In 2026, I believe AI is starting to do for real estate firms what Moneyball did in 2002 in baseball — rebuild the work around what humans actually do best.
Morgan Stanley estimates AI can do roughly 37% of real estate work overall [0]. My estimate after a 4 years of building AI systems inside real estate firms is that roughly 70% of the work could be moved to AI, if done right.
The other 30% — trust, the showing, negotiation, emotional intelligence, local market knowledge — is exactly what you want your agents to spend their time on.
This article is a short intro for key stakeholders in RE firms: which workflows to leave to human agents, which to let Agentic AI handle with human agents review (HITL) and which safely to hand over to AI.
Five things AI cannot do in a real estate sale (yet, in 2026)
I want to start with the limits. If you understand these, the rest follows naturally. It is part of the mindset I use rebuilding real estate firms around AI.
Trust at the high-stakes moment.
Buying a property is, for most clients, the largest financial decision of their life. They want a person they can hold accountable, eye contact at a coffee meeting in DIFC, in Limassol, or in the Phuket office — wherever the deal happens.
In 2024, Air Canada was held legally liable for a refund policy its chatbot invented and surfaced to a customer [1]. If your AI says it, you said it — that is the precedent now. Apply that to a $2M villa transaction and ask yourself how comfortable you are with the risk on the trust-building part of the funnel. None of the business owners I work with are willing to take it.
Tacit knowledge about the local market.
There is a kind of knowledge an experienced agent carries that doesn’t exist in any database — things like a developer in Dubai South having three handover delays in the last eighteen months, a particular tower’s elevators breaking down every summer, or a community’s HOA being in a legal dispute that makes resale impossible.
This information is stored in the heads of agents, in WhatsApp groups or in personal conversations. It is not in any LLM training set. It cannot be scraped from Bayut, Property Finder, or any local portal. The day an AI tells a client “I know that tower, my colleague closed three units there last quarter and two of them came back with handover issues” is the day this section of the article becomes obsolete. We are not there.
Physical presence at the showing.
You cannot replace a property showing. In residential real estate, the client wants to walk the unit, feel the marble, hear the traffic from the balcony, smell whether the previous tenant smoked. The agent watches the client — where they linger, what their face does in the master bedroom — and that observation can change the client profile for an experienced sales manager.
A 3D tour and an online walkthrough do not give you any of this. AI does not stand in the kitchen with the client.
Reading the emotional context of a deal.
I know agents who closed transactions where the client was going through a divorce, a parent’s death, or an immigration crunch. Half the work in those deals was not real estate. It was holding the client’s hand through a hard moment. AI is not suitable for this. The closest it can do is generate a sympathetic-sounding message, which is exactly the thing that makes it worse when the client is already vulnerable.
The real moment of negotiation.
Negotiating a price reduction with a seller’s agent is not a script. It involves reputation, long-term consequences, timing, and the read of whether the other side is actually about to walk. LLMs are not reliable in this kind of work — they are easy to manipulate in long, multi-party conversations with social stakes, and they fall apart on the multi-step coordination that real negotiation requires.
I’ll add a caveat to be honest about what’s happening. There are early experiments where AI takes part in genuinely high-stakes calls — AI-driven investors writing checks to startups is the most visible one. So this is not impossible in principle. It is not reliable enough though to use in a real estate transaction. The cost of such mistake is the deal and (worst case) the reputational damage for the firm.
These five things all sit at the high-trust and high-stakes part of the funnel. None of them changes in the next twelve months.
Where AI fits in the agent’s week
Knowing AI can’t do those five things focuses you. You can now target the workflows where ROI is highest and risk is lowest — which is exactly the rest of the agent’s week.
Your agents do not just close deals. In a typical week they do twenty different things — answer cold leads, qualify those leads, follow up over months, schedule showings, write listings, coordinate documents, summarize calls into the CRM, send thank-you notes, handle objections, advise on financing, walk units, and yes, negotiate and close.
A few of those twenty things are exactly the five we just listed. Most are not.
AI handles most of the others, and it handles them well — often better than humans, because the bottleneck on those tasks is (usually) not intelligence. It is either speed or consistency.
When AI answers at 2am on Sunday, doesn’t get tired by lead #47 of the day, and doesn’t forget to follow up on day 19 — it gives huge impact.
Let me show you what this looks like in practice.
Four workflows AI takes over
I’ll walk through four workflows where I’ve seen or built AI to take over an agent’s time successfully. Each one is something an agent currently spends hours on per week.
Inbound first response, around the clock
The statistic around first response time is infamously brutal. A real-estate-specific WAV Group study of 384 agents across 11 US states [2] found the average response time on a new inquiry was 917 minutes — fifteen hours — and 48% of inquiries went unanswered entirely.
Harvard Business Review’s The Short Life of Online Sales Leads adds the cost of that delay: qualification odds drop 21x once a lead waits longer than five minutes. AI is the only system that responds in under a minute, every time, around the clock.
I have not yet met a real estate firm where deploying this honestly fails. The failures I’ve seen come from teams that deploy a bot with no backup. Klarna did this in 2025 with their entire support team [3] and walked it back within six months because complex issues had nowhere to go. Done right — AI for the first response, human for anything substantive — this is one of the easiest wins available to an agency owner today.
Lead qualification before a human gets involved
This is where the time savings get big. In firms I’ve worked with, qualification absorbs eight to twelve hours per agent per week — and most of that time goes to leads that never close. Moving qualification out of human hours frees the working day for the leads that actually convert.
The trap here is to confuse “qualification” with “discrimination.” AI qualifies on whatever criteria you give it. If you have not thought clearly about who your real customer is, you will simply automate the wrong filter, just faster.
If you are not sure how to make it, play safe: let AI handle pre-qualification — and then pass it to an agent or first-line specialists (when either of them is back online).
Follow-up and nurturing across the long cycle
In Dubai, Cyprus, or Phuket, the typical sales cycle runs three to twelve months. A lead that didn’t buy in March might buy in September. The agent who kept the relationship alive without being annoying wins that deal.
AI’s edge here is reliability, not quality. Your best agent will not send 800 personalized check-ins per quarter. The bot will. In long-cycle markets, that compounds into deals you would otherwise lose.
Voice memos to structured CRM records
I rate this as the highest-leverage hour an agent will recover in 2026. Most CRMs in the firms I’ve audited over the last three years are graveyards because the agents don’t update them.
Voice-to-CRM with AI fixes the discipline problem at the root. It also gives the founder a real view of what happened in the field instead of an empty dashboard — which, if you’ve ever tried to manage a sales team you can’t see, is a category of pain on its own.
These four are not the full list. There are six more workflows that come up in real-world engagements: showing scheduling, document extraction and contract review, KYC compliance, property matching, listing generation, and post-show feedback collection. I’ll cover the full inventory in Article 2 of this series.
Inside Krenels: the Broker Studio project
One of our recent projects at Krenels — internal name Broker Studio — was built for an international real estate brokerage operating across seven regions. The bottleneck they came in with was time-to-offer. When a client expressed interest in a property, putting together a personalized offer was a manual job — a manager would search live inventory, pick out matching units, build a presentation, and tailor the language to the request. At their lead volume, this took hours per offer, and an offer that takes hours arrives after the lead has already looked at three other agencies.
What we built was an AI layer that sits in the middle of that workflow. The agent’s job stays the same — they decide what the client needs and review the final output before sending. The mechanical part — finding matching units in live inventory, assembling them into a presentation, formatting the package — is now automated.
Three numbers from the deployment:
- Time-to-offer dropped from 72 hours to 20 minutes. Same speed-to-lead lever as the inbound workflow above, applied at the offer stage.
- Manager hours recovered: roughly 6,825 per month across the full operation.
- Adoption: 22 weekly active users out of 41 deployed seats, across 7 regions.
I include the adoption number because it matters more than the speed gain on its own. Most AI rollouts in real estate fail because of adoption — the tooling works fine, but the team finds reasons not to use it. A 54% weekly active rate on a recently deployed product is healthy and tells me the workflow has been internalized by the agents who use it daily.
We did not track conversion lift directly on this project, and I won’t put a specific number on it without measurement. What’s defensible is that compressing time-to-offer by 200x in a market where speed is the dominant predictor of close rate is moving the largest lever in the funnel.
Outside Krenels: a McKinsey home builder case
McKinsey published a 2026 review of agentic AI in real estate [4] that I find more useful than most public case work in this category. Working with home builders, McKinsey’s team implemented agentic workflows around inbound lead engagement. Two outcomes from the report:
- Lead response times improved by more than 90%.
- The firms captured incremental home sales from after-hours AI agents that engaged with buyers around the clock.
The second outcome is the one I often mention in client meetings. Response time is the surface number. Underneath it is the question of what happens to the deals that would otherwise have gone to a competitor — or simply disappeared because the buyer’s first inquiry arrived at 10pm and got buried until Monday morning. McKinsey’s report counts those recovered sales explicitly.
I trust this case study more than most chatbot-vendor case studies for one reason: McKinsey is not selling the AI tool. They are advisors who watched the deployment work in production and reported the numbers.
Where to start
When founders ask me how to actually start, the order I recommend is the same in almost every intake session at Krenels.
The first deployment I would put in place is 24/7 inbound first response, with lead qualification on the way through. The risk is low, the payback comes inside the first sixty days, and the prerequisites are minimal — a website form, a CRM, and budget for one tool. You don’t have to be data-mature to get there, and the conversion lift on this single change usually justifies the rest of the work that follows.
After that I would add voice-to-CRM and follow-up automation. This is where a ten-agent team typically reclaims around a hundred hours of manager time per month. The catch is that the CRM has to actually be used by the team. That requires real change management — and that’s where most implementations stall, even when the tooling itself works. I’ll cover that in a separate piece.
The third layer, which I would only add once the first two are running cleanly, is document extraction, contract review, and KYC checks. These workflows always need human review on every output. Real estate documents are legally consequential, so “let the bot decide” is not an acceptable risk profile. The benefit is that the drudge work moves off the transaction coordinator’s plate, which removes one of the bigger sources of operational attrition in agency life.
A short list of things I would not yet attempt: full negotiation automation, AVM-only pricing for luxury units, and fully automated buyer support without an escalation path. I’ve seen all three tried at firms in 2025-2026 and walked back inside a quarter.
The decision in front of you in 2026
If you run a real estate firm — agency, brokerage, or developer — AI now does twelve to fifteen hours of weekly work per agent in seconds. Your call is whether to let your team keep doing it manually.
The founders I see successfully growing are reorganizing their teams around the parts of the dealflow where humans stay in the loop — trust, showings, negotiation. The founders I see struggling are either chasing the dream of replacing the whole team or pretending nothing changed. Both of those positions cost you lost profits.
If you want to talk about this for your specific firm, you can reach me through Krenels.tech. The next article in this series covers why most AI implementations fail at the organizational level, even when the tech itself works.
[0] — https://www.morganstanley.com/insights/articles/ai-in-real-estate-2025
[1] — https://www.americanbar.org/groups/business_law/resources/business-law-today/2024-february/bc-tribunal-confirms-companies-remain-liable-information-provided-ai-chatbot; gladly, the fine was around 1000$; but there were much worse cases like Deloitte charge back to Albanese goverment
[2] — https://www.wavgroup.com/2014/01/13/agent-responsiveness-study-reveals-critical-flaws-in-real-estate-lead-response/. Don’t get fooled that it is dated 2014 — the numbers I’ve seen with clients looks frighteningly similar.
[3] — https://strategicmarketingtribe.com/marketing-news/b/klarna-ai-backlash-human-support-trend-2025
[4] — https://www.mckinsey.com/industries/real-estate/our-insights/how-agentic-ai-can-reshape-real-estates-operating-model; great article, though lacking details; still recommended for business owners, CDTO and products working in RE firms