Plenty of business owners deliberately do nothing with AI. Out of concern rather than disinterest: for their brand, their client data and the reliability of what comes out. That stance is not a backlog. It is a reasonable response to a market full of big promises.
The scepticism is largely right.
The concerns we hear are always the same. What I put into a model like that, is it still mine? The answers sound confident, but are they right? And do I want a client to receive something no person has seen? Those are not fearful questions. They are exactly the right questions.
The figures back the sceptics up as well. According to BCG, only around 4% of companies get substantial value out of AI. The rest experiment, pay licences and see little in return. So anyone holding off is missing less than the advertising suggests.
Waiting for everything is a shame all the same. Within all that noise sits a small number of applications that demonstrably work today, in a company of five to fifty people too. And that distinction is easy enough to make.
What demonstrably works today.
1. Writing up conversations
After a client conversation you speak in what was discussed, or you record the conversation. Back comes a summary, a list of actions and a draft email for the follow-up. The work that would otherwise sit until the evening is ready before you are back at the office. This is the most underrated application, precisely because it sounds so unspectacular. Talking is faster than typing, and nothing gets lost between conversation and admin any more.
2. Drafts in your style, which a person always sees first
AI writes a first version of the email, the quote or the follow-up, in your tone and your words. Nothing leaves without your approval, though. The draft sits ready, you read it, adjust it and press send. AI does the groundwork, you hold the pen.
3. Handling low-risk, reversible work automatically
Work that can do little harm and can always be undone, AI can do by itself: putting information in the right place, adding to a file, creating a task or a reminder. If it goes wrong once, you undo it and adjust the rules.
4. Scouting and pre-sorting
AI is good at looking patiently. Scanning sources such as TenderNed, pre-sorting requests and opportunities, and preparing a fitting proposal as a draft in advance. You are left assessing what is worth it, instead of searching yourself.
What is hype.
- AI that handles your client contact on its own. A bot that emails or chats with clients under your name, with nobody looking on. One odd reply costs more trust than a hundred good ones earn. What goes out belongs in front of a person first.
- All-in-one magic. One system that supposedly understands your whole company and arranges everything by itself. That is not how it works. Working AI consists of small, well-bounded tasks with clear limits, never one miracle cure.
- Promises with no process under them. Every demo is impressive. The question is what happens in week six, when the exceptions arrive. Anyone selling you AI without looking at your process is selling you the demo.
The difference is not in the AI, it is in the process around it.
How can the same tool deliver nothing for one company and a great deal for another? A field study by Harvard Business School and BCG among 758 consultants gives a clue: use AI well and you work around 25% faster, at higher quality. The same technology, a different outcome. The difference sits in the arrangements around it.
The question is not which AI you choose. The question is what AI is allowed to do by itself, and what a person sees first.
Arrangements like that are called guardrails: AI handles low-risk, reversible work itself, and everything that goes out sits ready as a draft and leaves only with your approval. Those guardrails are what makes AI usable rather than what holds it back: your team dares to work with it, because everyone knows where the limit lies. And they are the reason the four applications above work, and the three promises above them do not.
And what about your brand and your data?
That leaves the concern this piece started with. You solve it with how things are set up, not with a privacy promise in small print. Three starting points. Your data belongs to you and should not serve as training material for somebody else's model. Everything that touches your brand goes past a person first. And sensitive steps can run locally, in a setting you manage, so client details never leave the building.
Set it up that way and you can be careful and move forward at the same time.
Where do you start?
Not with a big AI programme. Start with work that costs time now and carries little risk: writing up conversations, preparing drafts, keeping files up to date. Agree in advance what AI may do itself and what you see first. And after a few weeks, look at whether it delivers anything, in hours and in calm. Small, bounded and reversible: it does not sound like a revolution, and it is how it works.
Questions about what AI can and cannot do in your company? Email hello@bravio.nl. You will have an answer within one working day.