01 the business
Engineering and surveying firm
180 staff across five offices
Staff ask the same questions daily: where the latest rate card is, how to set up a site visit, which template to use. The answers sit in SharePoint, Google Drive or a senior colleague's head. Experienced people are interrupted all day, and new starters guess.
02 before
The problem
Staff ask the same questions every day: where the latest price list is, how to handle a refund, what the process is for a site visit. The answers sit somewhere in SharePoint, Google Drive or a senior colleague's head, so experienced people get interrupted all day and new starters guess.
- Senior staff answer the same questions repeatedly.
- New starters take months to find their way around.
- Out-of-date documents are used by mistake.
03 the build
What the AI does
- Step 1: Connects to the places your knowledge lives, such as SharePoint, Google Drive, Confluence, Notion and your help centre, and keeps their existing permissions.
- Step 2: Indexes the content and refreshes it as documents change, so answers reflect the current version.
- Step 3: Answers questions in plain language inside Slack, Microsoft Teams or a web page, citing the documents it used.
- Step 4: Says clearly when it cannot find an answer and points to the right person or team to ask.
- Step 5: Logs unanswered and frequent questions so owners can see which documents are missing or out of date.
paused: waiting for a person to approve
How it was put together
- Connected SharePoint, Confluence and the help centre, keeping existing permissions.
- Indexed content and refreshed it as documents changed.
- Answered in Teams, citing the documents used.
- Logged unanswered and frequent questions for knowledge owners.
04 human in the loop
Where a person stays in charge
A named owner for each knowledge area approves which sources the assistant may use and reviews the gap report each month. The assistant only answers questions; it does not change records or send anything outside the company.
- Each knowledge area has an owner who approves sources.
- The assistant only answers; it never changes records or sends anything outside.
- Staff only see answers from documents they can already open.
05 rollout
From kickoff to live in 5 weeks
- Weeks 1–2Choose sources and owners; retire outdated documents.
- Week 3Pilot with one office.
- Weeks 4–5Roll out firm-wide with a monthly gap review.
06 after
What changes
People find answers without waiting for a colleague, and the experienced staff who used to field every question get their time back. Gaps in your documentation become visible and get fixed.
Assume 180 staff each spending 30 minutes a week looking for information or asking colleagues, and the assistant saving a third of that.
180 × 10 minutes = 1,800 minutes, or 30 hours a week returned to billable and project work.
What to measure
- Questions answered with a citation.
- Interruptions to senior staff.
- Gap-report items fixed each month.
- New starter time to productivity.
07 evidence
What the research says
In a survey of 31,000 people in 31 markets, 75% of knowledge workers said they use AI at work, and 78% of AI users bring their own AI tools.
Knowledge workers spend roughly a fifth of their working week searching for and gathering information.
McKinsey Global Institute, Capturing business value with social technologies (opens in a new tab) · 2012
Across 5,179 customer support agents, an AI assistant raised issues resolved per hour by 14% on average and by 34% for novice and lower-skilled agents.
NBER, Generative AI at Work (Brynjolfsson, Li and Raymond) (opens in a new tab) · 2023
08 lessons
Mistakes to avoid
- Indexing everything. Old files cause wrong answers.
- No owners. Gaps never get fixed.
- Ignoring permissions.
09 questions
Common questions
- Do staff already use AI?
- Microsoft and LinkedIn found 75% of knowledge workers use AI at work, and 78% of them bring their own tools. A company assistant gives them a safe option grounded in your documents.
- Is our data used to train models?
- We use business-grade AI services that do not train on your data, with accounts in your name.