AI & Data Solutions
RAG & Enterprise Knowledge Search
Ask a question in plain English and get an answer from your own documents, with the source alongside.
Discuss Your ProjectIn plain terms
Instead of searching a pile of files and reading five of them, you type a question and get an answer that tells you which document it came from. It works from your own policies, manuals, tickets and records, and not from the general internet, and it only shows people what they are allowed to see.
Does this sound familiar?
- Answers are scattered across shared drives, wikis, email threads and one person's head
- New starters keep asking the same experienced colleagues the same questions
- A general chatbot sounds certain and has never seen your policies
- Some documents are confidential and the assistant has to respect that
A typical scenario
An engineering firm with 300 staff
Before
The answer to "what is our rule on subcontractor insurance?" is in a policy on SharePoint, an older version on a shared drive and an email from the finance director. New project managers ask the same three colleagues, who are interrupted a dozen times a week.
After
A project manager types the question into Teams and gets the current rule in two sentences, with the policy name, section and date, and a link to open it. When there is no answer the assistant says so, and the question goes on a list for the policy owner.
How we deliver it
From first call to running
- 1
Gather the documents
We connect to where your documents live and see what is there: how many, how old, which ones contradict each other.
You get: A document audit, with a list of what to retire or fix first
- 2
Build a first assistant
We index one set of documents, such as HR policies, and build the question box.
You get: A working assistant on a limited set of documents
- 3
Test with real questions
Staff give us thirty to fifty real questions. We check every answer and fix the causes.
You get: An accuracy score and a list of fixes
- 4
Open it up
More users, more document sets and permissions switched on.
You get: A live assistant in Teams or on a web page
- 5
Keep it right
We review unanswered questions each month and retest after changes.
You get: A monthly quality report
What we need from you
- Access to the places your documents are kept
- Someone from each department to check answers
- Agreement on which documents are the authoritative ones
- Rules about who may see what
Where it works best
- A lot of documents that people struggle to search
- The same questions asked repeatedly
- Content that changes, so a fixed FAQ goes stale
Worth sorting out first
- Only a handful of documents: ordinary search may be enough until the library grows
- Documents that are out of date or contradict each other: agree which are authoritative first, and our audit helps with that
- Questions that need calculating across large data sets: pair it with our database tools
How we work with you
A fixed-price first assistant on one document set, then monthly support as you add more. The audit of your documents is useful on its own.
Discuss Your ProjectTypical technology
Questions
How do you stop it making things up?
It may only answer from what the search finds, it has to cite the source, and it is allowed to say it cannot find an answer. Then we test it against questions we already know the answers to.
Does it learn from our documents?
No. Documents are searched when a question is asked and the model itself is not retrained. A side effect is that removing a document removes it from the answers straight away.
How long before staff can use it?
A first version on one document set can usually be in front of testers within a few weeks. Most of that time goes on the documents, not the AI.
What happens to confidential documents?
The assistant only answers from documents the person asking could already open. That is built in from the start, and we test it with people from different teams.
Related services
Intelligent Document Processing
Invoices, forms and contracts read automatically, checked against your rules and passed to the next system.
Learn more →AI Quality, Security & Governance
Checking AI work is accurate, keeping data safe, and agreeing sensible rules for how AI is used.
Learn more →Data, Analytics & Machine Learning
Data you can trust, dashboards people actually use, and predictive models where they are worth building.
Learn more →