Almost every company now has “something with AI” on the agenda. Most projects don’t fail on the technology, but on the question of where to start. Our answer is simple: not with a strategy, but with one task.
No strategy, but a task
AI is a tool. Like a drill, it is only useful if you know which hole you want to drill. So choose a task that meets three conditions: it comes back every week, it demonstrably costs time and the result is easy to check. That last one matters, because an AI model makes mistakes. If a staff member can judge the result in a few seconds, that is not a problem. If no one can check it, it is.
Five applications that already work today
All built into the software you already use: your web shop, your CMS or your CRM. No separate AI tool, no extra login, and always with an employee who has the final say.
1. Products into your web shop faster
A photo, an article number and a few attributes. The model writes the title, the description and the specifications, picks the categories and stages the product in your own shop admin. The employee reads it, adjusts where needed and publishes. Adding a product takes a few minutes instead of half an hour.
2. Content and product texts in your own tone of voice
Newsletters, blogs, product texts, job ads: the data is there, the writing is not. The model writes a draft in your CMS in the tone of voice we captured beforehand, with your own examples as the basis. Someone approves before it goes online.
3. Social media scheduled and posted automatically
For an estate agency we built this into the website itself: the agent pastes a listing link, the application reads the details, the model writes a fitting post per channel and the agent approves. Everything is then scheduled and posted automatically. The same works for new products, articles or offers.
4. A smart assistant in your CRM
Before a conversation with a client, the assistant summarises the history, prepares a draft email and suggests a next step. All inside the CRM you already use, with the data that is already there. The employee decides what actually gets sent.
5. Reading documents into your system
Invoices, orders, registrations, forms. The model reads the document, extracts the data and stages it in your own application. The employee checks and confirms. No more retyping, and every document is registered the same way.
How to start: measure before, measure after
The order matters more than the technology.
- Choose one task from the list above, or your own variant that meets the three conditions.
- Measure the current situation: how much time does the task cost per week, how often does it go wrong, how long until something is handled?
- Build a first version in your existing software. No new tool, no extra login. That is achievable in a few weeks.
- Let two or three staff members work with it for a month and measure again.
- Only when the numbers add up do you expand to the next task.
Those who skip these steps and start big right away usually have a demo after six months, and no result.
The human has the final say
Everything we build has an approval step. The model prepares, a staff member decides. That is not a brake on the gain, it is the reason it works: errors are seen before they have consequences, and staff trust a system they can correct. We also log every call, so you can see afterwards what happened and what it cost.
About costs: use of a model is billed per call and in practice is a small part of the total. The largest part is in building it properly into your process. That is why we start small.
Conclusion
AI works in a business process when you put it on one concrete task, measure beforehand, let a human decide and only expand when the numbers add up. The five applications above are not future music; we are building them now. If you are unsure which task qualifies first for you, that is exactly the conversation we like to have.
