Onboard · Managers
How to roll out an AI assistant to your company
Rolling out an AI assistant means deploying a chat-based AI tool to employees with access to approved company knowledge, clear permissions, and rules for human review. A successful rollout starts with one team and a defined purpose, connects only the data each role should see, and expands in stages with training and support.
- Reading time
- 8 min
- Chapters
- 6
- Last reviewed
- November 10, 2026
- Author
- Lingows AI Team
- Difficulty
- Intermediate
Before you start
- An AI use policy, or agreement to write one before launch
- An owner for the assistant with time to support the pilot
- A list of the documents and systems the assistant may use
The steps at a glance
- Step 01
Define the purpose
Write the purpose, owner, first user group, and measures.
- Step 02
Choose the assistant type
Decide between a general business assistant, a custom assistant, or both.
- Step 03
Connect approved knowledge
Clean and connect a small set of current sources with access rules that follow the user.
- Step 04
Set rules and review steps
Define permitted tasks, restricted data, human review, and how to flag bad answers.
- Step 05
Pilot with one team
Launch with training, a support channel, and close monitoring.
- Step 06
Expand in waves and operate
Add teams in stages, check permissions each wave, and review quality on a schedule.
Chapter 01
What should an AI assistant do for the company?
An AI assistant should have a defined purpose before rollout, such as answering staff questions from internal policies, drafting customer replies, or summarizing project records. A clear purpose decides which data the assistant connects to, who gets access, and how success is measured. A vague purpose produces a tool people try once and abandon.
Giving everyone AI is not a purpose. Helping support staff answer product questions from the documentation, with sources, is a purpose. The narrower statement tells the team which documents matter, which users to start with, what a good answer looks like, and what to measure.
Write the purpose down with the owner, the first user group, and two or three measures. Revisit the statement before each stage of the rollout, because the purpose often sharpens once real users start asking real questions.
Chapter 02
How do you choose between a general assistant and a custom one?
Choosing between a general AI assistant and a custom one depends on the data, the controls, and the workflow. General business assistants from major providers suit drafting, research, and document work. A custom assistant suits cases that need connections to internal systems, role-based access to company records, audit logs, or specific actions inside business software.
General versus custom assistant
| Consideration | General business assistant | Custom assistant |
|---|---|---|
| Setup | Accounts, admin settings, and policy | Design, build, connections, and testing |
| Company knowledge | Uploaded files or supported connectors | Retrieval from approved sources with access rules |
| Actions in business systems | Limited to supported integrations | Defined actions through APIs or MCP servers |
| Access control | Workspace and admin settings | Rules matched to the company's own roles |
| Audit and logging | Provider admin features | Logs designed for the company's review needs |
| Best fit | Drafting, research, and individual productivity | Workflows tied to company data and systems |
Many companies use both: a general assistant for everyday work and a custom assistant for one or two workflows that need company data and controls. Models from OpenAI, Anthropic, Google, and other providers can sit behind either option. The AI integration and custom AI chatbots pages cover the custom route in more depth.
Chapter 03
How do you connect an AI assistant to company knowledge safely?
Connecting an AI assistant to company knowledge safely means choosing approved sources, cleaning outdated documents, and enforcing the same access rules people already follow. An assistant should only retrieve documents a user is allowed to open. Retrieval with source citations lets staff check answers against the original document instead of trusting the assistant blindly.
Start with a small, high-quality set of sources: current policies, product documentation, and approved templates. Outdated or contradictory documents produce outdated or contradictory answers, so remove or archive stale material before connecting a source.
Most company assistants use retrieval augmented generation, which finds relevant passages in approved documents and gives those passages to the model along with the question. The RAG answer explains the approach. For assistants that take actions in business systems, the Model Context Protocol is one standard way to expose tools with controlled access.
Chapter 04
What rules should govern an AI assistant at work?
Rules for an AI assistant at work should state which tasks the assistant may help with, which data stays out, when a person must review output before use, how to report a wrong or harmful answer, and who owns the assistant. Rules belong in the AI use policy and in short guidance shown inside the assistant.
Human review is the most important rule. Anything sent to a customer, filed with a regulator, or used to make a decision about a person should pass through a staff member who checks the content. The human in the loop answer describes common review patterns.
Make reporting easy. A simple way to flag a bad answer, with the question and the response attached, gives the owner the examples needed to fix sources, adjust instructions, or change settings. The guide on training a team to use AI covers the policy and verification habits staff need before launch.
Chapter 05
How should an AI assistant rollout be staged?
An AI assistant rollout should be staged: a pilot with one team and a defined purpose, a review of usage and answer quality, fixes to sources and instructions, and then expansion to more teams in waves. Each wave gets training, a support channel, and a check that permissions match the new users' roles.
Rollout stages
- 01PilotOne team, one purpose, a small set of approved sources, and close support from the owner.
- 02ReviewRead a sample of real conversations, collect flagged answers, and compare results with the measures set at the start.
- 03FixUpdate sources, instructions, and settings based on what the review found.
- 04Expand in wavesAdd teams one or two at a time, with role-based training and a permissions check for each wave.
- 05OperateMonitor usage and quality, review sources on a schedule, and keep a backlog of improvements.
Checklist
AI assistant launch checklist
Chapter 06
How do you measure an AI assistant after launch?
Measuring an AI assistant after launch combines usage, quality, and outcome signals: weekly active users, the share of answers flagged as wrong, the tasks people use the assistant for, and changes in the measures set at the start, such as response time or time spent searching. Review the signals with the owner on a regular schedule.
Usage alone can mislead. High usage with frequent flags means people want help but the sources or instructions need work. Low usage with good answers usually means the purpose is unclear or training did not reach the right people.
Plan for ongoing operation from the start. Sources go stale, models change, and new teams bring new questions. Lingows AI supports this stage through AI onboarding and managed AI operations, and builds custom assistants on SpinFlow when the work needs company data and controls.
01Summary
Key takeaways
What to remember
- Give the AI assistant a written purpose before choosing a product.
- Use a general assistant for everyday work and a custom one where company data and controls matter.
- Connect a small set of clean, current sources first.
- Make access rules follow the user, and test with different roles.
- Pilot with one team, then expand in waves with training and support.
- Measure usage, flagged answers, and outcomes, and plan for ongoing operation.
02FAQ
Questions about this guide
What is the difference between an AI assistant and an AI agent?
An AI assistant responds to a person's requests in a conversation, such as answering a question or drafting a document. An AI agent pursues a goal across several steps and can take actions in connected systems with less direction. Many company rollouts start with assistants and add agent capabilities for specific workflows later.
Can an AI assistant use internal documents without exposing the documents?
An AI assistant can use internal documents without exposing the documents to unauthorized staff when retrieval respects existing access rules and the provider terms prohibit training on company data. Companies should confirm data handling terms in the provider agreement, test access with different roles, and keep restricted material out of connected sources.
How many employees should be in an AI assistant pilot?
An AI assistant pilot should include enough employees to produce varied, realistic questions and few enough for the owner to support closely. One team with a shared purpose usually fits. The pilot group should include enthusiasts and skeptics, because skeptics often surface the problems that matter most before a wider rollout.
What happens when an AI assistant gives a wrong answer?
When an AI assistant gives a wrong answer, the user should flag the response, and the owner should find the cause: a missing or outdated source, an unclear instruction, or a task the assistant should not handle. Human review rules keep wrong answers from reaching customers while the owner fixes the underlying issue.
Do we need a custom AI assistant if we already pay for a general one?
A company with a general AI assistant needs a custom assistant only when a workflow requires connections to internal systems, role-based access to company records, specific actions in business software, or logging the general product cannot provide. Many companies keep the general assistant for everyday work and add one custom assistant for a priority workflow.
Written by
Lingows AI Team
Engineering and strategy team, Lingows AI
The Lingows AI team has built AI systems since 2017 and software since 2012. Collective posts cover model landscape changes, tooling selection, and the direction of applied AI.
Related software
SpinFlow
Enterprise AI platform with 17 frontier models and the business apps to put them to work.
Put this guide to work with AI Onboarding.
AI onboarding puts the system described here into daily operations, from pilot through adoption, with human review built in.