Bertrandt AG¶

Company Overview¶

Business Overview¶

Bertrandt AG was founded in 1974 as a one-man engineering office in Baden-Württemberg and is headquartered in Ehningen, near Stuttgart. It is an independent, international development service provider — a company that OEMs and suppliers hire to do engineering work they do not want to staff internally — active mainly in automotive, plus aerospace & defence, mechanical engineering, energy and medical technology. It employs roughly 12,000 people at over 50 sites. [Evidence] ¹⁵

Listing and ownership:

Listed on the Frankfurt Stock Exchange (Prime Standard, Automobile sector); ISIN DE0005232805; IPO on 1 October 1996. [Evidence] ²
Shareholders (company disclosure, as of 9 December 2024): Porsche AG 28.97%, Friedrich Boysen Holding 14.90%, Frank Ferchau 10.00%, boards 4.00%, free float 45.76% (as published; these figures sum to more than 100% — the free-float figure needs re-verification). The Porsche anchor stake is strategically notable — Bertrandt is partly owned by a customer. [Evidence — inconsistent, verify] ³
The Management Board page lists three members and no designated CEO: Dr. Andreas Fink (Technology, since April 2023), Michael Lücke (Sales), Markus Ruf (Finance). Verify governance before citing — see Appendix C. [Evidence] ⁴

Key financials (fiscal year runs 1 October – 30 September):

FY2024/25 (reported 18 December 2025): total revenues €978 m, down ~18% (prior year €1.187 bn); EBIT −€35.5 m (prior year −€98 m, which was burdened by special items); FY2024/25 adjusted EBIT ≈ −€2 m; free cash flow +€18 m; equity ratio 41.8%; employees 12,185 (down 1,843 in one year). [Evidence] ⁵
H1 FY2025/26 (Oct 2025 – Mar 2026): revenues €456.7 m, down 11.6%; EBIT −€13.1 m (improved year-on-year); employees 11,744. On 27 April 2026 the board cut the full-year forecast from moderate growth to a moderate decline. [Evidence] ⁶
The "Fit for Future" profit-optimization programme (launched FY2023/24) has delivered annualized savings above €120 m — portfolio, structural and infrastructure cuts, i.e. site closures and job reductions. [Evidence] ⁶

Context, stated plainly: the German ESP sector is being squeezed — OEM R&D cost-cutting, project postponements since spring 2024, and work migrating to cheaper international R&D hubs. Bertrandt names all three and is responding with diversification (aerospace & defence growing double-digit; Volvo Cars nominated it a strategic supplier; ZEISS SMT granted "Strategic Partner" status). (Volvo named several service providers strategic suppliers in 2025 — see also the HCLTech and Tata Technologies reports.) [Evidence] ⁵⁶

Automotive Business¶

Automotive is Bertrandt's main customer sector [Evidence] ⁵. Unlike a platform vendor, Bertrandt does not sell software — it sells engineering hours and work packages across the whole vehicle: body/exterior, interior (a ~40-year practice covering trim, cockpits, seats, lighting and dimensional management), simulation, testing, electronics and software. [Evidence] ¹²¹¹

Engineering Business¶

Service lines relevant here: Design Solutions (body, exterior, interior), Simulation Solutions (crash, durability, CFD/thermal, multibody), Testing Solutions, Engineering Integration & Management (incl. quality management/FMEA), and Vehicle & Prototype Services — which contains the model making, rapid technologies and injection- moulding tool shop. [Evidence] ¹¹¹³¹⁰

Role in Automotive Moulded Parts Workflow¶

Workflow Stage	Relevant?	Notes
Requirements	Yes	RFQ/requirements processing is where Bernd (AI) actually lives⁷
Industrial Design	Yes	Design Solutions studios
Concept Design	Yes	Interior/exterior concept work (e.g. autonomous-vehicle interiors)¹²
CAD Modelling	Core strength	Programme work on OEM systems; tool shop designs in CATIA V5, Siemens NX¹⁰
Engineering Review	Yes	Quality management; FMEA Agent (AI) assists FMEA⁸
Simulation	Yes — but not moulding	Crash, durability, CFD, MBS; no injection simulation named¹³
DFM	Implicit, manual	Human know-how in interior/tooling teams; no automated DFM offering found
Tool Design	Yes — physical	In-house tool making, CAD/CAM (Tebis)¹⁰
Mould Flow	Not evidenced	No Moldflow/Cadmould/Moldex3D mention anywhere public
Prototype	Core strength	Rapid tooling (aluminium moulds), parts in series materials, MuCell¹⁰
Validation	Yes	Testing Solutions
Manufacturing Engineering	Partial	Light-MES, evoVIU camera, evoTrQ press-shop traceability⁹
Production Release	Via customer	ESPs hand over to the OEM/Tier-1

What makes Bertrandt structurally interesting: it is the only company in this study's ESP scan that both designs moulded interior parts and physically moulds them in-house — yet its AI enters the workflow only at the two paperwork ends (RFQs in, FMEA documents out). [Synthesis]

AI Strategy¶

Public AI Vision¶

Bertrandt presents AI as a practical efficiency layer for engineering services, under the slogan "AI amplified, human verified" — automation with human decision authority, traceable results, and security/data-protection compliance (it serves regulated, safety-critical industries). The foundation is the "Bertrandt AI Stack", described as a standardized, scalable platform — "quick to deploy, stable in operation, controllable costs, flexible expansion" — supporting everything "from assistant systems to complex multi-agent solutions in engineering processes." [Evidence + Marketing] ⁹

This is a strikingly different pitch from the platform vendors': no world models, no sovereignty narrative, no product SKUs — just billable-hours leverage. For an ESP whose product is engineer time, AI that speeds up quoting and documentation goes straight to margin. [Synthesis]

Investor Statements¶

No AI-specific investor statements were found. The FY2024/25 results release and the H1 2025/26 report are about survival economics — "Fit for Future," diversification, internationalization — and do not mention AI at all. That silence is itself a finding: AI is an operational tool at Bertrandt, not (yet) an investor story. [Evidence of absence] ⁵⁶

Engineering AI Strategy¶

What the public material actually shows, in three streams: [Synthesis]

Document/process AI (the centre of gravity): Bernd for RFQ analysis and routing; the FMEA Agent for quality documentation. LLM-based, human-verified, integrated "into the existing Microsoft environment." [Evidence] ⁷⁸
Simulation-adjacent AI (edges only): AI-supported material-card creation for FEM, AI-based motion tracking from test video, and "AI-supported development and process automation" in the methods team (tools named: Synera, Python, Copilot Studio, Windsurf — plus classical topology optimization, on which see the honest-AI note in §5). [Evidence] ¹³
Shop-floor digital products: Light-MES (lightweight manufacturing execution system), evoVIU (smart camera), evoTrQ (component traceability in the press shop). Manufacturing-facing, but stamped-metal/press-shop flavoured — not the injection-moulding shop. [Evidence] ⁹

Timeline of AI Evolution¶

Date	Event
FY2023/24	"Fit for Future" programme launched (context for the efficiency push)⁶
Feb 9, 2026	Bernd AI assistant presented — RFQ processing, >90% classification accuracy⁷
Apr 8, 2026	FMEA Agent presented — LLM + QM knowledge base⁸
2026 (undated page)	"Artificial Intelligence at Bertrandt" hub: AI Stack, agents, Light-MES, evoVIU, evoTrQ⁹

A short timeline — Bertrandt's public AI history effectively begins in 2026. [Synthesis]

Products Relevant to Engineering¶

Bernd (AI assistant). Processes incoming Requests for Quotation and engineering documents: extracts data, routes work to the right specialist departments, checks against standards. Claimed >90% accuracy in classification tasks, which the company says compares favourably to classification by non-specialist staff. Runs "exclusively within Bertrandt's secure infrastructure"; API-compatible with Jira and existing toolchains. Roadmap: auto-generating quotation documents, task packages and reference-project suggestions. Moulded-parts angle: none stated. [Evidence] ⁷

FMEA Agent. Supports creation, refinement and validation of Failure Mode and Effects Analyses: automates structural elements, flags inconsistencies, suggests standards-compliant wording, and proposes causes/effects/preventive actions "based on similar projects" from a curated knowledge base of internal expertise, project documents and external sources. Combines "state-of-the-art large language models" with QM methodology; integrated into the Microsoft environment; benefits stated qualitatively only ("significantly reduced" research, "shorter" review cycles). Moulded-parts angle: none stated — though FMEA for a moulded part is a real use case a customer could bring. [Evidence] ⁸

Injection-moulding tool shop (physical). In-house tool making with CAD/CAM design (CATIA V5, Siemens NX; Tebis for NC programming), milling centres, injection moulding with clamping force up to 750 t, MuCell microcellular-foam capability, aluminium rapid-tooling moulds, and parts in "all common series-production materials." AI angle: none stated. [Evidence] ¹⁰

Simulation Solutions. FE (LS-DYNA, Abaqus, Nastran, OptiStruct, Pam-Crash), multibody (ADAMS, Simpack), flow/thermal (STAR-CCM+, ANSA, PowerFLOW, TAITherm, Ansys AEDT, OpenFOAM). No injection-moulding simulation tool appears in the list. [Evidence] ¹³

AI Capabilities¶

1. Bernd — RFQ intake and routing. Stage: requirements/quoting. Inputs: RFQ and engineering documents. Outputs: extracted data, department routing, standards checks. Tech: LLM-based, on internal infrastructure. Announced: Feb 9, 2026. Limitation: the one hard number (>90%) is for classification, not for engineering judgement; the generative features (auto-quotes) are roadmap, not shipped. [Evidence] ⁷

2. FMEA Agent — quality documentation. Stage: engineering review/quality. Inputs: project documents, FMEA structures, knowledge base. Outputs: draft failure modes, causes, actions; consistency checks. Tech: LLM + curated knowledge base; Microsoft environment. Announced: Apr 8, 2026. Limitation: no quantified results; explicitly human-in-the-loop. This is honest AI by this study's test — it learns from data (an LLM plus retrieved project knowledge) — but the LLM itself is almost certainly rented, not owned (see §8). [Evidence] ⁸

3. AI-supported material-card creation. AI-assisted evaluation to build material cards (the material-property datasets FEM solvers need). Limitation: one line on a service page; no method, no dates, no numbers. [Evidence — thin] ¹³

4. AI-based motion tracking. Reconstructs movements in physical tests from video data. Genuine computer-vision ML, but for testing, not moulding. [Evidence] ¹³

5. "AI-supported development and process automation" (methods team). Named tools — Synera (low-code engineering automation), Python, Microsoft Copilot Studio, Windsurf — alongside topology optimization. The automation is real; note that topology optimization itself is classical physics-driven optimization, not AI — see the concept note What is Optimization technology. Where the "AI-supported" label leans on the optimizer rather than on the LLM tooling, treat it as [Evidence + Marketing]. [Evidence + Marketing] ¹³¹⁵

6. Shop-floor products (Light-MES, evoVIU, evoTrQ). Production transparency, smart-camera inspection, and component traceability in the press shop (sheet metal). The closest Bertrandt AI gets to a factory — and it is the wrong factory for this study: no injection-moulding deployment is described. [Evidence] ⁹

What is absent — first-class negative findings: no AI for mould-flow analysis or interpretation; no AI/automated DFM checking; no geometry-based retrieval of past designs; no learning from moulding-trial or production outcomes; no moulding-simulation software named at all. [Evidence of absence] ¹³¹⁰

Engineering Workflow Contribution¶

As an ESP, Bertrandt executes chunks of an OEM's programme on the OEM's chosen platforms. Publicly verified tooling of its own practice: CATIA V5 and Siemens NX for tool-shop design, Tebis for CAM, the CAE stack in §4, and Microsoft as the office/AI environment. [Evidence] ¹⁰¹³⁸

For a moulded interior part, Bertrandt can genuinely run:

RFQ intake (Bernd — AI) → interior/trim design (CATIA/NX — human) → crash, durability, thermal simulation (human) → FMEA (FMEA Agent — AI-assisted) → tool design and build (in-house shop — human) → prototype mouldings up to 750 t, incl. MuCell (human) → testing (human, with AI motion tracking at the edges).

The two AI-assisted boxes bracket the chain; everything geometric, mouldable and physical in between is untouched by the AI Stack, on public evidence. [Synthesis]

Public Customer Evidence¶

Case Studies¶

No named-customer AI case study exists. Bernd and the FMEA Agent are described on Bertrandt's own blog with internal use cases and one anonymous specialist testimonial ("saves time and measurably increases quality" — FMEA Agent). [Evidence — vendor-only] ⁷⁸
No public case study combining AI + moulded parts + a measured outcome was found — the same gap as every platform vendor in this study. [Evidence of absence]
Business-level customer evidence exists but is not AI: Volvo Cars nominated Bertrandt a strategic supplier; ZEISS SMT granted "Strategic Partner" status (both cited in the FY2024/25 release, no engineering detail). [Evidence] ⁵
Interior practice references — Level ⅘ autonomous-vehicle interior concepts, the "Carbon Carrier" lightweight structural concept, curved-display cockpits — are concept showcases, not quantified programmes. [Evidence + Marketing] ¹²

Measurable improvements¶

The only quantified AI claim anywhere: ">90%" classification accuracy for Bernd, plus the headline "seconds instead of hours" for document processing — the latter is a slogan, not a measurement. [Evidence + Marketing] ⁷

Technical Architecture (Inferred)¶

Evidence¶

Bernd runs "exclusively within Bertrandt's secure infrastructure," is LLM-based, and exposes APIs toward Jira and existing toolchains. [Evidence] ⁷
The FMEA Agent combines "state-of-the-art large language models" with a curated knowledge base and integrates "into the existing Microsoft environment." [Evidence] ⁸
The AI Stack is described as one shared platform under all applications, built for traceability and data protection. [Evidence] ⁹

Synthesis¶

The pattern is: a platform layer (AI Stack) hosting retrieval-augmented LLM agents (Bernd, FMEA Agent) over Bertrandt's internal documents, inside a Microsoft-centric IT estate, with humans approving outputs. [Synthesis]

Inference¶

"State-of-the-art LLMs" + "existing Microsoft environment" + no claim of in-house model training strongly suggests rented frontier models via Microsoft Azure (Azure OpenAI or similar) — consistent with this study's cross-vendor finding that every vendor rents its LLM. No source names the model supplier; treat as inference (Appendix C). [Inference]

AI Technologies¶

In one list: LLMs (rented, supplier undisclosed); retrieval over curated internal knowledge (FMEA Agent); agentic patterns ("multi-agent solutions" on the AI Stack — claimed, not demonstrated); computer vision (evoVIU camera, motion tracking); process automation (Synera, Copilot Studio, Python); classical topology optimization wearing an "AI-supported" banner¹⁵; MES/traceability telemetry (Light-MES, evoTrQ). Absent: geometry ML, simulation surrogates, moulding-process ML, outcome-learning of any kind. [Evidence + Synthesis] ⁹¹³

Research Publications¶

No peer-reviewed papers, patents on AI methods, or standards contributions by Bertrandt were found in this research pass. As a services firm, Bertrandt communicates via blog posts and trade events, not publications. (Its interior practice claims a seat-concept patent — mechanical, not AI.) [Evidence of absence — limited search depth; see Appendix C] ¹²

Open Source¶

Nothing found: no AI-related GitHub organisation, no Hugging Face presence, no released models or datasets. One blog post (Feb 23, 2026) advocates OpenEMS, an open-source energy-management system — energy domain, not engineering AI. The AI Stack is a proprietary internal asset. [Evidence of absence] ¹⁴

Engineering Service / Platform Mapping¶

Verified platform usage: CATIA V5 and Siemens NX (tool-shop design); Tebis (CAM); LS-DYNA / Abaqus / Nastran / OptiStruct / Pam-Crash (FE); ADAMS / Simpack (MBS); STAR-CCM+ / ANSA / PowerFLOW / TAITherm / Ansys AEDT / OpenFOAM (CFD/thermal); Synera / Copilot Studio / Python / Windsurf (automation); Jira and a Microsoft estate (process/IT). As an ESP it will additionally work inside whatever CAD/PLM its OEM customers mandate. No moulding-simulation platform named. [Evidence] ¹⁰¹³⁷⁸

Engineering Intelligence Stack Mapping¶

Intent — Bernd captures customer intent at the RFQ stage: the most concrete intent-layer AI of any company in this study so far, because quoting is an ESP's front door. [Evidence] ⁷
Knowledge — the FMEA Agent's curated knowledge base (internal expertise + project documents) is a real, if document-bound, knowledge layer. Nothing indexes geometry or tooling history. [Evidence] ⁸
Reasoning — none beyond LLM text reasoning; no simulation surrogates, no moulding prediction. [Evidence of absence] ¹³
Execution — human engineers and a physical tool shop; AI executes nothing in CAD or the shop. [Evidence of absence] ¹⁰
Feedback — Light-MES/evoTrQ collect production data in press-shop contexts, but no loop feeds moulding or design outcomes back into any model — the same missing feedback layer as every vendor in this study. [Evidence of absence] ⁹

Strengths¶

Both worlds in one company: moulded-parts design (interior trim, ~40 years) and physical moulding/tooling (750 t, MuCell) — unique among vetted ESPs. [Evidence] ¹²¹⁰
Real, shipped, dated AI with a hard number (Bernd, >90% classification) — rare honesty-passing AI in this landscape. [Evidence] ⁷
Pragmatic AI economics: agents target quoting and documentation, where an ESP's margin actually lives. [Synthesis]
Porsche anchor shareholding and broad OEM programme access. [Evidence] ³

Weaknesses¶

The AI and the moulding shop don't touch — no public workflow connects them. [Evidence of absence] ¹⁰⁹
Financial distress: two heavily loss-making years, −18% then −11.6% revenue, ~2,300 jobs gone in 18 months, forecast cut April 2026. Restructuring consumes management attention and capital. [Evidence] ⁵⁶
AI is document-bound: no geometry, simulation or process ML at all. [Evidence of absence] ¹³
LLM capability is rented and the supplier undisclosed; benefits beyond Bernd's one number are unquantified. [Inference] ⁸

Current Gaps (largely manual today)¶

Everything between the RFQ and the FMEA document: interior-trim DFM judgement (draft, wall thickness, sink risk), mould-flow analysis (not even classical simulation is publicly evidenced, let alone AI), tool-design decisions, moulding trials and process set-up (MuCell included), and any reuse of the tool shop's decades of tooling history — which today sits in filing systems and veterans' heads, not in any retrieval system. [Synthesis] ¹⁰¹³

Future Direction¶

Publicly stated: expand the AI Stack "from assistant systems to complex multi-agent solutions"; extend Bernd toward auto-generated quotations and reference-project suggestions; keep diversifying into aerospace & defence and medical while the automotive core recovers. [Evidence] ⁹⁷⁵

My read: the direction of travel is more document agents on more internal processes — the fastest payback during restructuring. Nothing signals investment in geometry or moulding AI; if that connection ever happens, it will likely arrive via a customer demand or an external tool, not internal R&D. [Inference]

Relevance to Automotive Moulded Parts¶

Capability	Strength	Notes
Plastic Part Design	Strong — human	Interior/trim practice (~40 yrs); no AI assist
Surface Design	Present	Interior/exterior design studios; tools not publicly itemized
CAD Automation	Emerging	Synera/Copilot Studio process automation; not part-design automation
DFM	Manual only	No automated or AI DFM offering found
Tool Design	Strong — physical	In-house tool making, CATIA V5/NX/Tebis
Mould Flow	Not evidenced	No simulation tool named; a genuine hole in the public story
Manufacturing Engineering	Partial	750 t moulding, MuCell, rapid tooling; Light-MES elsewhere
Quality	AI-assisted documents	FMEA Agent (generic, not moulding-specific)
Engineering Knowledge Reuse	Document-level only	FMEA knowledge base; nothing geometric

Critical finding. Bertrandt holds, inside one company, exactly the two assets this study's thesis says should be connected — an AI agent platform and a moulding/tooling operation with decades of history — and, on all public evidence, has not connected them. The RFQs Bernd reads and the FMEAs the agent drafts pass through the same programmes whose trim parts the tool shop moulds, yet no AI checks a draft angle, predicts a sink mark, or retrieves a similar past tool. If the connection existed and worked, it would be a marketable differentiator for a struggling ESP — its absence from marketing is therefore reasonably strong evidence of absence in fact. [Synthesis — the report's spine] ⁷⁸¹⁰

Key Takeaways¶

Bertrandt is the only vetted ESP with both physical moulding/tooling capability and dated, named AI. [Evidence] ¹⁰⁷
The tool shop is real: in-house tool making, injection moulding up to 750 t, MuCell, aluminium rapid tooling, CATIA V5/NX/Tebis. [Evidence] ¹⁰
The AI is real but document-bound: Bernd (RFQs, >90% classification, Feb 2026) and the FMEA Agent (Apr 2026) on the Bertrandt AI Stack, "AI amplified, human verified." [Evidence] ⁷⁸⁹
The two are publicly unconnected — no AI touches DFM, tooling, mould flow or moulding process anywhere at Bertrandt. The company is the cleanest single-company illustration of the gap this study keeps finding. [Evidence of absence]
No moulding-simulation software is named on Bertrandt's public site at all — its simulation portfolio is crash/durability/CFD/MBS. [Evidence of absence] ¹³
The business is under severe pressure: FY2024/25 revenue −18% to €978 m, EBIT −€35.5 m, ~2,300 jobs cut since Sept 2024, forecast cut again April 2026. [Evidence] ⁵⁶
Its LLM layer is rented, supplier undisclosed (Microsoft environment — inference), matching the study's every-vendor-rents-its-LLM finding. [Inference] ⁸
The only quantified AI benefit anywhere is Bernd's >90% classification accuracy; everything else is qualitative. [Evidence] ⁷
For anyone pursuing moulding AI, Bertrandt reads as customer and channel first, competitor a distant third — and as the ideal pilot site to close the AI-to-mould gap inside one building. [Inference]

References¶

Primary Sources — Bertrandt¶

"Seconds instead of hours – thanks, Bernd!" blog, Feb 9 2026 — https://www.bertrandt.com/en/blog/bernd
FMEA Agent blog, Apr 8 2026 — https://www.bertrandt.com/en/blog/fmea-agent
Artificial Intelligence at Bertrandt (AI Stack hub) — https://www.bertrandt.com/en/artificial-intelligence-at-bertrandt
Modelling & Rapid Technologies (tool making, 750 t, MuCell, CATIA V5/NX/Tebis) — https://www.bertrandt.com/en/vehicle-and-prototype-services/modelling-and-rapid-technologies/
Design Solutions — https://www.bertrandt.com/en/design-solutions
Interior development — https://www.bertrandt.com/en/design-solutions/interior
Simulation Solutions (tool list; no moulding simulation) — https://www.bertrandt.com/en/simulation-solutions
FY2024/25 press release, Dec 18 2025 (PDF) — https://www.bertrandt.com/fileadmin/files/files/00_Unternehmen/03_Presse/Pressemitteilungen/2025/2025-12-18_press_release_fy_2024-2025_bertrandt.pdf
H1 FY2025/26 report, Oct 2025–Mar 2026 (PDF) — https://www.bertrandt.com/fileadmin/files/files/00_Unternehmen/01_Investor_Relations/03_Finanzberichte/GJ_25_26/half-year-report_q2_2025-2026_bertrandt.pdf
Financial reports index — https://www.bertrandt.com/en/investor-relations/financial-reports
Company overview — https://www.bertrandt.com/en/company
Management Board — https://www.bertrandt.com/en/company/bertrandt-group/management-board
Bertrandt share (listing, ISIN, IPO) — https://www.bertrandt.com/en/company/investor-relations/bertrandt-share
Shareholder structure (as of Dec 9 2024) — https://www.bertrandt.com/en/company/investor-relations/bertrandt-share/shareholder-structure/
Blog archive, Feb 2026 — https://www.bertrandt.com/en/blog/archiv/archive/2026/february/page-1

Secondary Sources¶

Search-engine snippets (DuckDuckGo) used only for discovery; every cited fact above was verified on the linked primary page or PDF.

Appendix A — Timeline¶

1974 founded (one-man office) → Oct 1, 1996 IPO Frankfurt → Apr 2023 Dr. Andreas Fink joins board (Technology) → FY2023/24 "Fit for Future" launched; EBIT −€98 m → Dec 9, 2024 shareholder snapshot (Porsche 28.97%) → Dec 18, 2025 FY2024/25 results (−18%, EBIT −€35.5 m, −1,843 jobs) → Feb 9, 2026 Bernd AI assistant → Feb 12, 2026 Q1 report → Apr 8, 2026 FMEA Agent → Apr 27, 2026 ad-hoc forecast cut → May/Jun 2026 H1 report (−11.6%, 11,744 employees).

Appendix B — Glossary¶

ESP — engineering service provider: sells outsourced development work (engineer hours and work packages) to OEMs and suppliers.
RFQ — request for quotation: the customer document an ESP must analyse and price to win work; Bernd's home turf.
FMEA — failure mode and effects analysis: a structured quality method listing what could go wrong in a design/process, its effects, and preventive actions.
MuCell — licensed microcellular foam injection moulding (supercritical gas dissolved in the melt creates micro-bubbles): lighter parts, less sink and warp.
Clamping force (750 t) — the force holding the mould shut during injection; 750 tonnes accommodates fairly large automotive parts (e.g. door trim carriers).
Rapid tooling — prototype moulds cut from aluminium: cheaper and faster than steel, good for parts in real series material before series tools exist.
Trim — the visible moulded covering parts of a vehicle interior.
MES — manufacturing execution system: shop-floor software tracking production in real time (Bertrandt's "Light-MES").
"Fit for Future" — Bertrandt's profit-optimization (restructuring) programme, running since FY2023/24.
Bertrandt AI Stack — the company's internal platform on which its AI agents (Bernd, FMEA Agent) are built.

Appendix C — Notes (thinnest-evidence areas to revisit)¶

Whether any moulding simulation exists internally. Absence from the public site is not proof the tool shop never buys a Moldflow/Cadmould analysis; job ads and tenders were not exhaustively searched. The report claims only "not publicly evidenced."
The LLM supplier behind Bernd/FMEA Agent — "Microsoft environment" suggests Azure-hosted models, but no source names one. Stated as inference only.
Bernd's ">90%" — attaches to classification in the RFQ workflow; an earlier research note's phrasing "accuracy on standards checks" is looser than the source and could not be re-verified. Standards checking is a listed function, but the number is for classification.
Governance — the board page lists three members with no designated CEO/chairman; the recent CEO history behind this was not researched. Verify before citing leadership.
FY2024/25 press-release table carries column headers reading "until 30/06/2025" — an apparent typo; the release text states these are fiscal-year figures (year ended Sep 30, 2025), and the H1 report cross-confirms €977.9 m as the prior full-year figure.
Interior practice claims ("~40 years," seat patent, Carbon Carrier) are from the company's own service pages — no independent verification.
Volvo Cars / ZEISS SMT statuses come from one press release with no detail on scope or value.
Shareholder structure is dated Dec 9, 2024 — Porsche's stake may have moved since; check current voting-rights notifications before reuse.

Executive Summary¶

Who they are. Bertrandt is one of Germany's large automotive engineering service providers (ESPs): founded 1974, listed on the Frankfurt Stock Exchange, about 11,700–12,200 employees at 50+ sites, doing outsourced development work for car makers and suppliers. Porsche AG holds 28.97% of the shares. [Evidence] ¹²³
Why they matter for this study. Bertrandt is the only ESP found in vetting that has both sides of the story under one roof: a physical injection-moulding tool shop (in-house tool making, moulding machines up to 750 t clamping force, MuCell microcellular foaming) and a dated, named AI programme — the "Bertrandt AI Stack" with the Bernd assistant (Feb 2026) and an FMEA Agent (Apr 2026). [Evidence] ¹⁰⁷⁸
The central honest finding. After deliberate searching, no public evidence connects the two. Bernd reads RFQ documents; the FMEA Agent drafts risk analyses; the tool shop cuts steel and aluminium with CATIA V5, Siemens NX and Tebis. No AI capability touches moulding, tooling or DFM, and no moulding-simulation software (Moldflow, Cadmould, Moldex3D) is named anywhere on Bertrandt's public site. The AI stack and the moulding shop are, publicly, separate worlds inside one company — making Bertrandt the cleanest case study of the industry-wide gap this study keeps finding. [Evidence of absence] ¹³¹⁰
Business health matters here. Bertrandt is in a hard restructuring: FY2024/25 revenue fell 18% to €978 m, EBIT was −€35.5 m, headcount dropped by 1,843; the first half of FY2025/26 fell a further 11.6%, and the forecast was cut again in April 2026. Its AI is partly a survival tool (winning and processing quotes faster), not a research luxury. [Evidence] ⁵⁶

Bertrandt company overview (founded 1974; Ehningen; ~12,000 employees; 50+ sites), accessed Aug 1 2026 — https://www.bertrandt.com/en/company ↩↩
Bertrandt share page (Frankfurt Prime Standard; ISIN DE0005232805; IPO Oct 1 1996), accessed Aug 1 2026 — https://www.bertrandt.com/en/company/investor-relations/bertrandt-share ↩↩
Shareholder structure as of Dec 9 2024 (Porsche AG 28.97%; Boysen 14.90%; Ferchau 10.00%; free float 45.76%) — https://www.bertrandt.com/en/company/investor-relations/bertrandt-share/shareholder-structure/ ↩↩↩
Management Board page (Fink — Technology, since Apr 1 2023; Lücke — Sales; Ruf — Finance; no CEO listed), accessed Aug 1 2026 — https://www.bertrandt.com/en/company/bertrandt-group/management-board ↩
Bertrandt FY2024/25 press release, Dec 18 2025 (revenue €978 m, −18%; EBIT −€35.5 m; adjusted EBIT ≈ −€2 m; 12,185 employees; "Fit for Future"; Volvo Cars; ZEISS SMT) — https://www.bertrandt.com/fileadmin/files/files/00_Unternehmen/03_Presse/Pressemitteilungen/2025/2025-12-18_press_release_fy_2024-2025_bertrandt.pdf ↩↩↩↩↩↩↩↩↩↩
Bertrandt report on the 1^st half FY2025/26, Oct 1 2025–Mar 31 2026 (revenue €456.7 m, −11.6%; EBIT −€13.1 m; 11,744 employees; Apr 27 2026 forecast cut; Fit for Future savings >€120 m annualized) — https://www.bertrandt.com/fileadmin/files/files/00_Unternehmen/01_Investor_Relations/03_Finanzberichte/GJ_25_26/half-year-report_q2_2025-2026_bertrandt.pdf ↩↩↩↩↩↩↩↩
"Seconds instead of hours – thanks, Bernd!", Bertrandt blog, Feb 9 2026 (RFQ analysis, routing, standards checks; >90% classification accuracy; secure internal infrastructure; Jira API) — https://www.bertrandt.com/en/blog/bernd ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩
"Rethinking FMEA" (FMEA Agent), Bertrandt blog, Apr 8 2026 (LLM + curated QM knowledge base; Microsoft environment; qualitative benefits only) — https://www.bertrandt.com/en/blog/fmea-agent ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩
"Artificial Intelligence at Bertrandt" hub ("Bertrandt AI Stack"; "AI amplified, human verified"; Bernd; FMEA agent; Light-MES; evoVIU; evoTrQ press-shop traceability), accessed Aug 1 2026 — https://www.bertrandt.com/en/artificial-intelligence-at-bertrandt ↩↩↩↩↩↩↩↩↩↩↩
Modelling & Rapid Technologies page (tool building for plastic injection moulding; clamping force up to 750 t; MuCell; aluminium moulds; series materials; CAD: CATIA V5, Siemens NX; CAM: Tebis), accessed Aug 1 2026 — https://www.bertrandt.com/en/vehicle-and-prototype-services/modelling-and-rapid-technologies/ ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩
Design Solutions overview (body/exterior and interior structure development), accessed Aug 1 2026 — https://www.bertrandt.com/en/design-solutions ↩↩
Interior development page (~40 years interior equipment experience; trim, cockpit, seats incl. patented Ergo seat concept, lighting, dimensional management; Carbon Carrier; L4/L5 interiors), accessed Aug 1 2026 — https://www.bertrandt.com/en/design-solutions/interior ↩↩↩↩↩
Simulation Solutions page (FE: LS-DYNA, Abaqus, Nastran, OptiStruct, Pam-Crash; MBS: ADAMS, Simpack; CFD/thermal: STAR-CCM+, ANSA, PowerFLOW, TAITherm, Ansys AEDT, OpenFOAM; methods: Synera, Python, Copilot Studio, Windsurf, topology optimization; AI motion tracking; AI material cards; no injection-moulding simulation listed), accessed Aug 1 2026 — https://www.bertrandt.com/en/simulation-solutions ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩
Bertrandt blog archive, Feb 2026 (OpenEMS open-source post, Feb 23 2026; Bernd the only AI post that month) — https://www.bertrandt.com/en/blog/archiv/archive/2026/february/page-1 ↩
On why topology optimization and DoE are classical optimization rather than AI, see the concept note What is Optimization technology: these methods compute from physics equations and learn nothing from data. ↩↩