0%
Forward Deployed Engineering
The Role
Discovery and Scoping
Building in the Field
Running Deployments
Deploying Into Customer Environments
Where the software runs shapes almost everything else in a deployment: who can reach it, who operates it, how it is upgraded, how fast problems are fixed and what the security review will ask. Some customers are happy with a shared cloud service. Others insist the software runs in their own cloud account, in their data centre or on a network with no internet connection at all. This lesson covers the main deployment models, how to package software so it can run anywhere it needs to, how to roll it out safely, how to upgrade it and how to operate it when you cannot log in.
Multi tenant SaaS
In the multi tenant model the vendor runs one shared service and every customer uses it, with their data separated logically. It is the cheapest and fastest model for both sides: no installation, upgrades for everyone at once and one team that knows the system deeply. Most customers should start here if their security and regulatory needs allow it. The FDE work focuses on configuration, integration and data rather than infrastructure.
Single tenant hosted
In a single tenant hosted model the vendor runs a separate copy of the service for one customer, often in a dedicated cloud account or region. The customer gets stronger isolation, control over the upgrade schedule and sometimes a choice of region for data residency. The vendor gets more environments to operate. Single tenant works well for large customers with strict isolation requirements who still want the vendor to run the service for them.
Customer cloud
In the customer cloud model the software runs inside the customer own cloud account or virtual private cloud. Data never leaves their account, their security tools monitor everything and their network rules apply. The vendor usually operates it through limited, audited access or not at all. This model is common in financial services, healthcare and government, and it makes packaging, upgrades and support much harder because every environment is slightly different.
On premises and air gapped
Some customers run the software in their own data centre, and a few run it on networks with no connection to the internet at all. Air gapped deployments cannot download updates, call home or send telemetry. Everything, including container images, models, dependencies and documentation, must be delivered as a signed bundle through the customer transfer process. Support depends on what the customer can copy out, which is often only a small, reviewed text file.
Choosing with the customer
The choice is driven by the customer constraints, not by preference. Ask early about data residency, regulatory rules, security policy, internal operations skills and how they run comparable software. Many customers ask for the most isolated model by default and relax once they understand the trade offs: slower upgrades, more of their own staff time and higher cost. Write down the reasons for the chosen model, because they will be questioned at renewal.
Hybrid models
Many real deployments mix models. The application runs in the vendor cloud while a small agent runs inside the customer network to reach systems of record. Or the main service is shared but sensitive data is processed by a component in the customer account and only results leave it. Hybrid designs often satisfy security teams at a fraction of the cost of a full customer cloud deployment. Draw the data flow clearly so reviewers can see exactly what crosses each boundary.
Data residency
Regulations and contracts may require data to stay in a country or region. This affects where the service runs, where backups are stored, where logs and telemetry go and which support staff may access data. Residency often applies to sub processors too, such as model providers or email services. List every place customer data goes, including temporary copies, and check each against the residency requirement.
Cost of each model
Every isolated environment has a fixed cost: infrastructure, monitoring, upgrades, security patches and people time. A product with forty customers in the shared model and five in customer clouds may spend more engineering time on the five. Make sure the price of isolated models reflects this, and track the operating effort per deployment so the business can see what each model really costs over a year.
Designing for every model
Products that will be deployed in several models should be built so that the same code runs everywhere, with differences in configuration only. Avoid hard coded vendor services, assume no internet access unless configured, make every external dependency replaceable and keep telemetry optional. Building this way from the start is far cheaper than retrofitting a SaaS only product for its first air gapped customer.
What the FDE owns
In shared SaaS, the FDE owns configuration and integration. In isolated models the FDE often also owns the installation, the first upgrades and the relationship with the customer platform team. That is a lot of operational work for one person, so agree early which parts the vendor platform team, the customer team and the FDE each own, and write it into the deployment plan.
Questions to ask in discovery
A short list of questions settles most of the decision. Where must the data live? May it leave the customer network, and if so, to which regions and providers? Does the customer have a platform team that runs containers or Kubernetes today? Which cloud and which services are approved? How are changes to production approved, and how often? Is there internet access from the environment, directly or through a proxy? Who will be called at night if the system fails? The answers often rule models in or out before any technical design starts.
Operations maturity
A customer that asks to run the software themselves must be able to. Some have strong platform teams with clusters, monitoring and on call rotations already in place. Others have one infrastructure engineer and a ticket queue. Assess this honestly before agreeing a self run model, because a deployment that the customer cannot operate will fail in ways that look like product defects. Where maturity is low, propose a model where the vendor keeps more of the operation.
Contracts and responsibility
The deployment model changes who is responsible when things go wrong. In shared SaaS the vendor commits to uptime and security. In a customer cloud the customer runs the infrastructure and the vendor supplies software and support. Make sure the contract and the support terms match the model, including who patches the operating system, who monitors availability and how quickly the vendor must respond. A mismatch here turns every incident into a dispute about whose fault it was.
Moving between models
Customers sometimes start in one model and move. A pilot may run on shared SaaS with sample data before production moves to a customer cloud, or an isolated customer may later agree to move to single tenant hosting to get faster upgrades. Design data export and import so a move is a planned project, not a rebuild.
Reviewing the choice
Revisit the model at each renewal. Constraints change: a regulator publishes new guidance, the customer moves to a cloud provider or their security team gains confidence in the shared service. A model chosen for good reasons three years ago can become an expensive habit, and both sides benefit from asking whether it still fits.
Keep a short record of the original reasons for the model so that this review starts from facts rather than from memory, and so a new account team can see why the choice was made.
Update the record whenever the model changes, with the date and the reason.
Start customers on the shared service when their constraints allow, and move to single tenant, customer cloud or air gapped only for clear reasons that you write down. Consider hybrid designs, check residency for every copy of the data, price isolation for its real cost and build one product that runs in every model through configuration.