# Internal Knowledge Assistant: done-for-you AI automation by BufferSol

> A done-for-you internal knowledge assistant is a private AI agent that answers employee questions from your own Confluence, SharePoint, Google Drive, Notion and ticket history. It searches only what each person is allowed to see, gives a short answer with links to the source pages, and says clearly when it doesn't know. It also flags outdated or conflicting pages to their owners. BufferSol builds, connects and runs it in your tenant, with every question logged.

- Canonical: https://buffersoltech.com/done-for-you/automations/internal-knowledge-assistant
- Industries: Healthcare; SaaS & Technology
- Provider: BufferSol Technologies (New York; serving the US, UK, Ireland, Canada and the EU)
- Timeline: live in 3 to 6 weeks
- Pricing: quoted after a short scoping call

## Staff ask in plain English and get the answer from your own docs, with the source

**Who it is for:** Companies of fifty to several thousand people whose know-how is spread across wikis, shared drives, chat threads and tickets, where new and experienced staff alike lose time hunting for answers or asking the one person who knows.

### Signs you need it
- The same questions appear in Slack channels every week.
- Wiki search only works if you already know the page title.
- There are three versions of the same process doc and nobody knows which is current.
- New hires depend on a few long-tenured people for answers.
- Knowledge leaves when an expert leaves.

## Business outcomes
- **Answers in the flow of work.** People ask in Teams or Slack and get a short answer with links, instead of searching several tools or interrupting a colleague.
- **Experts interrupted less.** Senior staff stop answering the same questions in chat and spend that time on the work only they can do.
- **Docs that stay current.** Questions the assistant can't answer and pages that contradict each other go to the page owner, so the knowledge base gets better with use.

## How it works today
1. An employee needs the steps for a customer refund exception or a deployment rollback.
2. They search Confluence, then SharePoint, then Slack, and find several pages with different dates.
3. They message a colleague who knows, who stops what they are doing to answer.
4. The answer lives in that chat thread and is lost a week later.
5. Outdated pages stay up because nobody owns them.

Hidden costs:
- Hours each week lost to searching and waiting for replies.
- Mistakes when people follow an old version of a process.
- Constant interruptions for the few experts everyone relies on.

## How it works done for you
1. BufferSol connects Confluence, SharePoint, Google Drive, Notion and resolved tickets to a private index in your cloud, and keeps it in sync as pages change.
2. Each document keeps its original permissions. When someone asks a question, the assistant checks their identity in Entra ID, Okta or Google and searches only what they can open.
3. It reads the most relevant passages, writes a short answer, and links the exact pages and sections it used, with the last-updated date.
4. If sources conflict, it shows both and names the owners. If nothing covers the question, it says so and suggests the right channel or person.
5. Unanswered questions, thumbs-down answers and stale pages go into a weekly report for page owners, who fix the source rather than the bot.
6. Every question, answer and source is logged for review, and sensitive spaces such as HR cases or legal matters can be excluded entirely.

## What changes for the team
- **Subject-matter expert.** Today: Answers the same questions in chat and direct messages many times a week. Done for you: Fixes the source page once when the assistant flags a gap, and is interrupted far less.
- **New hire.** Today: Hunts across tools and is unsure who to ask or which doc is current. Done for you: Asks the assistant and gets an answer with links to read further.
- **Knowledge or operations lead.** Today: Has no view of what people can't find or which pages are stale. Done for you: Uses the weekly gap report to assign owners and retire outdated pages.

## KPIs we measure
- **Questions answered with a source:** Share of questions where the assistant gave a sourced answer and the user did not ask again.
- **Answer accuracy:** Page owners score a weekly sample of answers in their area.
- **Repeat questions in chat:** Count of common questions in key Slack or Teams channels, before and after launch.
- **Stale pages fixed:** Pages flagged as outdated or conflicting that owners updated or archived each month.
- **New-hire time to first contribution:** Manager-reported time until a new hire works without daily help, compared with recent cohorts.

## Industry data
- a quarter of the workweek: spent searching for information by executives and teams (Atlassian survey, 2025) ([UNLEASH](https://www.unleash.ai/artificial-intelligence/atlassian-ai-could-help-workers-stop-wasting-2-4-billion-hours-searching-for-information-every-year/))
- nine hours each week: average time UK workers spend locating the information they need to do their jobs (Atlassian research, 2025) ([IT Brief UK](https://itbrief.co.uk/story/uk-employees-lose-nine-hours-a-week-seeking-information))

## Systems it connects to
Confluence, SharePoint, Google Drive, Notion, Slack, Microsoft Teams, Jira, Zendesk, Microsoft Entra ID, Okta

## Guardrails
- Every answer respects the source system's permissions; people only get answers from documents they could open themselves.
- Answers always cite their sources, and the assistant says when it doesn't know instead of guessing.
- Your data stays in your cloud tenant and is not used to train public models.
- Admins can exclude spaces, folders or labels, such as HR cases, legal matters or board material.
- Questions and answers are logged with retention you set, and can be reviewed or deleted.

## What teams say about this work
- Knowledge workers and executives alike say they spend about a quarter of their workweek searching for information. ([UNLEASH](https://www.unleash.ai/artificial-intelligence/atlassian-ai-could-help-workers-stop-wasting-2-4-billion-hours-searching-for-information-every-year/))
- UK workers report losing more than a full working day each week tracking down the information they need, with about half the week going to coordination such as sharing details and chasing colleagues. ([IT Brief UK](https://itbrief.co.uk/story/uk-employees-lose-nine-hours-a-week-seeking-information))
- Employees depend on people instead of self-serve resources for answers, and forget what they were taught by the time they actually need it. ([HCM Tech Advisory](https://hcmtechadvisory.beehiiv.com/p/is-your-benefits-team-becoming-the-help-desk))

## Delivery: live in 3 to 6 weeks
1. Week 1: discovery, access and a baseline of today's numbers.
2. Weeks 2–3: build the agents and connect your systems, with your rules and approvals.
3. Weeks 3–5: test on your real history, then pilot on live work with your team reviewing.
4. Weeks 4–6: staged go-live, training, dashboards and handover.

## Executive Ask questions for this workflow
- What are people asking the knowledge assistant most, and which questions can't it answer?
- Which teams have the most outdated or conflicting documentation?
- What is our current process for approving a refund over the standard limit, and who owns that page?

## Frequently asked questions
### Can people see documents they're not supposed to?
No. The assistant checks each user's identity and searches only documents that user can already open in Confluence, SharePoint, Drive or Notion. Permissions are synced from the source systems, so a change there applies to the assistant too. Admins can also exclude whole spaces or folders, such as HR cases or board papers, from the index entirely.

### What happens when our docs are wrong or out of date?
The assistant shows the last-updated date with every source, and when two pages disagree it shows both and names their owners. Thumbs-down answers, unanswered questions and stale pages go into a weekly report so owners can fix or archive the source. Over time the documentation improves because people can see where it fails.

### How long does it take to go live?
Most assistants are live in 3 to 6 weeks. We agree which sources to include and exclude, connect them with read-only access, and sync permissions. A pilot group asks real questions for a short period while page owners check answers. We fix gaps in the sources and the setup, then roll out in Slack or Teams to everyone.

### What does it cost?
It is quoted after a short scoping call. The main factors are the number of users, how many sources and documents we connect, permission complexity, and where you want the assistant available. You get a written quote with a fixed scope before any build starts, so you can approve the budget knowing what is included.

### Is our data used to train an AI model?
No. Documents are indexed in your own cloud tenant, and the models we use are called through business APIs that do not train on your data. Questions and answers are stored under the retention rules you set, and you can delete the index and logs at any time.