Customers
Systems in production, in our clients’ own tenants.
Pexon publishes six enterprise AI case studies from delivery inside industrial companies, banks and asset managers. Each one names the starting position, the system built, the trade-off taken and the measured change. Customers are described by category, size and regulatory context rather than named, because that is what the approvals cover.
Six programmes, six different starting systems.
Filter by the industry closest to yours and by the part of the stack you care about. Every card links the full write-up, and every figure on it appears in that write-up with the method behind it.
Showing 6 of 6 references
- TransformermanufacturerClient reference
Research per enquiry, from thirty minutes to seconds
An LLM-independent assistant in the manufacturer's own Azure tenant, indexing SharePoint and Confluence behind Entra ID so permissions arrive with the document.
−90% research time · EUR 3–5 per user, per month
Read the reference → - Cooperativebanking groupClient reference
One self-hosted platform serving 800+ client banks
Semantic chunking and pgvector with HNSW indexing on the provider's own OpenShift, with tenant isolation under BaFin supervision and KRITIS obligations.
800+ client banks · around 300 active users each
Read the reference → - EnginemanufacturerClient reference
First-fix rate up 40% across more than 5,000 users
Permissions computed per data node, multilingual and multimodal embeddings, and a deliberately smaller model reached through prompt engineering rather than scale.
+40% first-fix rate · −30% response time
Read the reference → - EnginemanufacturerClient reference
Downstream cost of defects down by up to 80%
Custom CNNs on Databricks running two detection modes at once — classified known defects and unsupervised unknown anomalies — with alerting at the line.
up to −80% downstream defect cost
Read the reference → - AutomotivesupplierClient reference
DIN and ISO answers for 300 to 400 engineers
A LangGraph multi-agent system on Azure AI Search with MLflow and pseudonymised query tracing, and an accuracy split published by query type.
100% on single-shot definitions · 70–80% multi-standard
Read the reference → - Regulated assetmanagerClient reference
Three model tracks under one upgrade cadence
Development, main and legacy tracks on half-yearly upgrade windows, a Terraform module library for application stacks, and node right-sizing that keeps bank APIs off GPUs.
3 parallel model tracks · twice-yearly upgrade windows
Read the reference →
By industry
The regulator changes, the data problem does not. Pick the sector closest to yours.
What every engagement commits to
Whatever the starting system, we begin at the data layer, build inside your tenant, and hand the system to your team.
- Where it runs
- Your tenant, your subscription
- Who operates it after
- Your team, or ours if you ask
- What you keep
- Code, pipelines and prompts
- Model choice
- Yours — the architecture stays LLM-independent
- How it starts
- Two-week Readiness Blueprint, €4,900 fixed
- Governing law
- German
Industrial manufacturing
Maintenance and quality questions answered from the systems that already hold the answer, and inspection that runs at the line rather than after shipment.
Energy technology
Product and support documentation spread across SharePoint and Confluence, made answerable behind Entra ID inside the manufacturer's own tenant.
Financial services
Self-hosted platforms under BaFin supervision, KRITIS obligations, MaRisk and DORA, with tenant isolation and cost attribution per institution.
Automotive supply
DIN and ISO retrieval for engineering, with pseudonymised query tracing and an accuracy split published by query type rather than one headline number.
How to read these
Described, not named. Measured, not estimated.
The approvals we hold were given to Pexon Consulting GmbH for its own site, and a customer who agreed to appear there has not thereby agreed to appear here. So each reference carries the category, the size and the regulatory context instead of a logo — which, for a buyer deciding whether an engagement resembles theirs, is the more useful half of the information anyway.
Every segment the published references cover.
Read out of the references themselves rather than written here, so this list can only ever say what they say. Named references are arranged case by case and need the customer’s agreement first.
- Energy technology
- Transformers
- Industrial manufacturing
- Financial services
- Banking IT
- Critical infrastructure
- Engines
- Drive systems
- Production quality
- Automotive supply
- Precision components
- Asset management
- Fund services
13 segments · 6 published references · names withheld, regulatory context on every page
Common questions
Why are your case studies anonymous?
Because the approvals we hold cover an anonymous description, not a logo. A case study naming a customer without written consent for this specific site is a legal problem, not a marketing win. Each one carries the industry, the company size and the regulatory context, which is what a comparable buyer actually matches against.
Are the numbers in these case studies measured or estimated?
Measured, and taken from the delivery record. Where no hard figure exists we say what changed qualitatively instead of inventing a percentage. If a case study shows a number, that number came out of the engagement rather than a marketing workshop.
Can I speak to one of these customers as a reference?
Sometimes, and it depends on the customer. Named reference calls are arranged case by case and need the customer's agreement first. Ask on an architecture fit call and we will tell you honestly which engagements have a customer willing to talk.
Is six references everything Pexon has delivered?
No. Six are written up and published here. The Pexon group has been delivering since 2019 and the delivery record is considerably larger, but a reference we cannot describe accurately and with permission does not go on this page. What you see is the part that cleared both bars, not the part that exists.
Not a sales call. An architecture call.
Thirty minutes with the architect who would actually run the engagement.