Security Reviews and Data Handling

Topics Covered

Security Reviews in Enterprise Deployments

Why customers review vendors

Who is involved

Typical stages

Start early

Risk depends on the data and the design

Standard materials

Speaking the security team language

Honesty about gaps

Conditions and commitments

Reviews repeat

Threat modelling the deployment

AI specific concerns

Working with the vendor security lead

Keeping the sponsor informed

Testing by the customer

After approval

Before an enterprise customer lets a new system touch its data, its security team will review it. The review can take days or months, and it often decides whether the deployment happens at all. Forward deployed engineers are usually the people who know the system best and who talk most with the customer, so they end up at the centre of the review: answering questions, explaining architecture, agreeing controls and making changes. This lesson covers how security reviews work, how to answer questionnaires, how to agree data handling, how to manage access and secrets and how to handle retention, deletion and audits.

Why customers review vendors

Every vendor that touches customer data adds risk. A breach at a vendor can expose the customer data, damage its reputation and bring legal penalties. Regulators expect customers to manage these risks, and many customers have suffered from vendor incidents before. A security review is how the customer decides whether a vendor is safe enough, what controls are needed and what risks they are accepting.

Who is involved

The review usually involves the customer security team, sometimes a privacy or data protection officer, legal and procurement, the IT team that runs identity and networks and the business sponsor who wants the deployment. On the vendor side, the FDE often works with a security lead, legal and the account team. Knowing who decides what, and who can approve exceptions, saves weeks.

Typical stages

Reviews often follow a pattern. First, an intake form describes the system, the data and the use. Then the customer classifies the risk, often by the kind and amount of data involved. Higher risk leads to a detailed questionnaire, requests for evidence such as audit reports and policies, one or more architecture calls and sometimes a test by the customer or a third party. The review ends with approval, approval with conditions or rejection, and conditions become commitments to track.

Start early

Security reviews are one of the most common causes of delayed deployments. Start the review as soon as the customer is serious, ideally in parallel with discovery or the pilot rather than after the contract is signed. Ask the sponsor to introduce you to the security team early, find out their process and timeline and share standard materials before they ask. A review started late can hold a finished deployment for months.

Risk depends on the data and the design

The same product can be low or high risk depending on how it is deployed. A system that reads anonymised aggregates inside the customer cloud account is far less risky than one that copies full customer records with personal data into the vendor cloud. Design choices made during scoping, such as where data lives, what is copied and what is masked, often decide how hard the review will be.

Standard materials

Prepare a security pack that answers the common questions before they are asked: an architecture diagram showing data flows, a description of where data is stored and processed, encryption in transit and at rest, access controls, logging, incident response, subprocessors, certifications and audit reports and data retention and deletion. A good pack can cut weeks from a review and shows the customer that security is taken seriously.

Start security reviews early, design to lower risk and bring a prepared security pack to shorten the path to approval.

Speaking the security team language

Security teams think in terms of data flows, trust boundaries, threats, controls and residual risk. Explain the system in those terms: what data goes where, who and what can access it, how it is protected, what could go wrong and how that is prevented or detected. Avoid marketing language. A precise, honest answer to a hard question earns more trust than a confident vague one.

Honesty about gaps

Every system has gaps. If a control the customer expects is not in place, say so, explain the current mitigation and offer a plan or an alternative. Never claim a control that does not exist. Security teams compare answers with evidence and with what they see later, and a false answer discovered after approval can end the relationship and create legal exposure.

Conditions and commitments

Approvals often come with conditions, such as single sign on before go live, data masking for a field, a penetration test within six months or a named contact for security incidents. Record every condition in a tracker with an owner and a date, share progress with the customer security team and close each one with evidence. Missed conditions can lead to the approval being withdrawn.

Reviews repeat

Security reviews do not happen only once. Customers review vendors again every year or two, when the deployment changes significantly, when new data is added or after an incident. Keep the security pack current, note significant changes as they happen and tell the customer security team before changes that affect data flows, rather than after.

Threat modelling the deployment

Before the architecture call, walk through the deployment as a defender. For each data flow and component, ask what could go wrong: unauthorised access, data leaving where it should not, tampering with inputs or outputs, loss of service and actions taken without accountability. Note the controls that prevent or detect each risk and any gaps. A short threat model makes the call far smoother, because the team has already thought through the questions the security team will ask.

AI specific concerns

Security teams increasingly ask about AI features. Common questions include whether customer data is used to train models, where prompts and outputs are stored and for how long, which model providers act as subprocessors, how the system resists instructions hidden in documents or messages, how outputs are checked before actions are taken and how access controls apply to retrieved content. Prepare clear, accurate answers, backed by provider terms and your own design.

Working with the vendor security lead

Most FDEs are not security specialists, and they should not have to answer every question alone. Bring the vendor security lead into calls on difficult topics, check unusual answers with them and escalate requests that would change company policy, such as a customer wanting to audit the vendor offices. The FDE coordinates and explains the deployment; specialists answer for company wide controls.

Keeping the sponsor informed

Sponsors often underestimate how long reviews take. Share the review plan and progress with them, flag blockers early and ask for help when the review stalls, for example when the security team has not responded for two weeks. A sponsor who understands the review can push for priority inside their own organisation.

A short threat model of each data flow and component prepares the team for the review and tracks conditions to closure.

Testing by the customer

Some customers ask to test the deployment themselves or through a specialist firm before go live. Agree the scope, timing, environment and rules in writing, and make sure the test targets a dedicated environment where possible. Treat findings like any other condition: fix them by severity, share evidence of the fix and ask for a retest of serious ones. A test with findings fixed promptly is a strong signal of good practice.

Plan the remediation time into the go live date from the start.

After approval

Approval is a start, not an end. Keep the security contact list current and send the security team a short update when conditions close.

Start early, design for lower risk and be honest

Start the review during discovery, learn the customer process and decision makers, design data flows to lower risk and bring a current security pack. Explain the system in data flows, boundaries and controls, admit gaps with mitigations and plans, track every approval condition to closure and tell the security team before changes that affect data.