Key stats
- 3 permission types: read, export, and write
- 2 breadth dimensions: people with access and data each person can reach
- 4 fields per grant: scope, purpose, approver, and expiry
- 1 removal test: prove the access is gone in every connected system
Key takeaways
- Measure breadth by people and data domains, not by the number of accounts alone.
- Treat export as a separate permission because a copy can leave the controlled system.
- Replace shared credentials before measuring roles, or the count will understate real access.
- Review exceptions first and give every one an expiry and an owner.
- Prove one removal end to end so a partial revocation does not survive the next review.
Question and scope
This report examines the breadth of production data access held by an offshore support team. The unit is one account or role and the set of production data it can read, export, or change. The scope covers support engineers and maintenance developers who work under a buyer's data policy. It excludes service accounts used by the application itself. The boundary matters because a support engineer and an application account carry different review expectations.
The question is whether a buyer can see how much production data the support pod can reach and whether that reach matches the work. A broad grant is not automatically a failure, because support sometimes needs to inspect a record to resolve a case. The report treats breadth as a design choice that should be visible, reviewed, and reversible. The report also treats a narrow grant with a long duration as a breadth problem, because time multiplies exposure.
What breadth means
Breadth has two dimensions: the number of people with access and the amount of data each person can reach. A small team with a warehouse-wide read role can have more breadth than a larger team limited to a support console. Measure both dimensions rather than counting accounts alone. A count of accounts can look small while one role reaches every tenant in the warehouse.
Include the ability to export, not only the ability to view. An export turns a narrow view into a portable copy that can leave the controlled system. The same record seen on a screen and downloaded as a file carry different risks. Record which roles permit export and where the files can go. Export should be reviewed separately because a file can be copied after the session ends.
Why standing access grows
Standing access accumulates through small exceptions. An engineer needs one table for a case, and the simplest grant is a role that covers the whole schema. The grant stays after the case closes because nobody owns the removal. A year later the role is part of onboarding and no one remembers why. The renewal step should ask for the original purpose, not only for a signature.
Shared accounts and copied credentials also widen access without a visible grant. A team that shares an administrator login can reach anything that login can reach, regardless of the role assigned to each engineer. The audit sees one account and misses the number of people behind it. Named accounts are a prerequisite for measuring breadth at all. Replacing shared credentials is the first control because it makes every later measurement meaningful.
Task-scoped access
Task-scoped access ties a grant to a case, a data set, and an expiry. The engineer requests the view needed for the task, and the approver sees the purpose and the duration. The grant can be narrow because the task is narrow. This model costs more approvals, so it works best for sensitive data and exceptional requests. A task-scoped grant should also name the reviewer who can extend it if the case runs long.
A support console can provide a middle path. It offers a limited set of queries or masked views without giving direct database credentials. The console logs each lookup and can enforce a purpose field. The team gets the record it needs, and the buyer keeps a central record of who looked at what. The console should record the purpose field even when the lookup returns no rows.
Evidence and review
Each grant should record the requester, the approver, the data scope, the purpose, and the expiry. The record belongs in the same system that grants the access so the audit can follow the chain. A spreadsheet maintained beside the identity provider tends to drift from the real permissions. The grant record should be queryable so the next review does not depend on a manual spreadsheet.
Review the grant against actual use. A role that has not been used in a quarter may still be needed for an emergency, but that is a decision to record rather than assume. Look at the audit log for the queries and exports the role enabled. Usage evidence turns the review from a permission list into a risk conversation. A role with no use and no emergency note is a candidate for removal, not an automatic keep.
Measuring breadth
Build a matrix of roles and data domains, then mark which roles can read, export, or write each domain. Count the people assigned to each role. The matrix shows where a single role spans unrelated domains, which is often the clearest sign of over-breadth. The matrix should be reviewed with the support lead because the labels on paper may not match the work.
Track exceptions separately from standard roles. Exceptions are the grants most likely to be forgotten and the ones a reviewer should inspect first. Report the number of open exceptions, their age, and the domains they touch. A stable exception count with old entries is a sign that the removal step is not working. An exception with no expiry should be treated as a standing grant and reviewed as such.
A practical protocol
Start with the identity provider export and the database grant list. Reconcile the two so every named engineer maps to a known role. Identify shared accounts and service credentials used by people, and replace them before measuring anything else. The reconciliation should include service accounts that people use, because those are often the widest path. The reconciliation should be signed by the person who owns the production data.
Mark each role as read, export, or write for each data domain. Count the people in each role and the open exceptions. Pick the broadest role that is not required for daily support and scope it to the tasks the team actually handles. Set an expiry for the change and a review date. The scoped role should be tested with a real support case before the old role is removed.
Check one removal end to end. Revoke a grant for a test account and confirm the engineer loses access to the console, the database, and any connected tool. A removal that works in one system but not another leaves a path the next review will miss. The removal test should include the reporting tool and any saved query that caches credentials.
A worked example
A support pod of six engineers holds a read role on the production database. The role was granted for a single incident two years earlier and now covers every schema. The audit log shows that most lookups use three tables, while the rest of the role is unused. The unused part of the role was the part nobody could justify when the review asked for a purpose.
The buyer replaces the role with a support view limited to those tables and a masked account column. Two engineers keep a temporary exception for billing investigations with a ninety-day expiry. The next review finds no shared credentials and a much smaller export path. The support work continues without a new approval for every case. The exception should be reviewed at ninety days rather than renewed automatically.
Data to collect first
Record the account, the role, the data domains, the permission type, the grant date, the approver, and the expiry. Add whether the account is named or shared and whether export is allowed. Keep the identity provider export and the database grant list so the two can be compared. The record should be stored where the identity team can see it, not only in the support pod's notes.
Collect a quarter of audit activity to see which grants are used. If the logs are incomplete, mark the gap rather than inferring use from tickets. An access review built on missing logs should say so and focus on removing the broadest grants first. A missing log should be recorded as a gap with an owner, not filled with an assumption.
Counterevidence and limits
Broad access can be legitimate for a small team that handles urgent production support around the clock. Removing it without a replacement can slow a recovery or push engineers toward workarounds. The goal is a visible decision with a compensating control, not a uniform narrow grant for every team. The compensating control should be named in the decision so the risk is visible to the next reviewer.
Access breadth also changes with the data model. A new table can fall inside an existing schema-wide role without any new approval. Review the role after a schema change as well as on a calendar. The measurement is a snapshot, and the underlying permissions can grow between reviews. A schema change should trigger a breadth review even when no new account was created.
Evidence table
| Signal | What to inspect | Owner |
|---|---|---|
| Outcome | Acceptance evidence for the bounded task | Task reviewer |
| Control | Access, test, and approval boundary | Internal owner |
| Handoff | Open risks and next decision | Next owner |
Good distributed work is observable at the handoff: the result, evidence, limitations, and next owner are explicit.
Frequently asked questions
Is standing production access ever acceptable?
Yes, when the support need is continuous and the data is bounded. Make it a recorded decision with a review cadence and a compensating control such as query logging.
What should a support console log?
Log the named user, the record or query, the purpose, and the time. The log should be reviewable without exposing the record itself to a wider audience.
How often should breadth be reviewed?
At least quarterly, and after a schema change or a team composition change. A new table can widen an existing role without a new grant.
