AI can help an estate agency respond to new property enquiries, organise the relevant context and make sure the next action does not disappear into an inbox. The useful system is not a chatbot that tries to replace the agent. It is a controlled workflow connecting the portal or website, the property record, the CRM and the person responsible for the conversation.
The business question is whether that workflow helps agents hold more useful conversations and arrange more qualified appointments without creating inaccurate property claims, unwanted messages or a database nobody trusts.
This guide maps a proposed workflow for estate agencies, brokerages, letting agents and property developers. The examples and calculations are illustrative, not Binarify client results. See our AI consultancy for real estate for the wider implementation approach.
Start with the enquiry the customer made
A person who asks about 18 Market Street has made a specific request. The first response should address that request before the agency tries to start a long marketing sequence.
For example, the workflow may be allowed to use these approved facts:
- the property identifier and current listing status;
- the published price or rent;
- viewing availability held in the agency’s system;
- the branch or agent responsible for the property;
- the person’s message and communication channel; and
- previous conversations already attached to the same CRM contact.
It should not invent availability, describe a feature that is absent from the approved listing, imply that an application will succeed or answer a question that needs professional judgement. If the source systems disagree, the response should flag the conflict for an agent instead of choosing whichever value looks plausible.
The acknowledgement should be useful. “Thank you, somebody will contact you” confirms receipt but leaves the original question unanswered. A better draft can identify the property, answer a verified factual question, state when an agent will respond and ask for one or two missing details that will help the conversation.
The complete lead-response workflow
1. Capture the source event without losing context
Enquiries can arrive through a property portal, the agency website, email, telephone, social media or an open-house form. Preserve the original source, timestamp, property reference, campaign information, message, contact details and any communication preference supplied with it.
Do not assume an email forwarded from a portal contains every permitted field. Do not scrape extra data simply because it is technically accessible. The integration should use the authorised feed, mailbox or API and retain only what the agency needs for the defined purpose.
Zillow’s CRM integration documentation illustrates why the connection must be checked rather than assumed: its older Tech Connect route is described as one-way, while supported CRM integrations synchronise certain fields in both directions. Product availability, supported fields and terms vary. The same questions apply to every portal and CRM combination.
2. Match the person and property to existing CRM records
Before creating a record, search for an existing contact using the identifiers the agency has approved. Email and telephone matches can help, but shared numbers, spelling changes and formatting differences create uncertainty.
A practical deduplication result has three states:
- confident match: attach the enquiry to the existing contact and show why;
- possible duplicate: place both records in a review queue; and
- new contact: create the record with its source and consent evidence.
The property needs the same discipline. Use the stable listing or property identifier where available. A text match on an address can connect the wrong unit, development phase or relisted property.
Never merge contacts solely because an AI model says they appear similar. A bad merge exposes one person’s conversation to another and damages every later report.
3. Route the enquiry using explicit operating rules
The routing decision should combine stable rules with the information in the enquiry. Useful inputs may include:
- listing ownership and branch responsibility;
- sales, lettings, commercial or new-development team;
- language and stated communication preference;
- staffed hours and current agent availability;
- existing client ownership; and
- a declared urgency such as a viewing requested for the same day.
Use normal workflow rules for facts the business can state precisely. AI is useful for interpreting an unstructured message such as “I need something similar near the station, but my lease ends next month.” It can extract the location and timeframe for review; it should not use a person’s name, disability, family status or another protected characteristic to decide which agent deserves the lead or which housing opportunity to show.
Define the fallback before launch. If no responsible agent is available, the system can send an approved acknowledgement and move the enquiry to a covered queue. Silently assigning work to somebody on leave only makes the dashboard look fast.
4. Prepare a grounded response for review
Generate the draft from the enquiry, approved listing facts, branch service information and a small set of reviewed examples. The draft should distinguish three things:
- what the agency can confirm from its systems;
- what the prospective buyer or tenant has said; and
- what the agent still needs to establish.
For a straightforward enquiry, a pre-approved response may be sent automatically under narrow conditions. For example, it can confirm receipt, repeat the published address and provide a link to select from viewing slots already released by the agent.
Route the draft to a person when the enquiry involves negotiation, property suitability, an offer, finance, an application, legal interpretation, a complaint, a vulnerable customer or facts missing from the approved source. Give the reviewer the original message and the facts used to prepare the reply. A review button without the evidence is only an extra click.
5. Write the interaction and next action back to the CRM
The CRM should remain the operational record. Store the message, channel, time, responsible person, review decision and next action against the contact and property. Do not leave the only useful history inside a separate AI dashboard.
Use idempotent write-back: retrying a failed integration must not create another contact, another task or another viewing. Make failures visible to an owner, and reconcile records that arrived while an integration was unavailable.
Agree a small set of meaningful stages. One possible structure is:
| Stage | What it means | Required next action |
|---|---|---|
| New | Enquiry captured but not yet usefully answered | Validate, match and assign |
| Acknowledged | Relevant information or a clear response commitment sent | Agent review or contact attempt |
| Contacted | Two-way communication established | Confirm requirement and next step |
| Qualified for appointment | Agreed criteria for a viewing or consultation met | Schedule and confirm |
| Nurture | No immediate appointment, but future contact is appropriate | Follow permitted cadence |
| Closed | Completed, declined, duplicate, invalid or opted out | Record the reason and stop automation |
“AI qualified” should not be a vague stage. Write down what the agency needs to know before a viewing or consultation and which answers require an agent’s judgement.
6. Trigger follow-up from events, not a blind sequence
A fixed series of five messages can continue after somebody has replied, booked, opted out or chosen another property. Follow-up should respond to CRM state.
Examples include:
- remind the assigned agent when no meaningful action is recorded within the agreed service level;
- send a reviewed viewing confirmation after an appointment is created;
- stop the new-enquiry sequence when a two-way conversation begins;
- create a task when the prospect asks for a different type of property;
- pause when the property is withdrawn or the listing facts are stale; and
- suppress future marketing when the person opts out.
The system should record why each message was permitted and which event stopped the sequence. That makes an incorrect message diagnosable.
Separate requested information from marketing nurture
The first answer to a specific enquiry and a later campaign promoting other properties are not automatically the same activity. Record the source and scope of any consent, channel preference, opt-out and suppression rule.
For UK operations, the Information Commissioner’s Office guidance on electronic and telephone marketing explains that rules differ for solicited and unsolicited messages, channel and recipient type. It also says consent must be clear and specific where required, and that people must be able to withdraw it.
In the United States, the Federal Trade Commission’s CAN-SPAM guide covers commercial email, including accurate sender information and a working opt-out route. Calls and texts can involve separate federal and state requirements. Other target markets have their own privacy, spam, real-estate and consumer rules.
Implementation should therefore establish the applicable rules in each market and encode the agency’s approved communication policy. An AI model should not decide whether consent exists from the tone of a message.
Measure the whole conversion path
Speed matters, but “instant response” is not a sufficient success measure. An automatic receipt sent in one second can hide an enquiry that waits two days for a useful answer.
The National Association of REALTORS® 2025 Technology Survey reported CRM among the leading technologies for quality leads. That supports connecting the workflow to the CRM, where the agency can measure the downstream outcome rather than counting generated drafts.
Take the baseline before changing the workflow. Our general guide explains how to measure lead follow-up before automating it. For real estate, add property and appointment outcomes:
| Measure | How to define it |
|---|---|
| Time to first useful response | Arrival to a response that answers the request or gives a specific next step; report median and 90th percentile |
| Assignment within service level | Enquiries assigned to an available owner within the agency’s agreed time, divided by valid enquiries |
| Contact rate | Valid enquiries that reach two-way communication within a fixed period |
| Qualified appointment rate | Enquiries that meet the agency’s defined criteria and book a viewing or consultation |
| Appointment completion rate | Scheduled appointments that take place, reported separately from cancellations and no-shows |
| Incorrect routing rate | Enquiries reassigned because the original destination was wrong |
| Duplicate and merge error rate | Duplicate records created and incorrect merges found during review |
| Opt-out and complaint rate | Recipients who opt out or complain, divided by messages delivered |
| Agent handling time | Time spent finding context, drafting, checking and updating the CRM |
Break results down by source, branch, staffed versus unstaffed hours and enquiry type. Do not compare a portal delivering viewing-ready prospects with a broad awareness campaign as though their conversion rates should be identical.
Use a stable time window for later outcomes. A lead received yesterday has not had the same opportunity to reach an appointment as a lead received four weeks ago.
An illustrative business-case calculation
Suppose a team receives 1,200 valid enquiries per month. It currently spends an average of six minutes capturing, finding context and preparing the initial response. An assisted workflow reduces the reviewed average to three and a half minutes.
That releases 50 hours per month:
1,200 × (6 − 3.5) ÷ 60 = 50 hours
Value those hours using the agency’s loaded operating cost and realistic use of the released capacity. Then add any change in qualified appointments or completed instructions, using observed conversion and contribution rather than an assumed revenue multiplier. Deduct software, messaging, model, integration, review and support costs.
The calculation is not a forecast. It shows what must be measured during a pilot for the project to earn further investment.
Check the CRM before building custom software
Many property portals and real-estate CRMs already provide lead capture, templates, reminders, routing or two-way synchronisation. Configure and test those capabilities first.
Custom work may be justified when:
- enquiries enter through several systems that do not share a dependable record;
- the team repeatedly assembles property and conversation context by hand;
- existing routing ignores real availability or case ownership;
- a valuable review or quality-control step is missing;
- the platform cannot measure the agreed outcome; or
- the agency needs a controlled connection across products with supported APIs.
It is less likely to be justified when the real problem is that required CRM fields are unused, agents work outside the system or nobody owns unassigned enquiries. Those issues need process ownership and adoption before another interface is built.
A contained first implementation
Start with one branch, one enquiry source and one property category. Include normal enquiries, duplicate contacts, unavailable properties, out-of-hours messages, incomplete contact details, replies, opt-outs and integration failures in the test set.
Run the workflow in draft mode first. Agents review the proposed match, route, response and next action while the current process remains available. Record corrections and classify their causes. Expand automatic actions only where the test evidence and agency policy support them.
Binarify’s AI Impact Diagnostic maps that workflow, checks the portal and CRM capabilities, establishes the baseline and recommends a contained scope—or explains why configuration and process changes are enough.
Explore our real-estate AI consultancy, see our wider AI automation consulting service, or book a 30-minute conversation about the point where your enquiries currently slow down.