How to connect your ERP, CRM and Excel so nothing is entered twice

An order comes in from the online shop. Someone retypes it into the ERP, creates the customer in the CRM, issues the invoice in the accounting system and, at the end of the week, adds it all up again in a spreadsheet. The same data five times, by five pairs of hands. System integration has one goal: each piece of data is entered once and reaches the other systems on its own.

The short answer to the question in the title: first decide on a single source of truth for each type of data, then connect the systems in the simplest way available so that data flows in one direction. Classic integration through an API, a ready-made connector or a tool like Make or n8n usually does the job. AI only makes sense where the data is unstructured — emails, PDFs, free text — or where records need matching that do not line up exactly.

Excel does not have to go. It just becomes an output for reports rather than a place where data is retyped by hand.

Why the data drifts apart

Double data entry is not only a waste of time. It is a source of contradictions. When the same customer address is written into three systems, a year later you have three different addresses — someone corrected it in one system and not in another. The shop shows 12 in stock, the ERP shows 4, and a customer orders something you do not have.

The most expensive part is not the hours of retyping. It is the meetings spent not on what to do about the numbers, but on which spreadsheet is right.

Step 1: a source of truth for every piece of data

Before the first line of integration is written, you need to decide where each piece of data is created and where it may be changed. The other systems only receive it. A typical setup looks like this:

Your table may look different. What matters is that every row has one source. Two-way sync (“a change anywhere is copied everywhere”) sounds appealing, but in practice nobody knows which change won. Where it cannot be avoided, it needs a clear rule for conflicts.

Ways to connect systems

Once it is clear what flows where, you choose the technique. From the cleanest to the fallback:

To be honest: most integrations are classic integration with no AI. Moving an order from the shop to the ERP is a precisely defined task — field A goes into field B. It does not need a language model, and adding one would mean paying for uncertainty where you could have certainty.

A simple rule helps with choosing the tool: if a ready-made connector or Make can handle the flow, start there. Custom code makes sense when the number of scenarios grows, when retries after outages, state and error reporting need handling, or when data must not leave your infrastructure.

What to do when a system has no API

In Slovakia and the Czech Republic this is the rule rather than the exception. The good news: something can almost always be done.

• Pohoda has XML import and export, which can move orders, invoices, the address book and stock levels. Higher editions also include mServer, which accepts the same XML over HTTP — close to a real API.
• Money S3 also supports XML data transfer. The newer Money S4 and S5 offer an API as well, which makes integration simpler.
• Legacy and in-house systems usually sit on a database that can be read. We recommend writing straight to the database only in exceptional cases — it bypasses the checks the system does itself.
• The last resort is a scheduled file export dropped in an agreed location, or a robot clicking through the user interface (RPA). It works, but it is the most fragile option and needs thorough monitoring.

Every fallback route has the same requirement: transfers must be logged and someone must be alerted on failure. A silently failing import discovered at month-end closing is worse than retyping by hand.

Where AI helps

AI is not better integration. It is a tool for the places where rules fall short:

• Unstructured input. An order arrives by email instead of through the shop, a delivery note as a PDF, a request in free text. AI extracts structured data from them, which then flows the same way as everything else. We cover this in detail in the article on automating order processing from email.
• Matching ambiguous records. “ABC Ltd”, “ABC Limited” and “Abc Bratislava” are probably the same customer. The same goes for a product named differently in the shop and in the ERP. Rules handle exact matches; AI proposes the rest and a person confirms them.
• Questions across connected data. Once systems are connected, you can build an assistant that answers “which customers have an unpaid invoice and an open order”. This is where MCP — the protocol AI uses to plug into business systems comes in.

When it is not worth it

• A handful of records a month. If ten invoices a month move between systems, integration will not pay back. Discipline and one checking spreadsheet are enough.
• Too many systems. Sometimes the better fix is to switch one off — using the CRM module in your ERP instead of a separate CRM, for example — rather than connecting them.
• The process is about to change. If you are replacing the ERP in six months, connect only what is essential and design the rest for the new system.
• One person uses Excel for their own analysis. That is not a problem that needs solving.

What it costs

Indicatively, excluding VAT, for a small or medium company: a map of systems and data flows €900–1,800, the first connection between two systems (for example shop ↔ ERP) €2,500–6,000, further systems plus AI for emails, PDFs or record matching €5,000–12,000, operations and maintenance from €290 a month. The biggest cost driver is whether the systems have an API — and the state of the data being connected.

With two or three people spending 30–60 minutes a day moving data around, we plan for savings of around 30 hours a month. Part of the benefit does not fit into hours: fewer wrong invoices, stock figures that are right, and numbers nobody argues about.

Summary

• A source of truth for every piece of data first, technique second. Without it, integration just speeds up the chaos.
• Let data flow in one direction. Two-way sync only with a clear conflict rule.
• Start with the simplest route: an API, a ready-made connector, Make or n8n.
• No API is workable too — XML import (Pohoda, Money S3), database reads, file exports. With thorough monitoring.
• AI only where rules fall short: emails, PDFs, matching ambiguous records, questions across data.

We describe how we design this kind of integration on the system integration page, and the technical side (APIs, MCP servers) on the connecting AI to your systems service page.

How many systems does someone at your company type the same order into? Let’s go through it together — and if a connector for a few euros a month will do, we will tell you.