Lifecycle risk register
From requirements to post-release model updates. One row per failure mode, closed only with verified controls and evidence.
Independent engineering advisory
AI Product
Risk Assessment.
Your model will be wrong. The question is what happens next.
Risk assessment for products where a model's error reaches a person, a machine, or a balance sheet.
From model behaviour to product-level evidence
01Our perspective
Performance is a metric.
Control needs evidence.
Robust design joined to product safety. We examine the edges of your operating window - distribution shift, sensor degradation, rare cases and operator variation - and whether each control demonstrably works.
01Hardware & software
02AI component
03People & operations
04Operating environment
05Risk controls
02The assessment
Three concrete outputs.
Evidence you can use.
For your system and operating domain. Each failure mode needs a control, a test, an acceptance criterion and evidence - not just an agreement that it is unlikely.
Failure modes analysis
Data, model, integration and change-related failures: drift, label error, silent failure, degradation after retraining, and late or unavailable outputs.
Misuse scenarios
Foreseeable misuse paired with controls: out-of-domain use, over-trust, bypasses, untrained staff and erosion of human oversight.
Scope and deliverables agreed before engagement.
03Packages
Three engagements. Tangible deliverables.
AI PRODUCT RISK SCAN
Core AI function
- Failure modes for the critical path for core ai function
- Your existing controls mapped against them
- The gaps made explicit - which controls carry no test and no acceptance criterion
- A populated risk register you keep and extend yourself
- A recommendation on what a full assessment would need to cover
AI PRODUCT RISK ASSESSMENT
Full product
- Lifecycle risk register - from requirements to post-release model updates
- Failure modes analysis across data, model, integration and change
- Misuse scenarios, each paired with the control that answers it
- A test, an acceptance criterion and a recommended sample size behind every control
- A short method for keeping the register alive when the model changes
AI RISK ENGINEERING RETAINER
Ongoing support
- Change review whenever the model, the data or the operating domain moves
- The impact of each change on the risk file, decided and recorded
- Design input before new AI functions are committed
- Documentation reviewed ahead of audits or notified body submissions
- Agreed monthly hours support
Offer available in 2027
Complex or safety-critical systems: scope and quote agreed individually.
Pricing depends on the number of AI functions and the regulatory route. You get a fixed quote, not an hourly rate.
04The process
Focused from the start.
- 01
Understand
Your product, context, and key questions.
- 02
Define
Agree the scope, inputs, and deliverables.
- 03
Examine
Review evidence and challenge assumptions.
- 04
Prioritize
Discuss findings and practical next steps.
05Who is doing this
Ewa Marchewicz
Senior Robust Design and Product Safety Lead
Gdańsk, Poland
Over 17 years in regulated industries. From embedded firmware into robust design and product safety leadership. Now applied to machine learning.
MSc in Computer Science · Postgraduate qualification in AI
Dates that move the work forward
Start a conversation
Let's look at your product.
Share your product context and the decisions your AI makes.
Scope & boundaries
I produce the analysis and supporting evidence. The manufacturer declares conformity. This is not certification or a guarantee of compliance. Management-system implementations, AI policies, AI-literacy training and procurement questionnaires are outside this service.