AI is everywhere. Every software vendor claims to be 'AI-powered'. But when you dig deeper, it often turns out to be a chatbot that can search manuals. Or a system that makes 'smart suggestions' that nobody uses.
We work daily with AI at Klusmaat. This is honestly what we've learned about what works and what doesn't.
What works: speech to action
The killer app for tradespeople is speech recognition that actually understands what you say. Not transcription - but interpretation.
When an electrician says 'The Jansens need a new distribution board, three phase, thirty groups, next Tuesday', the software should understand: this is a new job, for an existing customer, with specific requirements, and a desired date.
This sounds simple, but it's fundamentally different from traditional speech recognition. It's not about converting words to text. It's about understanding intent and executing actions.
What works: intelligent scheduling
Scheduling is one of the hardest parts of an installation company. Who goes where? Which job first? How much travel time between them? Which technician has the right certification?
AI is brilliant at this. It can calculate thousands of combinations in seconds. It considers travel time, traffic, skills, and available materials. And it learns from what goes wrong: if a certain job always runs over, the system adjusts its estimates.
This isn't futuristic technology - this works now. Every day we automatically schedule hundreds of jobs, and the scheduling is better than what people can do manually.
What doesn't work: generic chatbots
The most overhyped AI application is the chatbot. 'Ask your question and get an instant answer!' Fantastic in theory. Frustrating in practice.
The problem is context. A chatbot doesn't know who you are, which jobs you have running, or what your company does. If you ask 'what does the material for tomorrow cost?', the system needs to understand: which job, which material, which prices apply for this customer.
Generic chatbots can't do that. They're trained on general data, not on your specific situation. The result: answers that are technically correct but practically useless.
What doesn't work: AI as a marketing term
Too much software sticks 'AI' on features that aren't AI. A filter that sorts invoices isn't AI. A search function that corrects typos isn't AI. Automatically sending reminders isn't AI.
This dilutes the term and makes users cynical. Rightly so. If everything is 'AI-powered', nothing is anymore.
The real promise
The real promise of AI in construction isn't one smart feature. It's the combination of dozens of small automations that together save hours per week.
A phone that's automatically answered. A quote that writes itself. An invoice that's sent without you thinking about it. Inventory that replenishes itself. A schedule that optimizes itself.
None of these things is spectacular on its own. But together they fundamentally change how much time you spend on administration. And that's what it's all about.