SHEET 06 / 06AUTOMOTIVE AI VALUE CASE · MODELLED

~€520M a year, built bottom-up.

The value at stake from AI across the vehicle development lifecycle, for a premium-OEM archetype — ~2M units/yr, ~€150B revenue, ~10K-engineer software organisation. Every number is cost pool × evidenced uplift, net of an explicit verification tax, on the conservative end of published ranges. The method follows DORA's 2026 ROI of AI-Assisted Software Development framework. Hype numbers don't survive CFO scrutiny; these are built to.

~€520M

Annual value at stake · base case

~7:1

Value-to-cost ratio (€520M value vs €75M annual cost)

<9 mo

Payback period, including adoption ramp

3 of 12+

Flagship use cases quantified in this model

WHAT THE BASE CASE EXCLUDES — DELIBERATELY The largest optionality sits outside the base case: recall avoidance (a single major recall is €100M+ direct, plus brand and regulatory exposure) and time-to-market revenue pull-forward. Both appear only in the upside scenario. The honesty is the point.

exhibit 01 --layer="frontier copilot + on-prem domain SLM"

Engineering productivity — €158M / yr.

EXHIBIT 01 / SHEET 06

Convert ~18% of an €880M engineering cost pool into delivered software throughput and faster time-to-SOP. The pool: 8,000 software & V&V engineers × ~€110K fully-loaded.

TABLE 01 — KPI MOVEMENTS · ENGINEERING PRODUCTIVITY
KPI	Baseline	Target	Impact
Individual effectiveness (DORA)	index 100	115+	▲ strongest effect
Lead time for changes (DORA key)	index 100	75	▼ 25%
Deployment frequency (DORA key)	50 / yr	56 / yr	▲ 12%
MISRA-C first-pass compliance	40%	94%	▲ 54 pts
Defect escape / rework rate	index 100	75	▼ 25%
Software feature time-to-SOP	index 100	85	▼ 15%
Delivery instability (change-fail rate)	5%	≤ 5%	⚠ guardrail · hold flat

€880M cost pool × 18% net uplift (gross ~45% phase-weighted × ~60% task adoption − verification tax) = €158M / yr · cost €25M · net €133M · return 6.3×

Net uplift is built as headcount-reinvestment capacity — freed time redirected into the backlog, not headcount cut. Evidence anchors: 2025 DORA report (>80% report AI raised productivity; individual effectiveness the strongest effect) E, GitHub Copilot controlled study (55.8% faster) E, PopcornSAR PARVIS (MISRA 40→94%, 3–4× test-effort cut) V, Simulink mutant generation 13× faster (arXiv:2602.04066) E.

exhibit 02 --layer="cloud ML + agentic root-cause"

Quality & warranty — €240M / yr.

EXHIBIT 02 / SHEET 06

Mine connected-vehicle telematics, DTCs and service data to detect emerging failure patterns weeks-to-months earlier — and shrink the affected population before it becomes a warranty wave or a recall. The pool: €3.0B/yr of warranty + goodwill accrual (~2% of revenue).

TABLE 02 — KPI MOVEMENTS · QUALITY & WARRANTY
KPI	Baseline	Target	Impact
Mean time-to-detect a field issue	months	weeks	▼ 60%
Warranty cost / vehicle	€1,500	€1,380	▼ €120
Recall scope before containment	index 100	60	▼ 40%
Repeat-repair rate	index 100	80	▼ 20%

€3.0B warranty pool × 8% net reduction (low end of the 8–15% evidenced range) = €240M / yr · cross-check: −€120/vehicle × 2M vehicles = €240M · cost €30M · return 8.0×

Telematics + DTC pattern mining is mature in production E; agentic root-cause and supplier-quality loops are the 2026 frontier I. Avoiding one major recall can exceed the entire base-case value — held in the upside only.

exhibit 03 --layer="scenario-gen + simulation surrogates"

Virtual validation — €120M / yr.

EXHIBIT 03 / SHEET 06

Generate validation scenarios and AI surrogate models to move 20% of physical prototype and test effort into simulation. The pool: €600M/yr of prototype builds, HIL benches, and test-track & road validation across active programs.

TABLE 03 — KPI MOVEMENTS · VIRTUAL VALIDATION
KPI	Baseline	Target	Impact
Physical prototypes / program	index 100	80	▼ 20%
Validation scenario coverage	index 100	1,000	▲ 10×
Validation cycle time	index 100	70	▼ 30%
Time-to-market (SOP)	baseline	−3–6 mo	▲ earlier

€600M prototype & test pool × 20% virtualized share = €120M / yr · direct cost only · cost €20M · return 6.0×

AI-generated OpenSCENARIO/CARLA scenes plus HIL automation displace physical builds. Evidence: requirements→CARLA configuration up to 98% pass (TUM, arXiv:2505.13263) E, dSPACE + AWS Bedrock scenario generation V, Simulink/Stateflow mutant gen 13× faster E. Pulling SOP forward 3–6 months brings contribution margin forward — frequently a nine-figure revenue-timing benefit, held in the upside.

sum exhibits/ --net --scenarios

The stack, and its brackets.

SEC 04 / SHEET 06

FIG. 01 — BASE-CASE VALUE STACK · €518M TOTAL

€240M

€158M

€120M

PREDICTIVE WARRANTY · 8.0× SOFTWARE & V&V ENGINEERING · 6.3× AI VIRTUAL VALIDATION · 6.0×

TABLE 04 — THREE SUB-CASES, EACH INDEPENDENTLY CLEARING A 6× RETURN
Use case	Cost pool	Net uplift	Annual value	Annual cost	Return
Software & V&V engineering	€880M	18%	€158M	€25M	6.3×
Predictive warranty & field detection	€3.0B	8%	€240M	€30M	8.0×
AI virtual validation (shift-left)	€600M	20%	€120M	€20M	6.0×
Total value at stake			€518M	€75M	~7×

FIG. 02 — SCENARIO RANGE

Conservative ~€320Mlow rates, slow adoption

Base case ~€520Mevidenced low-end, net of tax

Upside ~€780M+incl. recall avoidance & TTM

plan capture --horizon=36mo

Capturing it: three phases.

SEC 05 / SHEET 06

PHASE 1 · 0–6 MO · QUICK WINS

— Software co-pilot rollout (non-safety)
— Warranty early-detection pilot on the connected fleet
— Instrument the DORA baseline before go-live
— Budget for the J-Curve dip (~15% over ~3 months)

~€80M RUN-RATE

PHASE 2 · 6–18 MO · SCALE

— Fine-tuned domain SLMs for MISRA / AUTOSAR / test
— Virtual validation & scenario generation at program scale
— Agentic root-cause for warranty

~€300M RUN-RATE

PHASE 3 · 18–36 MO · EMBED

— Supervised agents across the V-model (gated)
— ISO 26262 / ISO-PAS 8800 evidence automation
— Closed-loop data flywheel

~€520M+ RUN-RATE

cat methodology.md

Method notes — built to be challenged.

SEC 06 / SHEET 06

ROI FORMULA ROI% = (Value − Investment) / Investment. Value = Σ (cost pool × net realized %). Net % = (gross gain × adoption) − verification tax − instability tax, shaped in year one by the J-Curve dip. Safety-critical control software contributes zero to the model.

COST POOLS & RATES Pools are illustrative for the premium-OEM archetype — swap in a client's actual baseline and every downstream number re-derives. Improvement rates use the conservative end of evidenced ranges; vendor-reported figures are flagged un-audited.

AI IS AN AMPLIFIER Returns depend on the organizational foundation — internal platform, data quality, small-batch and version-control discipline — not the tool alone. Investment scope covers platform, compute, fine-tuning, integration and change management, not just licenses.

PROVENANCE Method follows DORA's 2026 ROI of AI-Assisted Software Development framework (verification tax, J-Curve, headcount-reinvestment, four key metrics). Prepared as a discussion exhibit — every number is open to challenge; that is the intent.