An agent that carries a lead from the portals all the way to the first SMS to the owner.
Signature Estates, a real-estate investment operator, did not need another listing dashboard. They needed someone who watches the portals around the clock, remembers each listing's history, and carries a case all the way to the point where a single approval is enough. The agent sets each fresh listing against its full CRM history, scores the seller signal (price drops, relistings, owner versus agent), writes a short memo, and assembles the queue: fresh radar matches, changed listings, historical candidates worth revisiting, and prepared first contacts. On the strongest leads it reveals the number and writes a discreet SMS to the owner that the operator approves from Telegram.
For whom
This is for you if the best deals disappear faster than you can scan the portals, compare history, and message the owner.
Radar, memory, contact, approval
- 01Every listing kept in memory without flooding the operator
- 02The seller signal sets who to approach first
- 03The agent writes the owner SMS itself; sending waits for approval
- Sector
- Real-estate investment
- Sources
- OLX, Otodom, Gratka, Morizon
- Memory
- More than 6,000 listings with history in the CRM
- Seller signal
- Price, returns, owner versus agent
- First contact
- SMS to the owner, email, or export, on approval
Operating loop
How radars turn listings into decisions and a first contact.
The system starts with saved sources, but its value appears later: event selection, listing memory, the operator memo, a ready SMS to the owner, and the trace of every decision.
Saved portal URLs become radars
Scans store fresh and historical listings in the CRM
Agent scores changes, duplicates, fit, and seller signal
Reveals the number and writes an SMS to the owner for approval
Operator approves in Telegram, the CRM holds the decision trace
System architecture
Sources, memory, tools, and consent gates.
The agent watches the market, stores history, calls tools, and stops before actions that require approval.
- 01
Sources
Radars define what can be watched.
Each radar has saved portal URLs, geography, property type, area constraints, seller preference, and notification mode. The agent does not improvise the scope; it works inside agreed source contracts.
- 02
Memory
Old listings become context.
The CRM stores source, photos, change history, freshness, duplicates, and portal state. Backfill does not flood the operator, but it gives context for price, resurfaced listings, and changes.
- 03
Analysis
The agent explains why a lead matters.
The seller signal scores how ready a seller is to talk: whether the price moves, whether the listing returned after disappearing, and whether an owner or an agent posted it. The agent also weighs radar fit, duplicate history, and the next useful step. Telegram receives a short recommendation, not a raw portal record.
- 04
Consent
Contact, messages, and documents are controlled actions.
On a strong lead the agent reveals the available number and writes a discreet SMS to the owner in their language: no spam, no pressure. Sending starts only once the operator approves it from Telegram. The same goes for a proposal email, a document, or an export: the agent prepares the move, but it does not reveal contacts silently or send anything without the operator's approval.
Why this is agentic
The system moves through several work stages on its own: it watches sources, detects change, uses CRM memory, prepares a recommendation, and calls tools only inside the process boundary.
- Carries a lead from a portal listing to a ready SMS
- Combines portal scans, CRM memory, and a prepared SMS draft in one run
- Separates fresh events from historical listings
- Scores the seller signal from price moves, returns, and poster type
Operating proof
The agent does not stop at an alert. It carries the lead to the first contact.
Most real-estate tools stop at a notification: found a listing, go check it yourself. The value here is further on, in what the agent does after it finds a lead: it remembers the history, judges who is most likely to sell, writes an SMS to the owner, and waits only for approval.
- Memory
Every listing knows its own history.
More than 6,000 listings in the CRM with source, photos, price moves, and returns after disappearing. The agent never reads a listing in a vacuum; it reads its whole past, so it separates a fresh opportunity from old background and spots which seller just dropped their price.
- Contact
An SMS to the owner, not another alert.
On a strong lead the agent reveals the available number and writes a short SMS in the seller's language, with no spam. The operator approves it from the phone in one move, and once approved the message can go out through an SMS gateway, not a personal phone.
- Autonomy
The whole path, one agent.
Watching four portals, detecting a change, reaching into CRM memory, scoring the seller, the memo, and the prepared SMS are one run of work, not several tools wired by hand. A human steps in once, at the end, at the approval gate.
Agent work
The agent carries a lead from a portal signal to an SMS the operator approves.
Problem
Whoever messaged the owner first won, and manual portal-watching was simply too slow for that. The operator kept going back to the same pages, telling a genuinely fresh listing from a rehashed one, checking whether the seller actually wanted to talk, and only then deciding whether it was worth writing. The strongest signal got lost among duplicates and listings that looked identical but meant different things.
How it works
The agent watches four portals (OLX, Otodom, Gratka, Morizon) around the clock, so no one has to click through tabs every morning. It remembers each listing's history, tells a genuinely fresh opportunity from a rehashed one, judges whether the seller is open to talking, and hands the operator a short queue: only what needs a move. On the strongest leads it goes one step further, finds the number and writes a ready SMS to the owner that the operator approves with a single tap from their phone.
Boundaries
The system does not buy, make offers, reveal contacts silently, or send anything outside the agreed mode. It can recommend, analyze history, and prepare the full first contact: reveal an available number, write an owner SMS, prepare a proposal email, a document, or a Google Sheets row. Every step that touches a seller or transaction data waits for the operator's approval.
What broke
Portals change structure, throttle traffic, return old listings, and hide phone numbers behind a click. Early versions produced duplicates and treated a backfilled historical listing as a fresh lead. We added quiet baselines, incremental scans, deduplication, source-health checks, explicit contact-check jobs, and separate states for fresh, changed, missing, and historical listings.
Result
Signature Estates gets a short decision queue instead of a browsing routine: fresh radar matches, changed listings, historical candidates worth revisiting, source issues, and prepared first contacts awaiting approval. An SMS to the owner is approved from the Telegram card in one move, while the CRM holds more than 6,000 listings, radar filters, a hot-lead queue, and the trace of every decision. Telegram handles fast decisions; the CRM holds the proof.
Work surfaces
Telegram is the fast front. CRM is the memory and proof.
Signature Estates gets a short decision queue instead of a browsing routine: fresh radar matches, changed listings, historical candidates worth revisiting, source issues, and prepared first contacts awaiting approval. An SMS to the owner is approved from the Telegram card in one move, while the CRM holds more than 6,000 listings, radar filters, a hot-lead queue, and the trace of every decision. Telegram handles fast decisions; the CRM holds the proof.
Radars
Saved portal URLs are the contract. The system knows which sources to watch, how often to return, and which type of opportunity can move forward.
Telegram
The operator receives a short decision card: why this lead, what changed, and which first contact makes sense. From the card they approve the SMS to the owner, save, or pass, without leaving for the portal.
CRM
The CRM keeps source, photos, listing history, state snapshots, and every prepared contact alongside the decision record: who approved what, and when. It is the system memory and audit trail, not just a table.
System screenshots
The proof is in the work surfaces.
These frames come from the running lead-intelligence system for a real-estate operator: Telegram decisioning, CRM listing memory, a prepared SMS to the owner, the morning brief, radars, and the event queue.
Telegram on mobile
The operator decision fits on one screen.
The three supplied mobile frames are wrapped in a phone and anonymized only where private fields appear: address, names, channel names, and company names.
01Top of the lead card: the numbers, context, and why the offer is worth checking. 
02Bottom of the same card: checklist and operator actions without leaving Telegram. 
03Team channel after anonymization: the bot returns a shortlist and short rationale for deeper review.
- Screen 01re-lead-intel
01Telegram card with actions: why this lead, check phone, draft SMS, save, pass. - Screen 02re-lead-intel
02CRM listing memory: records with source, freshness, radar fit, seller signal, and filters. - Screen 03re-lead-intel
03SMS outreach: the agent writes the message to the owner, and the operator approves the send through an SMS gateway.
Client testimonial
The agent watches the portals for us, remembers each listing's history, and catches the sellers genuinely worth a call, with a first message already drafted. Instead of a browsing routine I get a short decision queue, and nothing goes out without me: the contact stays human. It reads like it was built by people who understand the deal, not just the software.
Autonomy boundary
Boundaries
The system does not buy, make offers, reveal contacts silently, or send anything outside the agreed mode. It can recommend, analyze history, and prepare the full first contact: reveal an available number, write an owner SMS, prepare a proposal email, a document, or a Google Sheets row. Every step that touches a seller or transaction data waits for the operator's approval.
Portals change structure, throttle traffic, return old listings, and hide phone numbers behind a click. Early versions produced duplicates and treated a backfilled historical listing as a fresh lead. We added quiet baselines, incremental scans, deduplication, source-health checks, explicit contact-check jobs, and separate states for fresh, changed, missing, and historical listings.
Recognize your own process here? Let's see what an agent could take over.
- 30 minutes with the engineer who would build it, not a salesperson.
- A review of the processes that cost you the most time and money.
- A written summary: what to automate, in what order, with cost ranges.
No sales deck and no obligations. If automation doesn't make sense, we'll write that too.