Results, not transformation theatre.

Situation

A tier-1 supplier network feeding a major European aircraft manufacturer was drowning in verification backlog. Parts sat for weeks waiting for quality disposition.

What was at risk

Programme milestones, penalty discussions on every late shipment, and the credibility of the entire verification chain.

Baseline

Verification lead time of up to 30 days; delays escalating to programme level; no single owner of a part’s waiting time.

What I changed

• Mapped the verification chain end to end — 80% of the delay was routing and ownership, not inspection time.
• Redesigned routing so every part had one accountable owner and a clock that never stopped silently.
• Standardised quality gates so decisions happened at the gate — not in escalation meetings after it.

Measured result

How the result was verified

Lead-time KPIs tracked before and after in the programme’s own reporting; internally audited within the quality organisation.

What remained after handover

The routing verification KPIs and gate discipline stayed in daily use — no new software or headcount was needed to sustain them.

Similar backlog on your floor? →

Situation

A metal-forming supplier to major automotive OEMs was in permanent escalation: recurring complaints, special status with two OEMs, audit findings piling up faster than they were closed.

What was at risk

Existing business with the OEMs, new-programme awards, and a leadership team consumed by customer calls instead of running the plant.

Baseline

Special status with two OEM customers; a year of recurring escalations; 8D reports closed on paper but not in the process.

What I changed

• Rebuilt the complaint-to-root-cause discipline: every 8D closed with evidence, not paperwork.
• Installed a VDA 6.3–based internal audit programme run as coaching, not policing.
• Negotiated directly with OEM supplier-quality engineers to reset expectations and rebuild trust with a visible roadmap.

Measured result

How the result was verified

Escalation status is the customer’s own scorecard — both OEMs formally lifted the special statuses. Complaint and audit trends were tracked in the plant’s QMS.

What remained after handover

The audit programme and complaint discipline survived after I left — escalations end when the customer sees the same defect never twice, and that takes a system, not heroics.

In escalation with a customer? →

Situation

Scrap, rework and warranty were treated as separate line items owned by nobody. Individually each looked tolerable; together they were quietly consuming millions — the classic invisible factory.

What was at risk

Margin erosion with no single owner, and improvement budgets impossible to justify because the loss was never visible in one place.

Baseline

No consolidated cost-of-poor-quality model; loss drivers unranked; no owner, target or review cadence on any of them.

What I changed

• Built a single Cost of Poor Quality model consolidating scrap, rework, sorting, warranty and expedited freight.
• Ranked the top 10 loss drivers and assigned each an owner, a target and a monthly review cadence.
• Trained line leadership to read and challenge the numbers themselves.

Measured result

How the result was verified

Quality and finance built the model together, so finance owned the numbers from day one. Savings were verified against the P&L, not a quality-department estimate.

What remained after handover

The CoPQ model and the monthly ownership rhythm outlived the project — when CoPQ hits the P&L review, improvement funds itself.

Suspect an invisible factory? →

Situation

Before recommending AI-enabled operating models to clients, I built one for real: Velin, an autonomous AI company that runs my own operation — agents that propose, execute and verify work under human authority.

What was at risk

Credibility. An advisor who has never run AI in production is selling slides. This system is the proof that the governance I recommend actually works under load.

Baseline

A one-person operation with knowledge trapped in files and history — the same institutional-memory problem every complex company has.

What I changed

• Built an agent workforce with separation of duties, independent verification and a permanent journal.
• Added Velin Sovereign: a private model on my own infrastructure, answering from the company’s own record with a source under every sentence.
• Applied quality-system discipline throughout: human GO on anything that matters, audit trail, kill switch.

Measured result

How the result was verified

Every answer the system gives is recorded, sourced and reviewable. The acceptance run included deliberate trap questions about things never written down — all answered with “I don’t have that” rather than an invention.

What remains in operation

The system runs my operation daily and is demonstrable live — not a lab demo. See Velin · Ask Sovereign a question

How I build AI for companies →