Automation & AI work companion
Intelligent digitisation for the trades
On the office shelf: twenty years of project folders. On the server: "Photos_new_final", 40,000 images with no assignment. In the desk: measurement pads with pencil notes only one person can read. All valuable – none of it findable.
Intelligent digitisation
01Intelligent digitisation
Today AI reads even difficult scans and handwriting: measurements become structured dimensions, delivery notes become bookable line items, site photos sort themselves by project and trade, and old project files become a searchable history – including the post-calculations your rules of thumb come from. We set this up as batch processing: the stock runs through, someone from the office checks samples, and in the end the data sits where you work – in your trade software, a clean table or as the basis for a knowledge base. For a business that concretely means: the question "what did we install at Berger's in 2019?" costs a search, not an afternoon at the filing cabinet.
Straight up: there's no 100% recognition – with handwriting especially, sample checking is always part of it. What may go to the cloud and what stays local we settle BEFORE processing. Audit-proof archiving (GoBD) stays the job of your specialist systems – we make stock usable, we don't replace an accounting archive. And if a stock isn't worth the effort, I'll say so.
02Demo
Three documents of the kind lying around in any business. Pick one – on the right you see what the recognition makes of it. Watch the flagged field: that's exactly what the spot-check is for.
The route behind it is a pipeline: recognition and rules work through the stack – and anything uncertain branches off to a human instead of quietly landing in the table. Only once the spot-check rate holds does the whole batch run through.
The dashed loop is the core: what's uncertain goes to the check and back – so the recognition gets better with each stack.
03How it works
- 01
Survey the stock
Project folders, photo server, measurement pads, delivery-note filing – what sits where, in what quantity, and what should the data be good for afterwards? An afternoon of surveying saves weeks in the wrong direction.
- 02
Set the target structure
Where should the data go – trade software, table, project filing? And which fields are needed: project, client, trade, dimensions, amounts, date? The target structure decides what the recognition has to look for.
- 03
Pilot with a real batch
We take a representative slice – one project year, one folder type, one photo month – and set recognition and sorting on it. The pilot shows what the stock gives up.
- 04
Check quality and sharpen
Someone from the office checks samples: are dimensions, line items, project assignment right? Where the recognition misses, we sharpen rules – until the rate holds.
- 05
Process the whole stock
Then the whole stock runs through and lands in the target system. On request the path stays in place: new delivery notes, photos and scans are processed along automatically from now on.
04Stages
The entry point is deliberately cut small – a single stock is a perfect, measurable pilot.
Entry: one stock, opened up once
A clearly bounded stock – a year's delivery notes, a photo archive, a shelf of project files – is processed once. The result: a structured, searchable filing.
Extended: the running path
The one-off action becomes a repeatable pipeline: new delivery notes, site photos and scans land automatically structured in the right place.
Full build: the stock works along
The opened-up project data flows into a knowledge base: post-calculations, material lists and project history become queryable – the business's experience, on call for everyone.
05Technical implementation
- Recognition: OCR plus AI extraction – modern vision models read even skewed scans, stamps, carbon copies and handwriting portions on measurement sheets.
- Classification: document types (delivery note, invoice, measurement, photo, note) are recognised automatically and sorted by project/trade; duplicates fly out.
- Photo stocks: image recognition sorts by site, site-log period and subject – "Photos_new_final" becomes a project gallery.
- Processing as a pipeline: scripts or a small processing path – local or in the cloud, depending on the data-protection situation of your material.
- Review interface: a simple view for samples and corrections. The human stays in the loop – that's the concept, not a compromise.
- Output: structured to where you work – trade software, CSV/table, project filing or knowledge base.
for context · This shows how something like this usually gets built – not how it has to be built for you. What actually fits is worked out in the project: around your existing setup, your team, and what stays easy to maintain.