Connecting AI to your existing software

Most companies already have their kit in place. An accounting package, an ERP, a CRM somewhere and a pile of Excel around it. And then a new AI tool arrives that knows nothing about any of it. So someone is retyping again from one screen into another. That is exactly the work you wanted to automate away. The gain is not in yet another standalone tool. The gain is in connecting: letting AI talk to what you already use.
Why a standalone tool rarely helps
A tool that sits alongside your existing systems does not make your landscape simpler but busier. You get an extra place to look, an extra place to keep up to date, and retyping between those places. That is double work with an AI sauce over it. A connection removes that retyping. The AI reads from your system, does its work, and writes the answer back to the place your team already looks.
What a connection actually is
In plain terms. Software talks to software through an API. Think of it as a hatch with fixed rules: you ask something politely, the other system answers politely. Accounting, ERP, CRM, most serious packages have such a hatch. If AI can plug into it, nobody has to move data by hand anymore. The system does it, the human checks the outcome. The nice thing is that you get to keep using your existing systems. You do not have to tear apart your entire accounting or ERP for an AI project. You put a layer on top that removes the boring back and forth. That makes the threshold a lot lower than people think, and the risk too, because you do not touch your source system.
Argo: contracts from CleanERP into a custom quote tool
At Argo, a cleaning company, clients and contracts sit in CleanERP, their own ERP. We are building a quote tool and contract management around it. The connection runs through that system's OData endpoints, so existing client data does not have to be re-entered. The old process was retyping: measurements in Excel, formatting in Word, printing, signing, scanning, emailing. Now the system reads what is already there, and the human keeps control over what comes out.
“A connection is only finished when nobody has to retype anymore.”
Be honest about the teething problems
I am not going to pretend connecting always goes smoothly. Our trickiest project was not about the technology of the connection itself, but about the party behind it. We were the first user of their API, so we hit every teething problem. Every hour of work, another new error. A job that was around 70 hours on paper took more than a year because of it. Constant ping-pong. That is the reality nobody tells you in a sales pitch.
Ask about the state of the API upfront
That experience did teach a hard lesson. Before you promise a connection, you want to know how mature the API on the other side is. Is there documentation? Are other parties already connected to it? Is there a test environment? If you are the first user, count on delays and say so upfront to the client. Do not promise a tight schedule on an API that still has its teething problems to work out.
The last 25 percent decides whether it holds
Prototyping a connection can be quick. But a connection that runs safely and stably in production, that is the last 25 percent. Setting permissions properly, not leaving secrets lying around in the code, handling errors from the other side gracefully, making sure it keeps working when the data flow grows. That is exactly the part vibe coders skip, and exactly the part our developers are for.
Start with the process, not the connection
As always: do not start with the connection but with the process. Map where the retyping really sits and what hurts most. Only once that is clear do you decide which connection removes that pain. Tinker with a connection without understanding the process and you build a fast link between two things that should have worked differently.
Stuck between your systems?
Recognise that retyping between your accounting, ERP or CRM? Chances are hours per week vanish into it that you never see back. Walk through it with us in an intro chat. We look honestly at whether a connection pays off, and if it does not, we simply say so.

Marc bouwt dagelijks met AI en helpt bedrijven om ideeen snel om te zetten in werkende tools.
Follow on LinkedIn ↗
5 AI quick wins that save time today

How do you measure the ROI of AI?

How to have your own internal tool built

12 concrete AI applications for SMEs

How do you get your team on board with AI?

What does custom software with AI cost?

What does implementing AI cost?

How to automate your processes with AI

Building your own CRM with AI

What does an AI workshop truly deliver?

Where do you start with AI in your business?

Is the code from your AI tool truly safe?

Why AI projects truly fail (it's not the tech)

Claude vs ChatGPT: what works better for business?

How we built the Bietenradar (and why)
