Service
AI Copilots for Staff
How do we give staff an assistant that knows our business?
Short answer
Index your authorised internal material and give staff retrieval with citation — the source passage, the document and the date — rather than a summary they cannot verify. The value is not that it writes; it is that institutional knowledge stops depending on which colleague is available and how much they happen to remember.
Definitions
What this is, and what it is not.
It is
- Retrieval over your own authorised documents, policies and precedent
- Answers that carry their source so they can be checked
- Drafting from your material, for a person to review and own
- Permission-aware, respecting the access rules you already have
It is not
- A general assistant with your logo on it
- A system that answers from a model's general knowledge
- Something that should produce anything customer-facing unreviewed
Fit
When this is the right answer — and when it is not.
Both lists matter. A firm that never tells you no is not giving you information.
Right when
- Staff interrupt senior colleagues to ask questions the firm has answered before
- Institutional knowledge is concentrated in a few long-serving people
- Onboarding is slow because context lives in people rather than documents
- The same analysis gets redone because nobody can find the previous version
Wrong when
- Your documentation is badly out of date — fix the corpus first or it will confidently repeat stale policy
- The knowledge genuinely is not written down anywhere
What we do
In this order.
- 01
Decide the corpus deliberately
What is indexed is an explicit decision with a documented inclusion process, not everything that happens to sit on a shared drive.
- 02
Retrieval with citation, not generation
Every answer returns the source passage and where it came from. Where the corpus does not cover it, the system says so.
- 03
Permissions at the data layer
Access rules enforced where the data lives, including barriers between teams or matters — never by instructions in a prompt.
- 04
A gap report
Questions the corpus could not answer become a ranked list. That list is how documentation finally gets maintained.
Boundaries
What stays with a person.
- Anything customer-facing or externally binding is reviewed by a person before it leaves
- The system does not answer from general knowledge — if it is not in your material, it says so
- Confidentiality and information barriers are enforced at the data layer
- Your documents are excluded from model training in writing
Is our internal data safe?
The terms are specified in writing at design stage — exclusion from training, retention, and where processing happens. Where sensitivity warrants it, retrieval runs inside your own cloud tenancy.
What if our documentation is a mess?
Then the gap report is worth more than the assistant, at least initially. It tells you precisely which documents are missing or wrong, ranked by how often people needed them.
The broader capability
Ask Your Own Company a Question
Most mid-sized companies are data-rich and answer-poor.
Related services
You know AI matters and you do not yet know where it would apply here.
Find out whether this is actually your highest-value fix.
Eleven questions, scored in your browser, with the arithmetic shown. It may well point somewhere else entirely.
Eleven questions, scored in your browser