Compass CRM for Real Estate — built for agents, not generic deals

The problem: a real estate agent isn't a spreadsheet

A real estate agent's day isn't made of to-do lists — it's made of people and properties: leads from Meta ads and listing portals, viewings arranged over the phone, offers, and owners who want to know every week how the sale is going. Real estate agents typically use Pipedrive or another generic CRM. The basics work — but most of the real-estate-specific work happens outside the system: in calendars, in Gmail, in spreadsheets, and in the agent's head. The biggest pain points:

📥

Scattered leads

Leads from Meta Lead Ads, listing portals, and the website land in different places and have to be collected by hand.

🔥

Who is the "hot" buyer?

Nobody can see who has opened the price list several times or clicked to book a viewing. Call order comes down to gut feeling.

📅

Manual viewing scheduling

Times agreed by phone, calendar entries added by hand, no reminders — and every cancellation or reschedule means more manual work.

🏠

Owner reports

Sellers ask every week: how many viewings, how much interest? Putting the answer together takes half an hour each time.

💰

Commissions in a spreadsheet

Commissions on each listing (base, mortgage, legal) live in a separate spreadsheet, not in the CRM.

A generic CRM can't solve this, because it has no concept of a property, a listing agreement, or a viewing. You need a real estate data model.

Why wasn't Pipedrive enough?

Pipedrive is an excellent general-purpose sales tool. But a real estate agent doesn't manage "deals" — they manage properties, listings, and viewings, and the connections between them are what matter.

GENERIC CRM

Who is the customer?
What stage is the deal in?
When should I call?

REAL ESTATE OPERATING SYSTEM

Which property did they view, and what was their feedback?
Who is the most engaged buyer for this listing?
Did they accept the viewing invite?
Who needs to hear about it when the price drops?
How much commission is expected this quarter?

The goal wasn't a Pipedrive clone — it was a system built around how real estate agents actually work, while still covering every Pipedrive feature they use daily, so switching wouldn't feel like a step back.

Architecture decisions

Four decisions shaped the system: two separate pipelines, measurable buyer intent, humans making the calls, and zero new vendors.

🏠

Two pipelines: buyers and sellers

Leads track buyer and renter inquiries; Deals track seller listings — from the first visit to handing over the keys. Adding a new property automatically creates its listing. Contacts, viewings, and offers connect to both sides on a shared timeline.

Why? A real estate agent works in two directions at once. A generic CRM squeezes both into a single "deal" concept — here, each side follows its own logic.

🎯

Event-based lead scoring

Every buyer action earns points, and the 0–100 score places each lead in COLD, WARM, HOT, or VIP. At the VIP threshold the system assigns the lead and creates a call task automatically. Point values and thresholds are configurable in the UI.

Why? Call order is driven by real behavior, not gut feeling — the hottest buyers move to the top.

🤝

The system flags, a human decides

When a lead replies to an email, the system neither scores it nor acts on it — it surfaces the lead in a "Replied" block, where the agent decides the next step. Nothing gets archived or closed automatically.

Why? A reply could be "when can I see it?" or "thanks, not interested." A person reads that better than any automation.

🧩

No new vendors

For per-minute scheduling we didn't add a paid cron service or a pricier hosting plan — we used the pg_cron and pg_net extensions of the database we already had (Supabase).

Why? Zero ongoing cost, and the sending logic stays in one place — the Next.js app. The database only handles timing.

Pipelines and automations

The system follows every client from the first inquiry to the closed sale and the commission — and does the work for the agent wherever it can.

THE TWO PIPELINES AND SCORING

Leads — buyers and renters

leads

NewEmail sentActiveCall scheduledContactedQualifiedViewingOfferWon

Deals — seller listings

deals

Wants to sellAfter valuationOn the marketPurchase offerSale agreementHandover

Lead categories (0–100 points)

intent_score

COLD 0–29WARM 30+HOT 60+VIP 80+

AUTOMATIONS

Meta lead → CRM

Trigger:A Facebook/Meta lead form is submitted

Action:Lead + contact created and linked to the property, optional welcome email

Frequency:Instantly (webhook)

VIP threshold

Trigger:The score reaches 80

Action:Assigned to the lead agent + high-priority call task + notification + Telegram message

Frequency:Per event

"I'd like to see it"

Trigger:A click on the viewing request button

Action:Immediate call task + notification + Telegram, regardless of score

Frequency:Instantly

Activity spike

Trigger:3+ property-related clicks within 30 minutes

Action:Immediate call task — no need to wait for the score to cross the threshold

Frequency:Per event

Post-viewing follow-up

Trigger:A viewing is closed with positive interest

Action:Follow-up email using the default template

Frequency:When the viewing is closed

Engineering deep-dive

📈

Lead scoring

Price list opened 15, booking click 20, "I'd like to see it" 50, offer 35 points
Email opens on a graded curve: 1 → 2 → 4 points, then 0 from the 4th open — reloading a tracking pixel can't inflate the score
No negative points, and the stored score never decreases — but the dashboard ranking "cools" with time since the last activity (100% within an hour, 10% after 14 days)
Own and test contacts are automatically excluded from scoring and every automation

⏱

Precise email scheduler

pg_cron calls the scheduler endpoint every minute (via pg_net) — processing never depends on someone opening the CRM
Recipients are claimed with FOR UPDATE SKIP LOCKED: two parallel runs can never send twice
Suppression list (bounces, complaints, unsubscribes) and normalized-address deduplication before every send
Viewing reminders run on the same engine — there is no second scheduler

📅

Google Calendar integration

Two-way sync with Google Calendar: anything added in the CRM shows up on the agent's phone
Real calendar invites for clients, with the address and a Google Maps link
Rescheduling updates the existing event — duplicate entries are never created
The client's RSVP (accepted / declined) is checked every 15 minutes and shown in the CRM

📦

Pipedrive Smart Importer

Pipedrive exports land in staging tables — analysis and dry runs never touch live data
File type detection by a required set of headers: no guessing on partial matches
Dry run, reconciliation report, and owner-based filtering (whose clients get migrated)
Reversible imports and idempotent activity loading — re-running never duplicates

Quality and security

A CRM holds client data — so quality assurance was never an afterthought, but part of every iteration.

Data security

All 54 tables are protected by row-level security (RLS), isolated per organization
The importer and settings are available to the admin role only
The outgoing email suppression list automatically applies to every sending path

Automated tests

203 tests: 48 end-to-end (Playwright) and 31 unit test files
A mock Google Calendar server for testing invites, RSVPs, and reschedules
A regression run at the end of every iteration, before each deploy

Independent audits

A completeness audit of 71 features, with a prioritized gap list
Every "it works" claim was verified against the live database — claims we couldn't reproduce were removed from the reports
Mobile (390 px) and accessibility visual QA

Human approval

Owner reports never go out automatically — the agent reviews, adds a note, then sends
Bulk campaigns only start after a dry run and approval
A real Pipedrive import only runs with a dry run, small batches, and the option to roll back

Key learnings

„Real estate is about relationships, not deals."
Without connecting buyers, properties, listings, and viewings, a CRM is just a nicer spreadsheet.

„Behavior decides who to call — not gut feeling."
Whoever opens the price list for the fifth time should get a call today — scoring makes that visible.

„When it goes to a client, a human decides."
Automation prepares, summarizes, and reminds — but the agent approves the owner reports and the campaigns.

„Cheap infrastructure isn't a compromise."
Per-minute scheduling runs inside the existing database — zero new vendors, zero extra monthly fees.

„A well-designed CRM is a reusable foundation."
Today, this architecture is the starting point for the CRM systems we propose to other real estate clients.

ARCHITECTURE SUMMARY

Real Estate CRMLead ScoringEmail OrchestrationGoogle Calendar SyncPipedrive MigrationOwner Reporting