Example · In-house AI systems
Internal AI Systems for Company-Specific Processes.
Your team checks every case in three programs: documents in the document store, master data in the ERP, notes in the mailbox. An in-house AI system pulls the data from the connected programs into one view, checks the clear-cut parts against your rules and hands your team the decision fully prepared.
Tasks, sources, status and approvals in one interface.
Validate, Research, Prepare
DMS, CRM, ERP, API
What you get
What an internal AI system takes off your team’s plate.
One view
Everything about the case in one place
Documents from the DMS, data from ERP and CRM, notes and status. Whatever is connected, your team sees in one view.
Fixed rules
The application checks the clear-cut parts
It checks completeness, deadlines and limits against your rules, the same way every time.
AI
Documents read, deviations flagged
The AI finds relevant passages and prepares drafts, with the source. It can misread or miss something, so your team checks.
Approval
Your team decides
Unclear cases go to the responsible person with a reason. Only after approval does anything go back into the ERP.
What it looks like
Checking delivery documents: three programs before, one view now.
The first delivery from a new supplier comes with a certificate and an inspection certificate. The application pulls both from the DMS and the purchase order from the ERP. Fixed rules check dates, the AI reads the documents. Your quality assurance makes the decision.
“Ordered: S355 round bar, 2 tonnes. Delivery arrives Thursday.”
Certificate on file, expiry date 31 Jan 2027 not passed.Fixed rule: standard and date
The inspection certificate for the delivery states S235, but S355 was ordered.Source: inspection certificate, page 1
The query to the supplier is ready as a draft.
The quality inspector decides on the delivery. Only then does the status go into the ERP.
Example with sample data, not a customer reference. When fixed rules are enough and when AI comes in is explained in our decision guide to AI agents and automation.
How it works
Four steps to the first usable version.
01
See how it works
We look at a real case: who does what, in which program, where does it get stuck?
02
Define the scope
What do fixed rules handle, where does AI help, who approves, what is deliberately left out?
03
Build the pilot
A first version runs with real cases, your team checks every result.
04
Decide
Review benefit, corrections and maintenance effort, then expand, improve or stop.
Is this right for you?
Off-the-shelf software, automation, knowledge base or internal AI system?
Off-the-Shelf Software
When your process is standard
If a ready-made program fits, it is usually cheaper. Then we tell you so.
Automation
When fixed rules are enough
Clear inputs, clear conditions, no custom program needed.
Knowledge base
When your team mainly needs answers
Questions in their own words, answers from your documents, with the source.
Internal AI System
When checking, searching and deciding come together
Your own rules, several systems and approval by your team.
Documents scattered or access rights unclear? Then get organised first. Or look at a worked example from sales:
Who is behind it
Advice and delivery from one source.
You talk to the person who actually builds your internal AI system.
Tested in-house
61 of 61 invoice numbers
In a test with our own documents, the process read all 61 invoice numbers correctly. Only the numbers were measured.
Already in use
A real project
APIORIUM runs a multilingual website chatbot with its own knowledge base.
Personal
Martin Trixner
Has been building software and integrations since 2002 and is a certified AI-Automations-Manager.
Details for IT and project leads
What is behind an internal AI system.
For anyone who wants the details: structure, permissions, operation and measuring success. Click to expand.
How is an internal AI system structured?
It has three parts. An interface where your team sees tasks, sources, status and approvals. The business logic, where fixed rules and AI tasks are kept apart and cases without a clear result are flagged for review. And the connection to DMS, CRM, ERP, databases and APIs, with the permissions that already apply there.
Who may do what, and what is recorded?
We do not copy the permissions from your programs blindly; we rebuild them and test them before launch, including the search index and logs. Inputs, sources used, suggestions, corrections and approvals are logged. Important or uncertain results have fixed review, escalation and stop routes.
Operation, costs and switching AI providers
Monitoring, error handling, maintenance and a possible change of AI model or provider are planned before the pilot. Running costs are shown separately from development.
How we measure success
Before the pilot we record the starting point and the goals. Which manual steps, search time and coordination go away, and where does new review work appear? How often are suggestions usable straight away? Do the connections run reliably, and does the team like working with it?
What needs to be settled before the start
The sources must be findable, technically accessible and approved for this use. And you need a responsible person from the business side who assesses rules, test cases and quality limits.
We are building our own internal AI system for sales and service by the same rules. It is still in development.
Frequently Asked Questions
Frequently asked questions about internal AI systems.
Do we have to replace our software?
No. The internal AI system works with your existing programs and accesses them through APIs. We check before the start whether a connection is possible.
What does an internal AI system cost?
That depends on scope, systems and number of users. After a look at your process you receive a written quotation, with development and operation listed separately. Not sure yet where AI pays off? Then start with the AI Opportunity Assessment at a fixed price of €1,900 excl. VAT.
Where does our data end up?
Where the AI runs, on your premises, on a server in the EU or with a provider, we decide together beforehand. Without explicit agreement, no company data goes to external AI services.
How much work is this for our team?
Mostly at the start: one person from the business side explains the process, provides sample cases and checks the results during the pilot. We handle the technical implementation.
Next Step
Which process costs your team the most time today?
Bring a typical case, anonymised if you like. In 30 minutes we work out whether an internal AI system fits or off-the-shelf software is enough, and what a first pilot looks like. In person with Martin Trixner, free and without obligation.