A Central Florida industrial developer ran a multi-park portfolio the way most do: an Excel leasing tracker as the source of truth, broker updates arriving by email and text, and a leasing meeting spent reconciling versions. This is how that spreadsheet became a governed system of record in a weekend — and why the AI came second.
What was actually in the spreadsheet?
Everything, which is the point. The workbook — 11MB of it — held properties, buildings, bay-by-bay square footage, tenants, lease dates and rates, and the pipeline. It wasn't bad data; it was good data in a fragile container. So instead of asking the team to re-key their world into a CRM, the tracker itself seeded the system: every property, building, bay, tenant, and lease imported as live records.
What changed for the team in week one?
- The bay map replaced the grid. Vacancy, prospect, and leased states visible per building, with rollups computed from the bays — so the numbers can't disagree with the map.
- Self-service edits. The team adds buildings, renumbers bays, and updates leases directly; every change is attributed.
- Expirations became facts. Lease-end dates now surface on a dashboard feed, ranked by urgency, instead of living in someone's memory.
Where did the AI come in?
Only after the record existed — and only as a proposer:
- Broker packets: the biweekly xlsx/PDF reports get read by the system into suggested rows (prospects, tours, comps), staged in a review queue. A person maps, accepts, promotes. Processing an 11MB packet takes about a minute; nothing goes live without approval.
- Text-message capture: brokers text updates to a number; each text is parsed into the same review queue. ("Acme toured Building 3, needs 20k SF" becomes a structured prospect suggestion.)
- Quick notes: the same capture box sits on the dashboard for thoughts on the go.
One review queue, one rule — propose, never commit — across every channel.
Why does the order matter?
Because AI output needs somewhere structured to land. Firms that pilot AI without a system of record get summaries beside the workflow; firms that fix the record first get suggestions inside it. The weekend was spent on the boring part — the import, the data model, the review queue — and that's precisely why the impressive parts stuck.
The team still works the way it worked. The difference is that the tracker now has memory, concurrency, and a review queue — and the spreadsheet is retired with honor.