Task 4.4: Scale AI from pilots to enterprise-wide deployments.

• ADeploy every proven solution to every department at once

  Incorrect

  Enterprise-wide rollout with reusable components belongs to the Scale level, after a few production solutions have proven value and operations.

• BPut a few proven solutions into production with governance and monitoring

  Correct

  At AWS's Launch level, organizations move from proofs of concept to methodical deployment of select, proven solutions into production with robust governance, monitoring and support.

• CRun four more proofs of concept before putting anything into production

  Incorrect

  More experiments delay value. The validated ones are ready for a controlled launch, which is how pilots avoid stalling.

• DReturn to envisioning new use cases, because the proofs of concept are complete

  Incorrect

  Going back to envisioning abandons validated results instead of delivering their value in production.

Transformation progresses through envision, experiment, launch and scale. After successful experiments comes a disciplined launch of a few production solutions that deliver measurable value and establish the governance and operations needed to scale.

• AA central team that builds every AI solution for every business unit

  Incorrect

  Centralizing all delivery creates a bottleneck and removes business units' ownership; a center of excellence enables them instead.

• BA ban on AI work in business units until a group strategy is written

  Incorrect

  Stopping all work loses momentum and value; the problem is a lack of sharing and standards, not too much activity.

• CAn annual conference where units present their AI results

  Incorrect

  Occasional showcases share stories but do not provide the ongoing guidance, standards and shared capabilities that scaling needs.

• DAn AI center of excellence that shares guidance and capabilities

  Correct

  AWS guidance recommends an AI center of excellence that offers guidance, best practices and technical capabilities and engages with business units to identify opportunities while maintaining standards.

A center of excellence is a cross-functional mechanism for scaling AI: it spreads best practices and reusable capabilities, sets standards and works with business units, so that each unit moves faster without repeating others' mistakes.

• AStart with a few well-defined projects that show clear value, then scale

  Correct

  AWS recommends starting with well-defined, limited-scope projects that demonstrate clear business value, documenting results and learnings, and scaling successful implementations gradually.

• BLaunch one enterprise-wide transformation program covering every function

  Incorrect

  A single large program delays visible results and concentrates risk, which makes it hard to show progress within six months.

• CSpend the first year building a platform before delivering any use case

  Incorrect

  Platform work with no early wins leaves sponsors without evidence of value and weakens momentum.

• DPick the most technically ambitious use case to prove what AI can do

  Incorrect

  Ambitious projects carry the most risk of delay and failure, the opposite of what an early win needs.

Short-term wins build credibility, funding and learning for larger efforts. Well-scoped projects with clear value, documented lessons and a gradual path to scale let the organization move fast without betting everything on one large program.

• AProduction governance, monitoring, incident escalation and service levels with a support team

  Correct

  AWS guidance for moving from experiment to production includes production-grade governance, observability and monitoring, an incident response and escalation process, SLAs with performance metrics and tiered support with clear escalation paths.

• BA training program that prepares all 3,000 dispatchers to use the assistant from day one

  Incorrect

  Training is part of a launch, but it prepares users, not operations: nobody would detect a degraded assistant at night or own the response.

• CA load test proving that the assistant can serve 3,000 dispatchers at peak volume

  Incorrect

  Testing under realistic load is part of preproduction readiness, but it proves capacity once; it does not watch the live service or define who responds when it fails.

• DA move of the notebook to a larger instance so that it can serve every dispatcher's requests

  Incorrect

  A notebook is an environment for experimentation. More capacity keeps the service dependent on two people, with no monitoring, escalation or service levels.

The step from experiment to production is about operations and governance as much as about the model. Thousands of users need a service that someone watches, that has a defined way to escalate incidents and that is held to service levels, so that the business can depend on it.

• AThe pilot least likely to fail, even if its effect is small

  Incorrect

  A safe success with little effect does little to shape the next round; the framework says Launch-phase pilots should be highly impactful.

• BMany small pilots across departments at once, to spread the risk

  Incorrect

  The Launch phase delivers pilot initiatives in production to demonstrate business value; spreading effort across many small pilots dilutes the impact any one result can show.

• CHighly impactful pilots whose success can shape next steps

  Correct

  The framework says Launch-phase pilots should be highly impactful so that, if successful, they help influence future direction.

• DThe pilot that needs the fewest changes to current processes

  Incorrect

  Ease of introduction says nothing about the business value a pilot will demonstrate, which is what the Launch phase is meant to show.

Early pilots are signals as well as projects. Picking ones that matter makes their success persuasive to the rest of the organization, and what is learned from them shapes the approach before full-scale production.

• AKeep building every AI solution in the central team

  Incorrect

  Central delivery of everything becomes a bottleneck once business units can deliver themselves.

• BShift to setting standards and curating training

  Correct

  In AWS's Consolidation phase, when practices are self-sufficient, the center of excellence shifts to a supporting role: setting standards and best practices and providing curated training.

• CDisband the central team because units no longer need it

  Incorrect

  Economies of scale in standards and training remain valuable.

• DTake back all projects to ensure consistency

  Incorrect

  Recentralizing removes the units' ownership and slows delivery.

Centers of excellence evolve with maturity. As capability spreads, the center moves from delivery to enablement, keeping consistency while units move fast.

• ALaunch to everyone and add capacity if customers report slow answers

  Incorrect

  This finds the limits only after customers are affected, in the busiest week; AWS guidance is to establish performance under heavy load before it arrives.

• BExtend the 300-customer pilot for another quarter to gather evidence

  Incorrect

  A small pilot cannot show how the system behaves under its heaviest expected load, and another quarter would miss the season.

• CMove the assistant to the largest, most capable model available

  Incorrect

  Larger models can cost more per request, and changing the model without measuring the load leaves the real bottleneck unknown; AWS guidance is to find it through load testing first.

• DLoad test it at the expected peak traffic and fix the bottlenecks found

  Correct

  AWS guidance is to load test under average and extreme workloads, with a test suite that simulates the heaviest expected load, then identify the bottlenecks and adjust the architecture before demand arrives.

Moving from a pilot to enterprise scale changes the load a system must carry. Testing at the expected peak before the rollout finds bottlenecks while there is still time to fix them, which protects the business when it matters most.

• AExpanding it to the desired scale and sustaining its benefits

  Correct

  The Scale phase focuses on expanding production pilots and business value to the desired scale and making sure the business benefits last.

• BRepeating envisioning workshops to find new opportunities first

  Incorrect

  Identifying and prioritizing opportunities is the Envision phase, which the bank has already completed for this pilot.

• CRunning a new gap analysis across every CAF perspective first

  Incorrect

  Identifying capability gaps and dependencies is the Align phase, which comes before pilots are launched.

• DDeclaring the program complete and moving the team to other work

  Incorrect

  Stopping after one regional pilot leaves most of the value unrealized and the benefits unsustained.

A successful pilot is a starting point, not the finish line. The Scale phase turns proven value into broad, lasting business benefits.

• AA single weekend cutover for all three systems to shorten the change

  Incorrect

  A combined cutover concentrates the risk in one event; AWS's guidance for changing business-critical systems is to minimize disruption with zero-downtime transitions.

• BSwitching each system over with no way back, to avoid running both

  Incorrect

  Without a way back, a misbehaving service cannot be undone quickly; AWS calls for deployment strategies with rollback capabilities.

• CZero-downtime transitions with rollback capabilities

  Correct

  AWS's generative AI guidance on changing business-critical systems stresses minimizing operational disruption with zero-downtime transitions, using deployment strategies and rollback capabilities.

• DPlanned outage windows announced to business units in advance

  Incorrect

  An announced outage is still downtime for units that rely on the systems every day, which the guidance on zero-downtime transitions avoids.

Scaling AI-driven change across many systems must protect the business that depends on them. Gradual, reversible transitions keep operations running while modernization proceeds.