Skip to content
qiyas.cc
Problem Solver
cc · blog · Business 2026-11-14 · 11 min

CRM for a real estate agency: what it actually needs

A spreadsheet, a WhatsApp chat and an agent's memory are not a CRM. What a real estate agency's CRM should contain, what I learned building them for the Azerbaijani market, and when HubSpot or Bitrix24 is enough.


In a lot of agencies the “CRM” is a spreadsheet, WhatsApp on each agent’s phone, and the agent’s memory. It works until the agent quits. Then the clients leave with them, together with the owners’ numbers and every “this flat frees up at the end of the month”.

I have built several systems for the real estate market in Azerbaijan: a platform for agents, an agency that runs over WhatsApp, a service that posts one listing to five portals, and Python clients for bina.az, tap.az and lalafo.az, the country’s big listing sites. This post is what I took away from that work: what a real estate agency’s CRM needs, what it can live without, and when you should just buy one.

A CRM is not a contact list

A generic CRM tracks one thing: a customer and the deal with that customer. Real estate has two sides. On one side there is the owner who sells or rents out, and their property. On the other there is the buyer or tenant, and what they are looking for. The agency’s actual job is to connect those two lists.

That is why a system that starts with a “contacts” table hits a wall early. A real estate CRM should start from the property and the requirement.

The parts

Part What it holds
Property address, map pin, district, metro, rooms, area, floor, renovation, title deed, mortgage eligibility, price, photos
Owner name, phone, which properties are theirs, agreed terms, last call
Client name, phone, where they came from (portal, Instagram, referral)
Requirement districts, rooms, budget, minimum area, “not the top floor”, title deed required, buy or rent
Deal which requirement is tied to which property, and at what stage
Showing date, time, agent, property, client, outcome
Conversation calls, WhatsApp and Telegram messages, notes
Listing where the property is published, its status, when it expires
Agent and commission who handles what, who earned what

That table is the skeleton. Everything else sits on top of it.

1. Property and owner

The property card is the form agents fill in most, so it has to be quick. Sale and rent need different fields: for a rental it is the deposit, utilities and whether families with children are welcome; for a sale it is whether there is a title deed (in Azerbaijan, the state register extract everyone calls “kupça”) and whether the flat qualifies for a mortgage. On emlakagenti.az the listing form has six steps, and the fields change with the deal and property type. The reason is simple: an agent who sees irrelevant fields leaves the form half done.

Keep the owner separate from the property. One owner can have three flats, one flat can have two owners. The owner’s phone number is the most valuable thing an agency has, so not every agent should see it (more on that under roles).

2. Client and requirement

Separate these too. The same person rents today and buys next year. A requirement also expires: the client found a place, changed their mind, moved away. It needs a status: active, paused, found.

A requirement can be written as free text (“3 rooms in Narimanov, up to 300k”), but to search with it you have to break it into fields. At Khalif Estate an AI assistant does this over WhatsApp: the buyer writes what they want in plain words and is told how many homes match. Underneath, it is still a structured requirement.

3. Matching

This is the part that pays for the CRM. When a new property comes in, the system should say on its own: “this flat fits these five clients”. And the other way round when a new requirement arrives.

The simplest version is hard filters:

from dataclasses import dataclass

@dataclass
class Property:
    district: str
    rooms: int
    area: float
    price: int
    floor: int
    top_floor: int
    has_title_deed: bool

@dataclass
class Request:
    districts: set[str]
    rooms: set[int]
    min_area: float
    max_price: int
    not_top_floor: bool = False
    needs_title_deed: bool = True

def matches(p: Property, r: Request) -> bool:
    if p.district not in r.districts or p.rooms not in r.rooms:
        return False
    if p.area < r.min_area or p.price > r.max_price:
        return False
    if r.not_top_floor and p.floor == p.top_floor:
        return False
    if r.needs_title_deed and not p.has_title_deed:
        return False
    return True

In practice that is too strict. A client with a 300k budget will look at a good flat for 305k, and “Narimanov only” often means “Narimanov or next door”. So after the filters comes a score: a slightly higher price lowers the score instead of removing the property. Matching on emlakagenti.az shows it as a percentage, 96% for this listing, 87% for that one, and the agent calls from the top of the list.

4. Deal stages

Rough stages for a sale:

  1. New inquiry
  2. Requirement clarified
  3. Options sent
  4. Showing booked
  5. Showing done
  6. Negotiation
  7. Deposit taken
  8. Closed at the notary
  9. Lost (with a reason)

Rentals are shorter, with one extra field that matters: the lease end date. In a year the same flat is back on the market and the same tenant is looking again. An agency that stores that date gets two deals out of one.

Make the reason mandatory when a deal is lost: too expensive, wrong district, bought through another agency, stopped answering. After three months those reasons tell you more about your pricing than any spreadsheet.

5. Calls and messages

In Azerbaijan the conversation with a client happens on WhatsApp almost every time. If the chat lives on the agent’s own phone, it belongs to the agent, not the agency. A CRM should bring every conversation into one inbox and tie each one to a client card. On emlakagenti.az, WhatsApp and Telegram messages are answered from one panel.

Speed matters here. A study published in Harvard Business Review in 2011 audited 2,241 US companies by sending them web leads. Companies that tried to reach the lead within an hour were nearly seven times as likely to have a meaningful conversation with a decision maker as those that waited one more hour, and more than 60 times as likely as those that waited 24 hours or longer. That was not a real estate study, but the logic holds: the agent who calls first wins. So the CRM should assign a new inquiry to someone right away and make unanswered ones impossible to miss.

6. Showings

A showing is an agent’s day. Each one should be tied to a property, a client and an agent, and the owner should get a reminder. Afterwards someone writes down the outcome: liked it, did not, and why. That makes the next set of options better.

7. Documents

Title deed, power of attorney, deposit agreement, lease, a copy of the owner’s ID. These belong to the property and the deal, with access set per document. A copy of someone’s ID should not travel around on every agent’s phone.

8. Publishing to portals

Real estate listings in Azerbaijan live on bina.az, tap.az, lalafo.az, yeniemlak.az and a few others. An agent types the same flat into five sites, and when the price changes, edits it five times. Doing that from the CRM removes the most time consuming part of the job.

The catch: none of these portals has a public API for listings. I know because that is why I wrote Python clients for bina.az, tap.az and lalafo.az. They talk to the same endpoints the websites use: GraphQL on bina.az and tap.az, the site’s own requests on lalafo.az. Estate Listing Pusher does it for five sites from one panel: write the listing once and it goes to bina.az, tap.az, yeniemlak.az, vipemlak.az and arenda.az.

One rule from that work: a write must never be retried automatically. If the connection drops while a listing is being sent, the listing may already be live. Send it again and the portal has two copies, and a paid boost gets paid twice. So my clients raise an error in that case and make you check the account first. Paid services (VIP, bumps) are never sent without an explicit confirmation.

And the listing status has to come back into the CRM: in moderation, published, rejected, expired. An agent should see a rejected listing the same day, not a week later.

9. Duplicates

The same flat shows up on the portals many times: the owner posts it, then three agents post it. Two questions matter to an agency: is this listing from the owner or from a middleman, and do we already have this property. The scraper on emlakagenti.az collects listings from 18 sites into one panel, separates owner listings from agents’ and flags duplicates. One field is never enough to find a duplicate. Phone number, address, area, floor and photos are checked together.

10. Agents, roles, commission

Put roles in from day one. Adding them later is painful. At minimum:

  • Manager sees everything.
  • Agent sees their own clients and properties, and other agents’ properties with the owner’s number hidden.
  • Operator takes incoming inquiries and assigns them.

A commission can be split between several agents on one deal: whoever brought the property, whoever brought the client, whoever closed. Record the split when the deal closes, not from memory at the end of the month.

Every agent gets their own login. With a shared password you will never know who deleted what.

11. Reports

You do not need many. These are enough:

  • Time to first response, per agent.
  • Stage conversion: out of 100 inquiries, how many reach a showing, how many a deposit.
  • Reasons for lost deals.
  • Sources: where clients come from, which portal actually brings deals.
  • Days on market, per district.

Khalif Estate calculates the median price per square metre for every Baku district at the start of each month with its own model and publishes it openly. It is a good example of CRM data doing work outside the agency too.

Personal data

A CRM is where names, phone numbers and ID copies live, so data protection law applies. In Azerbaijan it is the Law on Personal Data (11 May 2010, No. 998-IIIQ). If you are elsewhere, your local law (GDPR in the EU, for example) will have its own version of the same ideas. Under the Azerbaijani law:

  • Apart from cases set by law, personal data may be collected and processed only on the basis of the subject’s written consent or information the subject provided in writing (Article 8.1), and proving that consent was obtained is the data holder’s job (Article 8.4).
  • Confidential personal data may be passed to third parties only with the subject’s consent, except where the law says otherwise (Article 5.2).
  • Whoever holds the data must take technical and organisational measures against loss, unauthorised changes and unlawful access (Article 5.8).

For a CRM that means: record on the client card when and how consent was given, restrict access by role, log who looked at what, and keep backups. This is not legal advice, but building the system without these in mind gets expensive later.

Buy or build?

Honestly: for a small agency an off-the-shelf CRM is often enough. Prices from the official pages, November 2026, in US dollars:

Bitrix24 HubSpot Sales Hub Custom CRM
Entry price Free plan available. Basic: $69 a month billed monthly or $49 billed annually, up to 5 users Free for up to 2 users. Starter: $20 per seat a month ($7 as a new customer discount). Professional: $100 per seat a month, $90 billed annually, plus a required $1,500 onboarding fee A one time build cost, then hosting and support
WhatsApp Integration from the Basic plan Through integrations However you want
bina.az, tap.az, lalafo.az Not built in Not built in Can be connected
Property to requirement matching With customisation With customisation A core feature

Bitrix24 charges per company, not per user: Standard is $144 a month on monthly billing for up to 50 users. HubSpot charges per seat. For a ten agent agency that difference is large.

An off-the-shelf CRM is enough when your work is mostly talking to clients and moving deals through stages. A custom one makes sense when:

  • your agents spend their day posting listings to portals and you want that automated;
  • matching is your main tool, not a field somewhere;
  • you need tight control over who sees owner and client details;
  • you want the CRM, the website, the WhatsApp bot and the listings database in one place.

There is a middle road too: keep the bought CRM and connect portal publishing and matching to it as a separate service over an API. Estate Listing Pusher has a REST API for exactly that.

Where to start

Do not build everything on day one. I would go in this order:

  1. Properties, owners, clients, requirements. Import the spreadsheet.
  2. Matching. Even the simplest filter is something agents feel in the first week.
  3. Deal stages and the reason for lost deals.
  4. WhatsApp conversations in one inbox.
  5. Publishing to portals.
  6. Reports.

After each step let the agents work with it for a week and tell you what gets in the way. A CRM agents do not use is, at best, an expensive spreadsheet.

For the wider picture, see common technical problems in business. For the messaging side, what a WhatsApp bot does for a business. And for getting the data model right, what DDD is and how to apply it.

The work mentioned here: emlakagenti.az, Khalif Estate, Estate Listing Pusher, and the Python clients for bina.az, tap.az and lalafo.

If your agency needs a site or a CRM, here is what I build for real estate, or just write to me.

Sources

related