Real estate is a document-heavy, follow-up-heavy business with a long sales cycle. That is exactly the shape AI handles well — not the negotiation or the site visit, but the reading, drafting and reminding around them. Five workflows are worth evaluating first.
Where AI fits in a real-estate workflow
| Workflow | What changes | Guardrail |
|---|---|---|
| Lead qualification | Inbound enquiries are read, matched to budget and inventory, scored and routed with a drafted reply. | An agent approves the reply and the score before it is sent. |
| Listing copy | Structured property facts become a first draft for the portal, brochure and social post. | A person checks every factual and legal claim. |
| Document extraction | Agreements, KYC packs and invoices are read into structured fields and attached to the right record. | Low-confidence fields go to a review queue. |
| Follow-up drafting | Stale leads and upcoming milestones trigger a personalised reminder draft. | The assigned agent decides whether to send. |
| Internal Q&A | Agents ask questions across project brochures, price lists and past correspondence and get cited answers. | Permissions follow the CRM; no cross-team leakage. |
The data problem comes first
Most real-estate AI projects fail before the model is involved. Inventory lives in spreadsheets, price lists are PDFs, lead history is in WhatsApp, and two offices disagree about which version is current. An assistant trained on that mess produces confident nonsense. The unglamorous first step is choosing a source of truth for properties, prices and customers, then wiring the assistant to it.
What we learned building Anshwar
We built a real-estate CRM and ERP that brings properties, leads, appointments, attendance, payroll and WhatsApp updates into one web and mobile suite. The lesson that carries into AI work: the value is in the connected record, not the clever feature. An assistant is only as useful as the inventory and customer data it can see, and only as safe as the permissions around it.
You can read the Anshwar CRM & ERP case study for the build detail.
A 30-day starting plan
- Week 1: pick one workflow and measure the current time cost and error rate.
- Week 1: agree the source of truth for properties, prices and contacts.
- Week 2: connect read-only access and build a small evaluation set of real enquiries.
- Week 3: draft responses or extract fields, with a review screen for a person.
- Week 4: run in parallel with the manual process and compare quality and time.
What to measure in the first month
- Lead response time. Median minutes from enquiry to the first human-approved reply, compared with the old process.
- Draft acceptance. The share of generated messages sent with light edits, versus rewritten or thrown away.
- Data completeness. The percentage of listings with a current price, availability, location and owner on record.
- Escalations. How often the assistant hands over to a person, and whether the reasons are understood.
- Cost per enquiry. Model calls plus review minutes, measured per workflow rather than in total.
If those numbers are not moving after a month, the system is a demo with a login. Cut the workflow and go back to the one that pays.
Consent and WhatsApp
WhatsApp is the default channel for property enquiries in India, which makes consent part of the architecture. Use the official Business API, keep opt-in records, respect template-message rules and let a person approve outbound campaigns. An assistant that messages people who did not ask to hear from you is a compliance problem, not a growth hack.
Guardrails that matter
- No automated promises about price, availability or legal terms. Those come from an authorised person.
- Advertising and listing copy must follow local rules and fair-housing expectations; a human checks before publishing.
- Customer data stays within the CRM's permission model, including WhatsApp and email threads.
- Every generated message has an audit trail: who approved it, when and from which facts.
- Model and hosting costs are monitored per workflow, not discovered at the end of the month.
If you want to scope a first workflow, our AI app development page covers how we build and price these systems. If you are earlier in the process, start with the AI automation workflow guide.
Mayur Saptal
CEO at Oysty India. Writes about AI delivery, scoping and the unglamorous part of shipping software. You can check the public client reviews on Upwork.