Task 1.1: Designing a cloud solution infrastructure that meets business requirements.

• ALeave them on-premises and reach them from Google Cloud over a private hybrid link

  Correct

  This matches the stated plan not to move the integrations and the technical requirement to keep the legacy interfaces reachable from both on-premises systems and the cloud. The tiered hybrid pattern keeps such backends on-premises while new components run in Google Cloud.

• BRehost them on Compute Engine with Migrate to Virtual Machines before the lease expires

  Incorrect

  Rehosting is a real lift-and-shift path, but it spends migration effort and risk on systems the company has decided not to move and plans to replace.

• CRefactor them into Cloud Run services so they can scale with new provider traffic

  Incorrect

  Refactoring rewrites the integrations, which is the upgrade the company has ruled out for systems that will be replaced anyway.

• DRepurchase a SaaS integration platform and retire the existing provider interfaces now

  Incorrect

  Repurchasing replaces the interfaces with a SaaS product now, which contradicts the plan and would force every insurance provider to change how it connects.

Not every workload has to move in the first wave. When a system is scheduled for replacement and the business has decided not to touch it, the architect keeps it where it is and connects it to the new cloud components, here through a private hybrid link so the legacy interfaces stay available to both environments.

• AAgree on business KPIs, such as listing time and editor rework rate, and baseline them

  Correct

  Business value metrics tied to a stated objective, measured before and after, are what show whether the pilot achieved its goal and earned a return.

• BBuild a Cloud Monitoring dashboard of model latency, error rate and tokens per request

  Incorrect

  These are useful operational and cost signals, but they say nothing about whether editors list products faster or with fewer corrections.

• CScore each generated description for fluency and coherence with automated metrics

  Incorrect

  Quality scores describe the generated text, but a fluent description is not proof that the business outcome improved.

• DCompare the price per million tokens of several Gemini models in the pricing calculator

  Incorrect

  Unit cost is one input to ROI; without business KPIs there is no return to weigh it against.

The Well-Architected Framework's AI and ML guidance asks for SMART objectives and ROI-focused KPIs that cover both costs and business value, such as efficiency and accuracy. Agreeing on those metrics and their baseline before the pilot starts is what makes the end-of-pilot decision measurable.

• ACall the partner API synchronously from checkout with a longer request timeout

  Incorrect

  A longer timeout makes shoppers wait and still loses the order when the partner is down for minutes.

• BPublish each order to a Pub/Sub topic and deliver it to the partner from a subscriber

  Correct

  Pub/Sub decouples checkout from the partner: the order is stored durably once published, and the subscription redelivers it until the subscriber acknowledges a successful call.

• CAttach a dead-letter topic to the subscription with a maximum number of delivery attempts

  Correct

  After the set number of failed deliveries, Pub/Sub forwards the message to the dead-letter topic, where the team can inspect it instead of retrying forever.

• DRaise the maximum number of instances of the checkout service on Cloud Run

  Incorrect

  More checkout instances do not help when the downstream partner API is the part that is unavailable.

• EKeep each order in Memorystore for Redis until the partner API responds

  Incorrect

  Memorystore is an in-memory cache, not a delivery system: the team would have to build retries and failure handling itself.

Asynchronous messaging is the standard integration pattern for an unreliable external system. Pub/Sub absorbs the outage and retries delivery, and a dead-letter topic catches the messages that never succeed so they can be investigated.

• ABuild a custom CRM on GKE with Spanner as its operational database

  Incorrect

  Building a CRM from scratch takes long and needs new skills for a capability the company does not differentiate on.

• BBuy a SaaS CRM and integrate it with KnightMotives systems through APIs

  Correct

  A repurchase moves the capability to a cloud-hosted SaaS product, which is fast to adopt, needs no dealer hardware and avoids building a commodity system.

• CModify the outdated ERP so that dealers can record customer interactions

  Incorrect

  Extending the outdated ERP adds to the technical debt the case describes and delays a modern dealer experience.

• DRehost an on-premises CRM package onto Compute Engine VMs unchanged

  Incorrect

  Rehosting keeps the old software and its limits; it moves the problem rather than modernizing the experience.

Workload disposition asks whether to build, buy, modify or retire. For a commodity capability that must reach many users quickly with no new hardware, buying a SaaS product and integrating it is usually the right call.

• ARequester Pays turned on for the bucket that holds the dataset

  Correct

  With Requester Pays, requests must name a billing project, and the requester's project is billed for the access and network charges.

• BThe Coldline storage class set as the dataset bucket's default

  Incorrect

  A colder class lowers the storage price but adds retrieval charges, and the bucket owner still pays for every download.

• CLabels on the bucket that name each university that reads it

  Incorrect

  Labels organize and filter the owner's own costs; they do not move the charges to anyone else.

• DUniform bucket-level access, with read granted to each university

  Incorrect

  This controls who can read the objects, but the bucket's project is still billed for their requests.

The business requirement is about who pays, not how much storage costs. Requester Pays shifts operation and network charges for each request to the project that the requester names, which suits large datasets shared with many outside organizations.

• AAn Eventarc trigger on object finalization that invokes a Cloud Run service

  Correct

  Eventarc routes Cloud Storage object events to Cloud Run, which scales with the number of arriving files and runs nothing while idle.

• BA Compute Engine VM that lists the bucket every minute for new objects

  Incorrect

  Polling from a VM is the server the team wants to avoid, and it scales poorly when thousands of files arrive.

• CA nightly Storage Transfer Service job that copies the new files for loading

  Incorrect

  A nightly job misses the requirement to process each file within minutes.

• DA BigQuery Data Transfer Service schedule that loads the bucket hourly

  Incorrect

  An hourly batch load cannot validate each file as it arrives, and it delays processing.

Event-driven integration lets external parties push data at their own pace. Storage events delivered by Eventarc to Cloud Run give near-immediate, scale-to-zero processing.

• AMembers can upload photos of the damage when they file a claim

  Incorrect

  Uploading photos is a capability the portal offers to users, which makes it a functional requirement.

• BPages must load in under two seconds for 95% of requests at peak load

  Correct

  A response-time target describes how well the system performs, which is a performance requirement, not a feature.

• CAdjusters can approve or reject a claim and add a comment to it

  Incorrect

  Approving or rejecting claims is a business function of the portal.

• DThe portal must stay available during the failure of a single zone

  Correct

  Surviving a zone failure is a reliability requirement on how the system behaves, not on what it does.

• EThe portal emails members whenever their claim changes status

  Incorrect

  Sending status notifications is a feature, so it is functional.

Functional requirements say what the system does; non-functional ones say how well it must do it. The Well-Architected Framework's pillars cover the non-functional areas: security, reliability, performance, cost, operations and sustainability.