Once the information is in one place, the whole team can use it. The owner, the office, the field and the books ask different questions and should see different things. This stage covers who sees what, the habits that make it collaborative, and a one-page policy so everyone knows the rules.
What's inside
- An example of what each seat asks, sees and shouldn't see
- How to decide who sees what
- Habits that keep answers right for everyone
- A one-page AI use policy and a handoff prompt
Same information, different questions
An example. Yours will differ, so decide each one deliberately.
| Seat | Asks things like | Should see | Should not see |
|---|---|---|---|
| Owner | What's slipping this week? What did we promise this customer? | Everything | Nothing is off limits, but keep pay and HR out of shared answers |
| Office manager | What's our policy on this? Where is this job? | Policies, customer records, schedules, suppliers | Pay rates, HR notes |
| Field lead | What are the steps for this? What did we agree for this job? | How-tos, checklists, their own jobs | Margins, pay, the full customer list |
| Bookkeeper | Which invoices are overdue? Where does this bill go? | Invoices, bills, vendor terms | Customer conversations, HR notes |
| New hire | Where do I find this? How do we do that? | Handbook and how-tos | Everything else, for now |
Decide who sees what
- Define seats, not people. People change, seats don't.
- Mark every source with the seats that may see it. Default to the narrowest.
- Keep pay, HR, contracts and customers' personal details owner-only unless there's a reason.
- Where possible, reuse the sharing rules from where the documents already live, and add exceptions on top.
- Review it whenever someone changes role or leaves.
This only works if the rule is applied when information is retrieved, as in Stage 4. An assistant told to keep something quiet can still repeat it.
The build: Stage 4 →Habits that make it collaborative
- Fix the source, not the chat. When an answer is wrong, correct the note, or the next person gets the same wrong answer.
- Keep a shared prompt library: a document of questions and instructions that worked, grouped by seat.
- Log what it couldn't answer and review the log weekly.
- Name one owner for the assistant, who decides what goes in and who gets access.
- Treat every answer a customer will see as a draft until a person has read it.
A prompt for handoffs
Handoff summary
Using only our records, write a handoff for [job or customer] for the person taking over. Cover: what was promised, what's done, what's still open, who the customer prefers to hear from, and anything to be careful about. Name the sources you used. If something is missing, say what's missing instead of guessing.
Owner's pre-call briefing
I have a call with [customer] at [time]. From our records, give me: the history in five lines, anything open or overdue, what we last promised, and one question worth asking. Use only what's in our records and name your sources.
A one-page AI use policy
Short enough that people read it.
AI use policy template
WHAT IT'S FOR [the jobs we use it for] WHAT GOES IN [the kinds of information that are fine] WHAT NEVER GOES IN [passwords, card numbers, pay and HR details, anything confidential] WHO CAN SEE WHAT [the seats and what each one can ask about] BEFORE ANYTHING IS SENT A person reads it first when it reaches a customer, involves money, makes a promise or touches a sensitive topic. WHEN IT'S WRONG Tell [name]. We fix the note, not just the answer. WHO OWNS IT [name], reviewed every [month / quarter].
Roll it out
A 30-minute team session
- Show one real question and one real answer, with its sources.
- Have each seat try five questions from their own week.
- Collect where it failed. That's your first fix list.
- Walk through the use policy together.