Tenant, property, unit, request, vendor, appointment, lease
loading…
Do this on paper or in a short document. The point is to make the work visible before reaching for architecture.
Initial request A regional property-management company wants an AI assistant that answers tenant questions, schedules maintenance, and tells staff which requests are urgent.
Write short answers to these questions:
Create four lists:
Tenant, property, unit, request, vendor, appointment, lease
New, triaged, scheduled, assigned, completed, escalated
Property system, calendar, SMS, email, vendor system
Tenant, coordinator, technician, manager, administrator
Consider at least one failure in each category:
A strong first version may answer approved policy questions, classify maintenance requests, detect emergency keywords, and create a draft work order for staff approval. It should not automatically dispatch vendors, interpret complex lease disputes, or close emergency cases without human review.
Create a table with workstream, range, assumption, and risk.
| Workstream | Your range | Assumption | Risk |
|---|---|---|---|
| Workflow discovery | _____ | Staff can demonstrate the current process | Different properties use different rules |
| Knowledge assistant | _____ | Policies are available digitally | Documents are stale or inconsistent |
| Maintenance classification | _____ | Historical tickets exist | Emergency labels are unreliable |
| Work-order integration | _____ | API access is available | Vendor system limitations |
| Pilot and monitoring | _____ | One property can pilot first | Operational adoption is low |
1. Restate the outcome.
2. Name the recommended MVP.
3. Give the estimate range.
4. State the assumptions.
5. Name the top risks.
6. Explain the first technical spike.
7. Describe the rollout and success metric.Show a sample lead answer
“I would not begin with a fully autonomous maintenance agent. I would pilot a staff-assisted workflow for
one property: approved policy Q&A, request classification, emergency detection, and draft work-order creation.
For a small team, I would estimate six to nine weeks, assuming the property platform has usable APIs and
policies are centralized. The largest risks are emergency-classification quality, differences between properties,
and integration access. I would test the property API and build an emergency-case evaluation set in the first week.
The pilot would be successful if routine handling time falls while emergency recall remains above the agreed threshold.”
Take apart two products each week. Save the first scope and estimate, then revisit it after learning more about the domain. Over time, you should catch operational work, dependencies, and risky assumptions earlier.
Practice standard End-to-end judgment improves through repeated teardown, estimation, and comparison—not through memorizing one architecture.