Contact nameInquiryPrivate staff records Excluded
What can it read?
Choose the records and details each task needs. Leave the rest out.
SECURITY, FROM THE START
AI built for your business.
Boundaries set by you.
Choose what your AI can read, what it can do, and when it must ask your team.
We agree the rules, AI services, and responsibilities for your specific solution.
Customer records
Approved messages
Selected files
Your AI assistantRead · Understand · Draft
Only the action you allow
Information you left out? No connection.
Example only. Rules are agreed for your setup.
See how TAG builds AI with you in control.
1:27 walkthrough · Captions includedRead the video overview
Agree what information the AI can use and which services can receive it. Test what happens when information is missing, permission is denied, or a service is unavailable.
Review the recipient, message, and facts before approving an action that needs a person’s decision. Other routine tasks can run within rules agreed in advance.
After launch, review activity, pause tasks when needed, and control access. The video uses fictional records to illustrate the process.
FROM FIRST CONVERSATION TO DAILY USE
How we set up
your AI.
Start with one useful task. Agree the limits. Check it works before your team relies on it.
- 01
Learn how you work.
Start with your team and examples you approve, not access to everything.
A map of the work
Your teamYour processOne useful task - 02
Agree the rules.
Choose the information, AI services, and people involved.
An agreed access plan
Contact detailsCustomer inquiriesNot payroll - 03
Connect what is needed.
Give access only to what the task needs. Allow reading without changes where possible.
Limited access
Selected recordsRead onlyAI assistant - 04
Test before real work.
Use approved test data to check answers, mistakes, and requests the system must refuse.
A test record
Normal requestMissing permissionService offline - 05
Start small.
Approve a limited set of tasks and people before expanding.
A launch checklist
PeopleAllowed actionsFirst tasks - 06
Keep your team in control.
Name who reviews activity, handles issues, and can turn off access.
An owner + review plan
Review activityPause actionsRemove access
BEFORE WE CONNECT
Know what is shared.
And who has access.
Four decisions to make before your AI uses business information.
Your systemsAgreed AI serviceWhat is shared? What is saved?
Where does it go?
Agree which AI services receive information, what they may use it for, and what they keep.
AI suggestsYour reviewer decidesSoftware checks permission
Who has control?
Name who can read information, approve actions, and change the rules.
SuggestedReviewedResult recordedPause future actions
How do you check it?
Agree what activity to record, who reviews it, and how to stop future actions.
BEFORE AI ACTS
Choose what it can do.
And when it must ask.
Some tasks only prepare an answer. Others can run automatically. Important decisions can wait for your team.
Read & draft
Selected informationAn answer or draft
Looking up a record does not allow the AI to change it.
Run an agreed task
Rules you approveAn allowed action
Routine tasks can run automatically within the limits you set.
Ask before acting
A suggested actionYour team decides
Actions that need approval wait for the person you choose.
Try reviewing a reply.
The AI has drafted a follow-up for Casey. Edit it, review the message, then approve or decline.
INTERACTIVE EXAMPLEFictional customer. No real sending.
SUGGESTED ACTION
Reply to Casey.
- To
- Casey · casey@example.com
- Action
- Send one follow-up email
Edits require a fresh review. This example stays in your browser and resets when you leave.
This is a practice example. In a real setup, the connected software checks permission before an action can run.
CHECK BEFORE LAUNCH. REVIEW AFTER.
Test the mistakes.
Plan the response.
Check what happens when information is missing, permission is denied, or a tool stops working.
Example checks for your project, not completed test results.
Discuss a test planBEFORE LAUNCHExample test plan
Test this situationExpected response
- Send without approvalWait for review
- Read a private recordBlock access
- Answer without a sourceAsk a person
- Repeat the same sendPrevent duplicates
- Use access you turned offStop the action
- Continue when a tool is offlinePause and notify
Know who to callName the person responsible for questions and issues.
Check changesRetest when information, tools, or permissions change.
Be able to stopPause future actions and turn off access.
No system is risk-free. Keep reviewing how it works. Pausing does not undo actions already completed.
YOUR QUESTIONS
What happens
to your information?
Here are the details to agree with your team before starting.
Read our website privacy policyDoes our information stay in our own systems?
Not always. An AI service may need selected information to complete a task. Before connecting it, we agree which information it can receive, where it is used, and whether copies are saved. Sending it somewhere else needs your review and approval.
Can an AI service use our data to train its models?
Using information to answer a question is different from using it to train AI. We review each service’s terms and settings, including training use, where it handles data, and how long it keeps copies. The agreement must fit your setup; a general promise about AI is not enough.
Who gets access, and can we turn it off?
The plan names the systems, records, tasks, and people involved. Where supported, your business controls the accounts and can remove access. We agree how to disconnect tools, remove staff access, and keep your records separate from other clients’ records.
How do you protect sensitive information?
Leave out information the task does not need. Use sample records or hide personal details for testing where practical. Check access, storage, backups, and encryption, which protects data while it is sent and stored, on the services chosen for your business.
What if a message tells the AI to ignore our rules?
A customer message or file is information to read, not permission to act. The connected software must check what is allowed. Test attempts to reach private records or trigger unapproved actions. When a simple rule can do the job, there is no need to ask AI to decide.
How do we check an answer or see what happened?
Plan for answers that show the records they used. Agree what to record about suggested actions, who approved them, and what actually happened. Decide who can see those records and how long they are kept. Activity records and backups need protection too.
What if something goes wrong or our setup changes?
Agree who to contact, how to report an issue, and how to pause affected tasks. Investigate, fix, and retest before restarting. Review changes to information, tools, and permissions. Pausing stops future work; it does not take back an email already sent.
What happens when we stop using the solution?
Plan this before launch: turn off access, stop scheduled tasks, return the agreed records, and delete copies according to your agreement. Include AI services, activity records, and backups. Some copies may need to be kept for a set time or deleted on a different schedule.
PLAN YOUR FIRST STEP
What would you improve?
What needs protecting?
Bring your questions or your IT team. We can plan the right setup together.