KINTO gave a real estate firm locked inside a sector ERP an agent layer that reads the ERP, shows lease by lease the status of each document, and writes only when a person approves.
Problem
The whole operation (leases, invoicing, treasury, maintenance) lived in a property-management ERP with thousands of undocumented API services, the old website was maintained apart from the ERP, and WhatsApp support depended on the office.
Signals
- Market: It takes listings from across the country, but its channel only offered its home city and lost people asking about other cities.
- Business: The old site and the ERP did not match.
- User experience: Management needed to ask the ERP in plain language instead of navigating menus.
- Support: Some conversations in the last year ended because the inventory had nothing that matched.
How we understood it
- ERP inventory before any automation: every API was cataloged and the real authentication was debugged error by error.
- Read-only audits: leases and properties in the ERP were cross-checked against documents kept across email, shared folders and scans (with OCR), and against the old site’s catalog. Result: a verifiable inventory of leases, properties and their documents.
- WhatsApp history mining: More than four years of conversations classified with a closed taxonomy, voice notes transcribed. This turned what the team knew into verifiable facts.
- Measure before fixing: every dashboard improvement was first reproduced with automated browser tests.
What we built and why
- An agent layer on the ERP: a command-line tool built for agents, read-only by default, with curated operations and error messages that suggest the next step. The ERP stays the source of truth.
- Close-the-books dashboard: shows lease by lease the status of each document; ERP writes go through a queue a person approves, with dry run, read-back and duplicate protection.
- Internal conversational assistant with per-person memory and explicit permission for any write.
- WhatsApp agent that shares listings with photos and the official ERP price (synced daily), takes damage reports, schedules visits and escalates to an adviser, with a health monitor that alerts when the channel cannot operate even if everything looks green. The official price “is law.”
- New listings site fed by the ERP, with a map per property, a WhatsApp button prefilled with the property, redirects driven by Search Console data, and a guard that blocks publishing a catalog without photos.
- Old-site rescue: verified backup and a human review tool to decide what to keep from the old site.
Why this way: In real estate the costly mistakes are writing wrongly to the books or promising a wrong price, so every write goes through a person and everything the agent states comes from authoritative data.
Transformation
-
The close runs through a dashboard with an approval queue, with the ERP and the ledger matching document by document.
-
The listings site is rebuilt from the ERP on every build, so what is published comes from the source of truth.
-
The WhatsApp agent reached production with the real inventory and produced two by-products: a list of neighborhoods with unmet demand and a catalog data-quality report.
-
ERP endpoints inventoried and curated for agents: thousands (measured)
-
rent invoices reconciled, ERP vs. ledger: dozens, 100% matching (measured)
-
WhatsApp history analyzed: over a hundred thousand messages, more than four years (measured)
What we don’t know yet
- Commercial impact of the agent (visits, conversion): there was no sustained traffic to measure yet.
- Accounting hours saved: not yet timed.
