AI systems for business processes.
We connect the programs you already run and set up AI systems for business processes - on their own or alongside miphu. The recipe behind it is the same every time.
- ChatGPT
- Claude
- Gemini
- Mistral
- DeepSeek
- Perplexity
Every process is built to the same recipe.
In almost any office routine, information gets entered, worked on or filed. That is where we come in, and what we build for it is made of the same five parts.
01
Data source
Where the information comes from: a mailbox, a form, a folder, a database, an interface.
02
One job
The model gets exactly one task, and the rules it decides by while doing it.
03
Tools
What it may use for that - calculate, search, read an invoice, produce a document.
04
Connections
The programs the work ends up in: ERP, bookkeeping, mailbox, file storage.
05
Result
A draft, an entry, an answer - with the source it rests on next to it.
Where AI already takes work off the table.
Four processes, built to this recipe. Each one says which data goes in, what the AI does with it, what it is connected to and who releases the result.
A quotation from a drawing and an enquiry
- Input
- An enquiry by email, with a CAD drawing, a bill of quantities or a photo attached.
- Job
- Read the attachments, look up comparable jobs and propose the costing basis from them.
- Connections
- The ERP, the CAD archive, prices and availability, your own quotation template.
- Result
- A draft quotation in your template; whatever stays unclear is flagged rather than guessed.
- Release
- Sales. A quotation never leaves the building unseen.
Reading and posting invoices
- Input
- Invoices and delivery notes from mailbox, post and portal - PDF, scan, photo.
- Job
- Read them out, match supplier, order and cost centre, reconcile against goods received.
- Connections
- Mailbox, orders and delivery notes, DATEV or your own bookkeeping.
- Result
- What is unambiguous goes through posted. Every discrepancy lands in the query loop with a reason.
- Release
- Bookkeeping, on every discrepancy.
An answer out of your own records
- Input
- Network drive, mail archive and DMS, filed exactly as they are today.
- Job
- Understand the question, find the relevant passages and assemble the answer from them.
- Connections
- The existing archives. Permissions stay as they are.
- Result
- An answer, with document, passage and revision beside it.
- Release
- None - nothing is created. Whoever passes the answer on can look it up beforehand.
Sorting the shared inbox
- Input
- One shared mailbox holding orders, enquiries, invoices and advertising side by side.
- Job
- Sort every message, attach files to the right job, prepare the standard cases.
- Connections
- The mailbox, jobs in the ERP, who is responsible for what.
- Result
- A sorted inbox and draft replies that stay unsent.
- Release
- Whoever is responsible. On anything that leaves the building, a person stays the sender.
Connecting to your ERP
We add to the ERP you already have, or we connect other programs to miphu. Where there is no ERP yet, there is no system for quotations, orders and invoices to come together in - then the ERP comes before the AI project. miphu is our own: a finished core for customers, quotations, orders and invoices, one instance per business.
How the ERP fits inStandard functions are maintained as part of the product updates. Custom extensions are looked after as a separate service.
How a project runs
- 01
Discovery call
We go through the routine in question: where the time goes and which programs hang off it. The call is free, and afterwards we put in writing what is worth automating.
- 02
Audit or build
Where the routine is clear, we build it and connect it to the programs you already run. Where several routines are in play and the order is open, the AI process audit takes them in first.
- 03
Operations
Models change, interfaces disappear, processes move. We monitor, keep up and keep developing.
The business we can show is our own.
We run our company on software we wrote ourselves: quotations, invoices, projects, bank reconciliation and time tracking all sit in miphu. The routines around it we automated ourselves - the same work we do for clients.
We cannot show a client project from this line yet. What stands here is our own operation.
Contact
In the discovery call we go through your routines and put it in writing: what is worth automating, and roughly what that costs.
