How Zrico AI is controlled
Zrico AI only takes the steps you allow, within money caps per step and per day, at the properties you choose. Anything over a limit or with legal weight waits in Zrico approvals, every step is logged, and one switch stops everything.
These are the controls as they work in the app today, with the labels your team will see.
What can Zrico do without asking?
Each area is set to Off, Assist or Autopilot: Leasing, Collections, Maintenance, Bookkeeping, Owner reporting, Renewals, and Resident messages and calls. On Assist, every step waits for a person. On Autopilot, Zrico takes only the steps you ticked for that area, and only inside your limits.
- Hours: always, only while the office is open, only while it's closed, or custom days and times.
- Properties: every property, or only the ones you list. If Zrico can't tell which property a step is for, it waits.
- A daily cap on how many steps it may take in each area, so nothing runs away.
How do approvals work?
Anything waiting for a person goes to Zrico approvals, with Waiting, Decided and Activity log tabs. Each card says what Zrico wants to do, why, and its risk: Low risk, Contacts someone, or Money or legal. The card reads "Nothing happens until you approve". Approving runs it once, as you, with your permissions.
How do money limits work?
Money steps, like entering a bill, offering a payment plan or recording a bank line, wait for approval unless you allow them in that area. If you do, you set a cap per step and a cap per day. A step over either cap waits, and so does any step whose amount Zrico can't tell.
Which steps always wait for a person?
Legal steps are off unless you allow them, such as opening a collections case with a late notice for staff to serve. Some steps wait for a person whatever the settings say.
- Giving a job to a vendor, which emails the vendor and commits a cost.
- Written replies to vendors, and adding a job to a vendor's visit already booked.
- Sending an account to a collections agency.
- Messages to many tenants at once.
- Renewal offers.
Whose permissions does Zrico use?
Zrico works with permissions, never around them. When you approve a step it runs as you, with your role's permissions. On Autopilot it acts as the staff member an admin names for that area, and a step that person's role can't take waits for approval instead.
What does the log show?
Zrico keeps a separate log of everything it did, next to the append-only audit log of staff actions. The AI log records every chat turn, every tool Zrico called with what went in and what came out, and every action asked for, approved, rejected or run. The Autopilot activity log shows each step with its outcome, Did it, Asked a person or Didn't work, and the reason.
How do I stop everything?
Stop everything is one switch. Every agent stops at once and nothing runs on its own until an admin turns each area back on, one by one. Each area also has its own kill switch.
Can I try an agent before it sends on its own?
Yes. Shadow mode lets an agent draft everything as usual while a person approves every message and change before it goes. Zrico scores each draft against what your staff decided, so you can see how often it was approved as written before you let it send on its own.
How does Zrico check its own work?
Checks run in plain code before anything goes out, not in the AI's judgment.
- Facts: every amount, date, time, link and phone number in a reply must come from your records, or it waits for a person.
- Fair housing: leasing replies pass a fair-housing check before they're sent. The leasing agent never sees a lead's name, email or phone. This supports your team's own compliance work; it is not legal advice.
- Emergencies: gas, carbon monoxide, fire, flooding, sewage and no heat in the heating season are caught by fixed rules in English and Spanish, even outside the agent's hours.
- Outside text: anything tenants, vendors, emails or documents say is treated as information, never as instructions.
What data does the AI see, and is it used for training?
Social Security and tax ID numbers, card numbers, bank account and routing numbers, and keys never reach the AI model: those fields are dropped and numbers with those shapes are masked. Zrico's AI provider states: "Anthropic doesn't train its models on API data by default." A monthly AI budget is a hard cap, checked before each call.
Do residents know they're talking to AI?
Yes. Resident message answers say they're from the AI, and a resident can write "person" at any time. On the phone, the assistant tells callers it's an AI, that the call is transcribed, and how to reach a person.
Is Zrico AI on by default?
No. Autopilot levels are off by default, and agents such as the resident messenger and the owner messenger are off until an admin turns them on. New areas start on Assist, where a person approves every step.
Who can change the AI limits?
Admins set the levels, caps, hours, properties and who Autopilot acts as. Every change is recorded in the audit log with who made it and when.
What happens when Zrico isn't sure?
It asks a person. A bank match it's less than 90% sure of, a bill scan with warnings or a letter with a number that isn't in the statement always waits, whatever the settings say.
Can a tenant's message trick Zrico into doing something?
Text from tenants, applicants, vendors, emails and documents is marked as outside information, and Zrico is told never to follow instructions in it. The money caps, approval rules and permission checks run in code, so they apply whatever the AI reads.
Does Zrico train AI models on my data?
Zrico's AI provider states that Anthropic doesn't train its models on API data by default. Sensitive numbers like Social Security, bank account and card numbers are removed before anything reaches the model.
See Zrico run your portfolio.
Bring a property. We will show you, on your own data, what Zrico would have handled last month.