Inference Public

AWS Quick guideTools you can use

A sample decision record for AWS Quick

A sample decision record for the AWS Quick choices that cannot be undone: account, sign-in, encryption keys and knowledge base permissions.

An officer signing a decision record at a meeting table

Last checked against AWS documentation on 20 November 2026.

Some AWS Quick settings cannot be changed once they are made. The AWS account, the sign-in method, the key that protects Quick's AI data and the permission setting on each knowledge base all fall into that group. Our AWS Quick FAQ explains why.

Because these choices are permanent, they deserve a written record: what was decided, on what evidence, and by whom. This page shows a filled-in example for a fictional council, so you can see what such a record looks like in practice. A blank template is available to download at the end.

This is a fictional example. Sample Unitary Council is invented, and so is everything in this record. The decisions below are illustrations of reasoning, and the right answers for your organisation depend on your own estate, policies and risk appetite.

How the record works

The record supports two separate decisions:

  1. Pilot authorisation. Before any configuration, the named officers approve the permanent and high-impact decisions and authorise a bounded pilot. This permits configuration and testing only.
  2. Operational decision. After the pilot, the same officers review the test results and remaining risks, then approve, condition or withhold wider use.

In our own engagements, the first decision is taken at the end of Discovery, before a Lighthouse demonstration, and the second is the Signed Promotion Decision at the end of the programme.

Each person approves only what is within their remit. Advisers, including the data protection officer and any external consultant, record their advice without approving. The record is amended and re-signed whenever an assumption changes, for example a new source, a new capability or a test result that contradicts what was expected.

The sample: Sample Unitary Council

Record control

FieldEntry
OrganisationSample Unitary Council (fictional)
Use caseStaff policy and procedure answers for HR, customer services and housing officers
Record versionv1.0, prepared for pilot authorisation
AWS accountNew dedicated account in the council's landing zone ("quick-prod")
AWS RegionEurope (London), eu-west-2
In scopeOne Quick account; two knowledge bases (staff policies; service procedures); SharePoint Online connector
Data controllerSample Unitary Council

Part A: Permanent and high-impact decisions

#DecisionChoiceReasoningApproved by
A1AWS accountNew dedicated accountAWS allows one Quick subscription per account. A dedicated account keeps Quick billing, logs, keys and admin access separateTechnical design authority
A2Sign-in methodIAM Identity Center, with an Enterprise edition accountStaff sign in with existing Microsoft accounts. The method cannot be changed laterTechnical design authority
A3Identity Center instanceExisting organisation instance, administered from the delegated accountAlready used for AWS console access; avoids a second identity siloTechnical design authority
A4Identity sourceEntra ID, with SCIM provisioning and SAML sign-inEntra ID is already authoritative. Quick groups rebuilt without nesting, because nested groups are not provisionedTechnical design authority
A5Encryption for Quick's AI dataAWS-managed keyThe council's security policy does not require customer-managed keys for OFFICIAL content of this kind. Chat memory is wanted, and customer-managed keys disable it. Revisited if sensitive sources are addedSIRO, with technical design authority
A7aPermissions, knowledge base 1 (staff policies)Document permissions offEvery document is published to all staff on the intranet already. Confirmed by the HR information asset ownerService owner, with HR information asset owner
A7bPermissions, knowledge base 2 (service procedures)Document permissions onSome housing procedures are restricted to named teamsService owner, with housing information asset owner
A8ResearchResearch allowed over knowledge base 1 onlyResearch cannot use permission-controlled knowledge bases. Knowledge base 1 holds only all-staff contentService owner
A9RegionLondonData stored in London. AI processing may use other European regions; recorded in C5Technical design authority

A6 (key safeguards) is not applicable, because A5 uses an AWS-managed key.

Part B: Configuration decisions

#DecisionChoice
B1RolesPilot users are Readers (Essentials). Two content leads are Author Pro (Enterprise). One Admin Pro
B1aLicence mix and billing owner22 Essentials, 3 Enterprise (two Author Pro content leads and one Admin Pro); billed to the digital services cost centre
B2Knowledge base ownersTwo owners for each knowledge base: the content lead and a deputy
B3Permission refreshDaily, the default. SharePoint is checked in real time when results are returned
B4ConnectorsSharePoint Online only for the pilot
B5Web searchOff for the pilot
B6LogsChat, feedback and agent-hour logs delivered to CloudWatch from day one; 12-month retention; access limited to two named analysts
B7AutomationNone in the pilot
B8OutputsUploads to Spaces off. Downloads allowed, with guidance that generated text is a draft
B9Cost controlsMonthly budget alert at 80 per cent; agent-hour report reviewed weekly

Part C: Information governance

#ItemOutcome
C1DPIA screeningScreening complete. Sources hold minimal personal data, but chat logs record staff questions against named users for 12 months. Full DPIA to be completed before staff access, covering prompts and logs
C2ATRS screeningInternal staff tool with no direct public effect. The mandatory scope covers central government departments and some arm's-length bodies. ATRS is recommended across the wider public sector, and government intends to extend the mandatory requirement over time. The council will publish a record voluntarily if use expands to public-facing work
C3FOI and subject accessRecords team can retrieve CloudWatch chat logs (delivered from day one) and downloaded files. Native history is visible only to each user and cannot be shared, so it is not relied on. Procedure written and tested on a dummy request
C4RetentionGenerated content is not a record unless filed by an officer. Native history (up to 90 days, visible only to each user) is not treated as a record
C5ClassificationOFFICIAL only. No case files, social care records or content above OFFICIAL
C6Staff guidanceTwo-page guide and a 30-minute briefing for pilot users
C7 to C9Health and social care itemsNot applicable. Social care content excluded

Part D: Known gaps and residual risks

#Gap or riskEvidence classMitigationOwner
D1Restricted housing procedures must not reach other teamsNOT_RUNPositive and negative access tests with representative users before any staff useTechnical design authority
D2Some HR policies on the intranet are out of dateDOCUMENTED (content audit)14 superseded policies withdrawn before connection; quarterly reviewHR information asset owner
D3Behaviour when SharePoint is unavailableNOT_RUNTest during pilot; record resultTechnical design authority
D4AI processing may take place in other European regionsDOCUMENTED (AWS)Accepted for OFFICIAL content in scopeSIRO

Part E: Pilot authorisation

RoleRemitDecision
SIROInformation risk acceptanceAuthorise with conditions: D1 tests passed and full DPIA complete before staff access
Service ownerUse case, scope, usersAuthorise
Technical design authorityAccount, identity, keys, regionAuthorise
Information asset owners (HR, housing)Content and permissionsAuthorise
AdviserAdvice
Data protection officerAdvises a full DPIA covering prompts and logs before staff access. Asks to be consulted before any case-related source is added
External adviserEvidence and recommendations provided; no approval role

Further reading

Using the template

The blank template has the same parts, with guidance in each row. Use it at the end of a readiness assessment or Discovery, before anything is configured, and keep it up to date through the pilot.

Work through the AWS Quick readiness checklist first, and read what public bodies need to check before switching on AWS Quick for the reasoning behind each decision.

If you would like help preparing a record like this, see AWS Quick Readiness, Implementation & Assurance.