5.4.6. Business Continuity and Performance at Scale

💡 First Principle: As AI becomes embedded in core processes, its failure becomes a business failure. Scaling therefore changes the risk question from "will this pilot work?" to "what happens to the business if this system fails, slows, changes, or becomes unaffordable?" Continuity planning and performance management protect operations while AI scales.

Factors to evaluate throughout scaling, as the exam guide's skill 4.4.6 requires:

• Dependency and failure modes: Which processes stop if the AI is unavailable? Is there a fallback (manual process, rules-based alternative, degraded mode)? Can the AI feature be switched off quickly (kill switch)?
• Vendor and model dependency: What if a provider changes pricing, retires a model version, or has an outage? Are there contractual protections, alternative models, or multi-model strategies? Has the impact of model updates been tested?
• Capacity and performance: Will response times and throughput hold at enterprise volume and peak times? Are quotas and capacity planned (for example, provisioned capacity for predictable high-volume workloads)?
• Cost at scale: Will consumption-based costs grow faster than value as adoption increases? Are budgets, alerts, and optimization in place?
• Quality at scale: Do accuracy, grounding, and fairness hold across new regions, languages, customer segments, and use patterns?
• Rollback and release management: Can a change to a model, prompt, or knowledge base be rolled back quickly if quality drops?
• People continuity: Do enough people retain the skills to operate the process if AI must be paused? Is knowledge concentrated in a few individuals?
• Regulatory and contractual continuity: Will expansion into new jurisdictions or customer segments change compliance obligations?

These factors should be evaluated at each expansion wave, not once. A system that performed well for one region may behave differently with another language, regulation, or customer base.

⚠️ Exam Trap: When asked how to protect the business as AI scales, choose answers that evaluate multiple factors (fallback and rollback plans, vendor and model dependency, capacity and performance, cost at scale, quality across new populations, retained human skills) and revisit them at each expansion. Options that assume the AI will always be available, or that eliminate manual capability entirely without a fallback, are distractors.

Reflection Question: A bank's AI agent now handles 70% of customer inquiries across three countries. The model provider announces that the model version in use will be retired in 90 days. What continuity and performance factors must the bank evaluate, and what should already have been in place?