Use case — Knowledge & Research
Make approved knowledge easier to find, verify, and use.
Internal knowledge is spread across drives, wikis, and inboxes with no clear owner or version. Retrieval only works when the content set is curated, permissioned, and dated.
Controls in this workflow: source citation, freshness metadata, permission mirroring, an explicit 'not found' behaviour, human review, user feedback, and a defined removal path for retired content.
The outcome
People get answers they can check, from content they are permitted to see, with a visible source and date — rather than a confident paragraph from nowhere.
The friction today
What teams describe
- The same question is answered differently depending on who is asked
- Documents have no owner, version, or review date
- Search returns file names, not answers
- Experts spend hours re-explaining settled material
Current state, and the workflow afterwards
Current-state workflow
- SearchAnswers live across drives, wikis, and inboxes, in several versions.
- AskPeople interrupt colleagues because search does not surface the current version.
- ReuseDocuments are rebuilt from the last similar one, carrying old claims forward.
- MaintainNobody knows which content is still approved, or who owns it.
Future-state workflow
- CurateA defined content set is agreed with named owners and review dates; everything else stays out of scope.
- PermissionAccess rules are mirrored at retrieval time so nobody sees content they could not open directly.
- AnswerQuestions return a short answer with citations to the passage and its review date.
- RefuseWhen approved content does not cover the question, the system says so and routes to a person.
- MaintainUnanswered questions become the content backlog, reviewed with owners on a schedule.
Example workflows
- Access-aware knowledge assistant
- Answers drawn only from content the person is already permitted to see, with citations and dates.
- Research synthesis with source links
- A synthesis across permitted material where every claim links back to its source.
- Policy and procedure retrieval
- The governing clause returned directly, with its version and effective date.
- Proposal and document drafting from approved content
- First drafts assembled from approved building blocks for author review.
- Expert-finding and content routing
- Pointing a question at the right owner when the answer is not documented.
- Content gap and freshness reporting
- Surfacing unanswered questions and stale content so owners can act.
What AI assists, and what people decide
What AI assists
- Retrieving and citing approved content within existing permissions
- Synthesising across permitted sources with links preserved
- Drafting from approved building blocks
- Routing questions to the documented owner
What people decide
- Whether the cited source actually answers the question
- What is approved, current, and citable
- Whether a draft is fit to send
- When content should be corrected or removed
Approved data and systems required
Systems involved
- Document storage and intranet
- Identity provider and group membership
- Content ownership register
- The interface people already work in
Approved data inputs
- Approved policies, procedures, and reference material
- Ownership and review metadata for each document
- Question logs, used for gap analysis rather than profiling
Controls and failure modes
Human-control pattern
- Only curated, owned content is indexed
- Retrieval enforces the same permissions as the source system
- Every answer is citable and shows how current the source is
- An explicit 'not covered' response instead of a plausible guess
Failure modes
- Over-broad indexing exposing content beyond intended audiences
- Stale content answering confidently — mitigated with review dates and owners
- Adoption failure if the answer lives outside the tools people use
What this does not promise
- No answers from content nobody owns or has reviewed
- No claim to replace expert judgement on novel questions
- No indexing of restricted material to improve coverage
Implementation path
- 1. InventoryAgree the corpus, the owners, and what is explicitly out of scope.
- 2. Permission-mapMirror existing access rules; retrieval never widens them.
- 3. Ground and citeCitation and freshness metadata are required on every answer.
- 4. Define 'not found'The system says it does not know and routes onward rather than guessing.
- 5. MaintainFeedback, review cadence, and removal are owned by named people.
Measures
These are evaluation targets you agree and track together, not promised results. Baselines are measured before launch so any change is attributable, and figures stay specific to your data, volumes, and process.
What you can measure
- Answer coverage against real questions asked
- Citation accuracy in sampled review
- Share of answers marked helpful by the asker
- Time to find a current, approved answer
- Volume of stale or removed content over time
What each stakeholder needs to know
- Executive sponsor
Why does this need governance?
The value comes from trust in the answer. Curation, ownership, and review dates are what make the answer usable, and they need an owner in the business.
- Functional leader
What changes for the team?
Repeat questions stop consuming expert time, and content gaps become visible instead of anecdotal.
- Operations leader
What is the ongoing work?
A recurring content review against the gap backlog, with named owners per content area.
- Technology & security reviewer
What does this touch, and how is it controlled?
Permission-mirrored retrieval evaluated per query, index scoped to the agreed content set, no restricted repositories, and logging of query and retrieved passages for audit and evaluation.
- End user
How do I know it is right?
Open the citation. If the answer is wrong or the source is out of date, flag it — that flag goes to the content owner.
The right AI solution is not selected by trend. It is designed around the workflow, approved data, systems, people, risk, and measurable outcome.
Find the AI opportunities worth implementing in your business.
Start with a structured assessment of your workflows, systems, and data, and leave with a prioritized view of where AI can create real value.