ERP, accounting, TMS). In the old situation the bottleneck sits in blocks two and three, where a person reads and retypes. In the new situation an AI step reads, and the person only looks at what deviates.
In steps, it looks like this:
- Pin down the source. Which mailbox, which folder, which portal. One entry point.
- Read. An AI step pulls the fields out of the source: client, item number, quantity, date. Even when the layout differs per sender.
- Match. The fields are compared with what is already in the destination system: does this client exist, is this item number right.
- Review. Everything that is certain goes through. Everything doubtful lands on a screen for an employee.
- Write. After approval, the order, invoice, or request goes into the destination system, with a log line.
- Confirm. A confirmation to the client or colleague, automatically.
Where people remain in control
A good automation removes the retyping, not the judgment. People stay in three places:
- Approval. Before anything goes out the door or costs money.
- Exceptions. A new client, an unusual price, an unreadable attachment. Those go to a queue.
- Improvement. The employee sees which exceptions keep coming back and passes that on; the rules are adjusted.
That is also why teams are rarely against it. You take away the dullest part of their day and leave the part they are good at.
Tool categories
You do not need to know the tools to choose a process, but it helps to know there are categories:
| Category | Among others | When |
|---|---|---|
| Workflow platforms | n8n, Make, Microsoft Power Automate | Connections between email, files, and systems with an API |
| AI models for reading and drafting | Models from OpenAI, Anthropic, Google, and EU-hosted open models | Pulling fields out of unstructured documents, draft texts |
| Integrations | APIs of Exact, AFAS, Twinfield, TMS/WMS; file import where there is no API | Writing to the destination system |
| Custom code | Small scripts or a small web app | When the review screen or the logic is too specific for a platform |
The choice follows from the process and your existing systems, not the other way around. More about what we connect is on the system integration page.
Measurement: hours before and after
Without measuring, you do not know whether it worked. Our yardstick is simple:
- Hours a week on the process, counted with the people who do it, before the build.
- Lead time from arrival to processed.
- Error rate (wrong lines, returns, disputed invoices).
- The same three, four weeks after go-live.
Write down the baseline before you start. That is also the moment you decide whether it is worth it: from about 5 hours a week of repeated manual work, a prototype is usually defensible.
Example
At a technical wholesaler with 45 employees, three sales support staff together spent 42 hours a week retyping orders from email and PDF into Exact. After a scan and a 10-business-day prototype on one client group, extended to all email clients, sales support is at 9 hours a week and 78% of orders go through without correction. The figures are illustrative; the process is one you will recognize. Read the example case on order entry from email. A similar pattern, but for collecting documents from clients, is in the accounting firm case.
FAQ
FAQ
Do I have to replace my ERP or accounting package before I can automate?
No. Almost every automation in a small business sits between existing systems. Exact, AFAS, Twinfield, and most TMS and WMS packages have an API or an import. Where one is missing, a file or a mailbox works as the bridge.
How many hours a week should a process cost before automating pays off?
A rule of thumb: from 5 hours a week of repeated manual work with a structured source, a prototype is worth it. With fewer hours it can still be useful when errors are expensive.
Do jobs disappear when we automate a process?
Not in our experience. The retyping disappears. People stay for review, exceptions, and client contact. Those hours go to work that makes money.
