Three deployment models, not three guarantees
Software can be classified by where and how it runs: on-premises, in the cloud or hybrid. On-premises software runs on an organization’s computers or servers; cloud services use a provider’s resources and are accessed over the internet; a hybrid environment combines on-premises resources with cloud services. These categories describe deployment models, not the features of any particular program.
In the cloud, the provider manages the physical infrastructure and some service components. That does not mean it automatically manages the customer’s permissions, configuration or obligations. The specific division of responsibilities depends on the product and contract. AWS, for example, describes a shared responsibility model: it protects its infrastructure, while the customer remains responsible for the data they upload and how access is configured. So “on-premises” does not mean absolute control, and “cloud” does not guarantee security or continuity.
Google Cloud includes hybrid deployments in the scenarios covered by its architecture recommendations. Its framework also recommends documenting the architecture and favoring simple designs and loosely coupled components. Compare the actual architecture and its associated responsibilities, not just the offering’s label.
Access, offline work and recovery
A useful first criterion is what happens when the internet connection fails. In a service available exclusively online, an outage can prevent access or delay communication with the remote service. Check each workflow to identify which data and operations require a connection; CISA notes that cloud services are accessed over the internet.
Continuity also depends on what has been backed up, how often and how it can be restored. A copy on the same device can be lost along with the original in a failure, theft or incident affecting both. CISA recommends the 3-2-1 rule: keep three copies of important data, on two types of media, with one copy off-site. Physical or logical separation helps limit common points of failure, but a backup is useful only if it can be recovered.
CISA recommends keeping important backups offline and encrypted, and regularly testing their integrity and availability in a recovery scenario. Periodic restoration tests show whether backed-up data can actually be recovered.
Data, privacy and responsibilities
Data location is an architectural question and, sometimes, a compliance issue. A cloud service may let customers choose storage regions, but whether that option is available—and which services can be deployed in each region—varies. AWS says customers can select regions for storing content and that not all services are available in every region. For organizations with geographic or regulatory requirements, the decision should include where backups, logs and integrations are processed, not just where the main database is stored.
In the cloud, it is essential to divide responsibilities. AWS describes a model in which the provider protects its global infrastructure, while the customer remains responsible for uploaded data and access settings for services and resources. Measures AWS recommends for its users include protecting credentials, assigning permissions by role and using multifactor authentication. Managed infrastructure does not replace identity management or customer decisions about data classification, retention and access.
An on-premises installation does not remove privacy or security obligations either: moving a server into the office does not determine who can see the data, how it is encrypted or how incidents are handled. AWS’s privacy guidance for Argentina says its customers retain control of their content and must implement additional measures based on their needs; it also describes region selection and controls such as encryption and access management. Applicable rules depend on the country, sector, data and contract. Check jurisdictional requirements for the specific situation, with appropriate advice when needed.
Maintenance, updates and integrations
CISA recommends applying timely, regular updates to software and operating systems, prioritizing servers exposed to the internet. Before choosing a deployment model, check who is responsible for keeping systems updated and what support the provider offers.
CISA says moving certain systems to managed providers can reduce, but not eliminate, maintenance work for small and medium-sized organizations. The agency also recommends keeping software and operating systems up to date. Less in-house operation does not mean zero work.
In a hybrid environment, connections between systems add dependencies: synchronization, authentication, interface versions and network availability. Google Cloud’s architecture framework recommends documenting the architecture and favoring simple designs and loosely coupled components, which can be updated or managed more independently. In practice, request an inventory of integrations and clarify what happens if one stops working. It also matters how data can be exported and which formats the provider supports: do not assume a backup is equivalent to a usable migration.
Total cost: beyond the license or subscription
A useful cost comparison covers the expected period of use, not just the initial price. Request estimates based on comparable assumptions and review how each provider presents its pricing and terms. Google Cloud’s Well-Architected Framework includes cost optimization alongside reliability, security, operations and performance. It is a reference for structuring an evaluation, not a price list or a savings guarantee.
Google Cloud’s Well-Architected Framework includes cost optimization among its design areas, alongside reliability, security, operations and performance. Use it to structure an evaluation, not as a price list or a savings guarantee.
Request an estimate based on comparable assumptions, such as the number of users and data volume. Ask the provider about pricing terms and options for exporting data or ending the service. Compare proposals using the same criteria before choosing.
Which model best fits each need
Teams can compare deployment models against their operational needs. A team working where connectivity is unreliable may prioritize verified local features and independent backups; a distributed organization that needs shared access may value remote services, provided it also reviews network dependencies, permissions and recovery. Check what each solution actually offers and which responsibilities remain with the customer.
Confirm requirements for each application and test critical processes before committing data or equipment.
| Main need | What to check first | Question for the provider |
|---|---|---|
| Work with unreliable internet | Offline features and later synchronization | Which operations and data remain available during an outage? |
| Collaboration across locations | Role-based access, availability and recovery | How are access rights revoked and service or data restored? |
| Data-location requirements | Regions for storage, processing and backups | In which regions are this offering’s data stored and processed? |
| Limited in-house administration | Tasks retained by the customer and scope of support | Who handles updates and manages identities, integrations and backups? |
| Future migration needs | Export options, formats, dependencies and exit charges | Which data can be exported, and at what cost and on what timeline? |
Checklist before signing or migrating
A sound decision starts with an inventory of applications and data, and a definition of what must keep working during an outage. For each critical process, describe users, devices, internet dependencies, integrations and recovery objectives. If the organization handles personal, financial or regulated data, add the applicable requirements and verify whether the selected region and specific service meet them; a provider’s general compliance page does not replace an assessment of the specific situation.
Before signing, compare the technical and commercial proposal against concrete questions. Ask the provider to distinguish its responsibilities from the customer’s, describe backup and restoration options, specify how accounts are controlled and explain version changes. For an on-premises system, identify the people and resources responsible for maintenance and recovery; for the cloud, review access, export and termination terms; for hybrid systems, document which components depend on others.
A short checklist can keep a comparison from focusing only on the monthly price or server location:
- Continuity: Which critical features work offline, and how is service restored?
- Backups: Are copies separated, encrypted and tested through regular restorations?
- Access: Who manages permissions, multifactor authentication and account revocation?
- Operations: Who applies updates, monitors integrations and responds to incidents?
- Data and exit: Where is data processed, how can it be exported and what terms apply when the service ends?
- Cost: Which recurring, variable, support and migration costs are excluded from the base price?
Then compare two or three proposals using the same period, volume and service level. A limited trial can reveal whether offline work, restoration and integrations perform as needed, provided it uses agreed criteria and suitable data. The choice should match the team’s actual capacity and the impact of an outage: the best model is one the organization can govern, maintain and recover, not one that promises to eliminate every responsibility.