The Real Estate Sales Assistant Is Becoming a System

A lot of agents started using AI in the obvious places first... listing copy, email drafts, call summaries, social posts, CRM notes. That was always going to be the entry point because the work is visible, easy to test, and usually annoying enough that even a small improvement feels useful.

That first version was mostly about answers. Ask the model a question, get a draft, take the output, move it somewhere else, then keep managing the process yourself. Helpful, but still fairly shallow. The agent still has to remember the next step, update the CRM, set the follow-up, connect it back to the property, and make sure the whole thing does not disappear into the day.

The more useful version is AI moving from answers to actions.

A recent NBER field experiment across 66 firms and 7,137 knowledge workers found that people using an integrated generative AI tool spent about two fewer hours a week on email, but the study did not detect broader changes in the quantity or composition of their work from individual AI access alone. Another NBER paper looking at Danish labour-market data found rapid AI adoption and reported productivity benefits, but no large near-term effects on earnings or recorded hours. The movement was underneath the surface, in task reorganisation, AI oversight and AI integration. A third NBER paper is more pointed again, finding that higher AI exposure is associated with longer work hours and reduced leisure time, largely because AI complements human labour rather than simply replacing it.

The time saved on one task does not always turn into spare time, often it turns into more output, more checking, more follow-up, more client work, more production, or simply a higher expectation of what one person should now be able to carry. The Guardian recently covered the “workslop” problem, where shallow or flawed AI-generated material gets pushed through workplaces and staff are left to clean it up. In the piece, 92% of executives said AI improved productivity, while only 40% of non-managers agreed. That gap is useful because it shows the difference between producing more and running better. If AI creates more drafts, more notes and more things for someone else to fix, the business has not really improved. It has just shifted the admin around.

Real estate is very exposed to that version of the problem... an agent can transcribe every call, summarise every note, draft every follow-up and still run a loose business if the information does not move cleanly into the next step. A buyer says they want to come back through the property, the note goes into the CRM, but the call still depends on someone remembering. A vendor worries about price, the comment is recorded, but it never becomes part of a clean feedback pattern. An appraisal request comes in, the reply is drafted quickly, but the preparation, comparable sales, past contact history and follow-up rhythm are still held together manually.

The sales assistant comparison is useful because it brings the idea back to how real estate actually works. In a good sales team, the assistant is not there to replace the agent. They prepare the day, keep track of buyers, organise inspection feedback, help with vendor reports, check what needs attention before the next open, keep the diary under control, and make sure important threads do not rely entirely on the agent’s memory.

A buyer follow-up loop is a simple example. The agent takes the enquiry or has the conversation, but the system then links that buyer to the property, checks previous enquiry or inspection history, classifies the likely intent, records the key notes, drafts the follow-up, sets the next task and brings that buyer back into view at the right moment. If they asked for the contract, inspected twice, mentioned finance, or showed interest in a similar property three months earlier, that should not sit as scattered fragments across memory, CRM and email... it should become a sharper next move.

An open-home loop works in much the same way. The agent records comments after the inspection, then the system groups feedback into pricing, presentation, layout, location and urgency. It prepares a cleaner vendor update, flags the serious buyers, identifies who should be called before the next open, and starts turning loose inspection notes into campaign intelligence. The agent still reads the people, the system makes sure the intelligence does not evaporate by Monday.

Vendor reporting is another obvious one. Most agents know the weekly report is more valuable when it carries judgement, not just numbers. But the assembly of the report is often repetitive. Enquiry numbers, inspection numbers, buyer sentiment, price feedback, campaign activity, contract requests, second inspections and recent conversations can all be pulled into a structured draft. The agent then adds the part that actually matters to the client: interpretation, recommendation, tone, and the next conversation.

Appraisal preparation might be the cleanest example of capacity gained. A potential seller asks for a view on price. Overnight, the system can prepare recent comparable sales, suburb activity, nearby competition, ownership history, previous database interactions, likely buyer segments and a suggested pre-appraisal note. The agent still does the appraisal, reads the owner, handles motivation, timing and trust. But the work around that moment is much sharper before the agent even walks through the door.

There is a prospecting version of this as well, and I do not mean cold calling. Domain’s LeadScope is a blunt but useful signal of the direction. It looks at CRM and property data to identify properties in an agent’s database that may be more likely to come to market. The broader point is that many agencies already have useful signals sitting in their database, but the system has not been very good at bringing them forward at the right time. A past appraisal, an old buyer enquiry, a long-held owner, a missed follow-up, a property type someone kept asking about... those signals are often there. The opportunity is not just storing them... it is turning them into tomorrow morning’s useful work.

This is also why the pendulum around AI is starting to look more sensible. The early story was often framed as replacement. Replace the support team, replace the customer service layer, replace the junior worker, replace the assistant. But the last year has shown a more uneven picture. Klarna became one of the poster children for AI customer service, then publicly shifted back toward using humans for parts of the service where quality, complexity and premium support still matter. The better lesson is not that AI failed. It is that work has layers... and not every layer should be treated the same way.

Real estate fits that pattern neatly. The agent should not be replaced in the parts of the job where trust, judgement, negotiation, taste, local context and emotional handling matter. But the admin layer, the research layer, the reminder layer, the preparation layer and parts of the campaign intelligence layer are all much more open to being carried by systems. The point is not to remove the human from the relationship. It is to make sure the human walks into the relationship with better context and fewer loose ends.

Microsoft’s April 2026 Copilot Studio update is useful here because it describes agents and workflows as complementary. Agents bring reasoning and adaptability, while workflows bring structure and consistency. Microsoft talks about workflows that can call agents at the moment reasoning is needed, and agents that can call workflows when a repeatable process needs to run reliably. McKinsey’s March 2026 real estate work points in much the same direction, arguing that agentic AI should not be treated as chatbots bolted onto existing processes, but as systems that can execute workflow steps with approvals and logging. Their line is especially useful for real estate: automate the steps and protect the thoughts.

You can see the same pattern outside real estate as well. Adobe’s April 2026 Firefly AI Assistant is designed so creators describe an outcome and the assistant orchestrates multi-step workflows across Creative Cloud apps. The user gives direction, taste and judgement, while the system carries more of the execution. Zillow is also pushing this from the consumer side of real estate. Its March 2026 AI Mode connects live listings data to actions like tour scheduling and connecting with an agent, while REA’s realestate.com.au ChatGPT app lets Australian property seekers search live listings inside ChatGPT using natural language ("Find me a 3 bedroom home with water views within a 15 minutes drive of medical facilities"). These are different use cases, but the direction is similar: AI is moving closer to action, not just answers.

The Australian enterprise data tells a similar story. Deloitte’s 2026 Australian AI report says 61% of Australian organisations are seeing efficiency gains from AI, but only 30% are using it to deeply transform how work gets done. It also says 69% are using autonomous AI agents, while only 22% have advanced agent-governance models. Thomson Reuters’ 2026 professional-services report gives a similar white-collar signal: organisation-wide GenAI use has nearly doubled to 40%, more than 80% of current users engage with it weekly, but only 18% say their organisation tracks AI ROI. Plenty of businesses are using the tools, but far fewer have worked out exactly where the value is being created.

For real estate agents, the practical difference is fairly clear. One approach is using AI to produce more disconnected material: more emails, more summaries, more notes, more drafts, more content. Some of that is useful... but it can easily turn into more noise. The better approach is using AI to carry more of the assistant layer underneath the agent, so the work moves from conversation to context to action without being rebuilt manually every time.

That is where the individual agent starts to feel more like a small operating unit... not because they are doing everything themselves, and certainly not because judgement disappears. If the assistant layer becomes more system-driven, the agent can spend more time on the parts that still need a person: the appraisal conversation, the vendor wobble, the negotiation, the call before auction, the moment where a client needs confidence rather than another update.

I do not think the best use of AI in real estate is simply getting two hours back. That may happen, and it is welcome when it does. The larger opportunity is being able to carry more live work without letting it leak. More buyers followed up properly. More vendors updated with context. More appraisals prepared well. More database signals brought back into view. More mornings where the agent starts with a sharper list of what needs attention and why.

That is less about convenience and more about capacity. In a business where a lot of advantage still comes from being a little sharper, a little faster, and a little more consistent than the next agent, that may end up being the more useful change.