Skip to main content

# AUTOMOTIVE · DATA & AI · DACH

Data & AI for the plant floor, not the pitch deck.

OEMs and Tier-1/2 suppliers in DACH sit on more sensor and quality data than almost any other sector — and trust it least. We turn production data into quality decisions people act on, with EU AI Act classification built into delivery.

 book-call→ read the full case study

# WHERE IT BREAKS

Where automotive data breaks down.

  1. False-positive overload in quality

    The quality system flags everything — real defects, sensor noise, calibration drift. Operators stop trusting it, and real defects slip through at the same rate as before the system existed.

    ▸ how we solve it

    We tune classification models for precision over recall and ship the predictions into the Power BI layer your quality team already uses — no new tool. A data audit up front usually surfaces the hidden calibration and maintenance logs that explain most of the noise.

  2. Sensor & production data in silos

    MES, quality checkpoints, line sensors and ERP each hold a fragment. Nobody can answer "which line, which shift, which supplier batch?" without a week of manual reconciliation.

    ▸ how we solve it

    We build an EU-hosted data platform that unifies sensor, MES and quality data, with data-quality gates at ingestion so a bad load is caught before it reaches a dashboard — not by an operator three weeks later.

  3. Production AI with no AI Act classification

    AI that scores or gates a safety-relevant part can fall under the AI Act as an embedded high-risk system (Annex I) — and most plants have no risk classification or documentation on file.

    ▸ how we solve it

    We settle the risk tier and the required documentation in week one, not after go-live, and design the human-oversight controls the regulation expects. Compliance is part of delivery, not a bolt-on.

# PROOF

Predictive quality for an automotive OEM.

A leading automotive OEM was drowning in false-positive defect alerts. We shipped precision-tuned predictions into their existing Power BI layer — no new tool to learn — and caught data-quality issues at ingestion instead of by end users.

40%

Fewer false positives

2.5×

Faster triage

90%+

Dashboard adoption (month 1)

12 wk

End to end

→ read the full case study

# REGULATORY

The regulatory angle: AI Act Annex I, August 2028.

Under the EU AI Act, AI that acts as a safety component of a regulated product — machinery, vehicles, their embedded systems — is high-risk under Annex I. The Digital Omnibus moved that deadline: Annex I embedded high-risk obligations now apply from 2 August 2028 (Annex III stand-alone systems from 2 December 2027). That is runway, not a reprieve — risk classification, technical documentation and human-oversight design take months to build into a production system, so the inventory-and-classify step belongs in this year's roadmap, not 2028's.

// from the journal

Relevant reading.

# FAQ

Common questions.

  • Does the EU AI Act apply to our production-line AI?
    If the AI is a safety component of a regulated product — for example a model that gates or classifies a safety-relevant part — it can be high-risk under Annex I of the AI Act, with obligations applying from 2 August 2028 after the Digital Omnibus re-phasing. The way to know is a risk classification, which is week-one work in our engagements, not a post-launch scramble.
  • Do we need to replace our quality system to add predictive analytics?
    Usually not. In our automotive work we shipped predictions into the Power BI layer the quality team already used — no new tool, no new login. The hard part is not the model; it is calibrating against false positives so the team trusts the alerts, and running in parallel with the old system until the numbers agree.
  • Our sensor and MES data is a mess — where do you start?
    A data audit: mapping every sensor feed, quality checkpoint and historical defect record, plus the maintenance and calibration logs everyone forgets. That audit is where the hidden causes surface — in one project a buried calibration log explained a 3× alert rate on one line.

Have a plant-floor data problem?

A 30-min call with the two people who would run it. We will tell you within 48 hours whether we are the right fit.

 book-call

// or write: hello@saloid.com · gräfelfing · de