DMADV Project A3 – Automated Risk Assessment Tools

1. DEFINE

Charter:
Create AI-powered tools (Bayesian stats & outlier detection) for instant process, material, and supplier risk scoring in pre-validation—improving compliance and mitigation.
Scope: Supplier, material, and in-process data streams prior to process validation for regulated manufacturing.

2. MEASURE

SIPOC:

Suppliers	Inputs	Process	Outputs	Customers
Data scientists, QA, suppliers	Historical & live data Deviation logs Material specs	Real-time ingestion Model training Dashboards	Risk ratings, alerts Predictive reporting	Validation team QA Regulatory Affairs

Data Collection & Analysis:

• Aggregate historic and in-process material/supplier data
• Apply Bayesian and SPC-based outlier analysis per SPC Glossary Table.html

Data Integrity Analysis:

• Completeness and double-check audits
• MSA/Gage R&R studies to validate accuracy
• Control charts for ongoing signal checks

3. ANALYZE

Cause & Effect Table:

Effect	Likely Root Causes
Late risk detection	Manual data, static models, lagging indicators
False positives/negatives	Poor integration, bad input quality, blind spots

5 Whys:

1. Risks missed until late? Data isn’t integrated
2. Why? Tools are manual/static
3. Why? No real-time input
4. Why? Varied data format
5. Why? Risk analytics not core to process

4. DESIGN

Design of Experiments (DOE):

• Test effects of supplier rating, environmental and process variables using factorial DOE frameworks from BlackBelt Glossary.pdf

Variation Reduction Planning:

• Standardize data ingestion and risk rating format
• X-bar & R, I-MR charts to detect/monitor variation
• Automated anomaly-check for high impact outliers

Hypothesis Testing:

• Null: AI model does NOT improve risk accuracy/speed
• Compare pre/post tool deployment—metrics: F1, accuracy, speed, using t-value/ANOVA as outlined in SPC Glossary Table.html

5. VERIFY

FMEA (Sample):

Failure Mode	Effect	Severity	Occurrence	Detection	RPN
Data entry error	Missed risk signal	9	3	3	81
Model drift	False security	8	4	4	128

Control Plan:

• Weekly model output audits/alert tracking
• Quarterly MSA and revalidation of risk detection tools
• SOP for retraining and data correction

Lessons Learned:

• Cross-functional team engagement accelerates adoption
• Data quality gaps must be corrected at the source
• Iterative, statistical validation is key to confidence

Overall Impact:

• Faster, more accurate risk detection and supplier oversight
• Pre-validation becomes predictive & compliance-first
• Culture of proactive, data-driven improvement