AI integration for your app or web application
Already have an app or web application, and want AI in it? Then we start with one feature you can measure. If it works, we build on.
You hear about AI everywhere. Your CRM vendor suddenly has a button for it, and someone on your team asks whether you should be doing something with it too. Fair question. So where do you start? With one task in your software that costs time today. We make that one smarter. Then we look together at whether it pays off.
Which AI features fit your product?
What can AI actually do in your software? In practice we see a few kinds of features come back again and again.
Answers from your own data
A language model doesn't know your business. Your price list and your terms aren't in it. Ask about them anyway and it makes up something that sounds plausible. With RAG, the AI first looks up the right pieces in your documents. Only then does it answer, with a source attached.
It all depends on your sources. Three versions of your terms in the folder? Then it can grab the wrong one. Cleaning up is often more work than the tech itself. More in RAG explained.
Summaries and drafts
A customer file with twenty notes. A ticket with a long email thread. The AI puts a summary on top, and your colleague only digs into the details when needed.
Drafts work the same way. The AI writes a first version of a reply or a quote text. A person reads it and edits it. Only then does it go out.
Connections to the systems you already have
Does the AI need to look into your CRM or your planning? Then you need a connection. More and more often that goes through MCP, an open standard. You build one layer on top of your software, once. You decide which actions the AI may take there.
We almost always start with read-only. Nothing can break that way. If the AI needs to write something, it makes a draft and a person confirms it first. More in MCP: how to connect AI to the software you already use.
Where in your product does it fit?
In a web application or customer portal, often best. That's where your team works all day. In an app it works too, for instance smart search for your customers. And on a website, an assistant that answers questions from your own information is a logical first step.
One feature first, then measure
Why not put it everywhere at once? Because you don't know yet how AI behaves with your data and your users. You only learn that by doing it.
So pick one task. Ideally one that happens often and where you can check the outcome. A mistake should also be easy to fix. Measure how it goes now, first. How long does the task take, and how often does it go wrong? Without a baseline, all you can say later is that it feels better.
With us, a first working AI feature is usually ready in 2 to 4 weeks. We put it behind a switch, so you can go back if it disappoints. A small group works with it for a few weeks first. Then we compare with your baseline.
Is the result clearly positive? Then we roll it out. In doubt? Then we adjust it or stop. That's an outcome too, and cheaper than letting something run for a year. The full roadmap is in adding AI to your existing web application.
Privacy and picking a model
What happens to your data? Everything you send to a model ends up with the company behind that model. So don't send more than needed. Arrange a data processing agreement and read the business terms. Those usually rule out training on your data. With free versions that's often different.
Which model do you pick? That depends on the task and the cost. Where your data is allowed to live counts too. The first pick matters less than you'd think. We always put the model behind a layer of its own in your software. That's where the prompts and the cost tracking live. Want to switch models later? Then you change one place.
Count on running costs for the model as well. With normal SME use it's often around €50 to €500 a month.
Which AI feature should you build first in your app or web app?
A task that happens often and where you can check the outcome. Summarising is often a good candidate, for instance of a customer file or a long email thread. A mistake there is quick to spot and easy to fix. Ideally pick the task your team complains about most. You measure that one first, before the AI comes in.
Can AI give answers from our own documents?
Yes, with RAG. The AI first looks up the right pieces in your documents and only then answers, with a reference to the source. It works as well as your sources do. So clean up outdated versions of your terms first. Test it with a fixed list of real questions, and run it again after every change.
Does our customer data go to the AI model?
Only what you send along. Everything in a request to the model ends up with the company behind it. So don't send more than needed. Arrange a data processing agreement and read the business terms. Those usually rule out training on your data. With free versions that's often different.
What does AI in my software cost per month?
With normal SME use you're often at around €50 to €500 a month in model usage. That depends on how many people use it and how heavy their questions are. A feature that calls a model on every page view drives that up fast. So keep track of it per feature.
Can AI in our software take actions too?
Yes, if you start carefully. Through a connection like MCP, the AI can look up data and prepare actions. We almost always start with read-only, so nothing can break. If the AI needs to write something, it makes a draft and a person confirms it first. Once that runs well for a while, we expand.
Where do you start with AI?
Book a call. We'll walk through your software together and find the best first task. Don't need AI yet? Then we'll just tell you.