Sources: all 3 figures are from the National Association of Realtors 2025 Profile of Home Buyers and Sellers, NAR's own research deck. The 21x response time figure that used to sit here cited a page that is now a paid link farm, and the 42 hour figure beside it cited a Harvard Business Review article whose text is paywalled, so no reader could check either. Both are gone. Read your own numbers off your own customer relationship manager before you act on anybody's average, including ours.
The three places deals leak out of a brokerage
None of these are marketing problems and none of them are technical problems. Every one of them is a follow-up problem, which is the only category of business problem that software genuinely fixes. A brokerage with a lead problem usually does not have a lead problem. It has a pile of leads it already paid for and never reached.
1. The lead nobody reached inside the window
Somebody fills in a form on a listing at 9 on a Sunday evening. They are not shopping carefully and they are almost never talking to only you: portal inquiries go to several agents at once, and the buyer's own attention lasts about as long as the browsing session that produced it. The study Dr. James Oldroyd ran at MIT with InsideSales.com, across more than 15,000 leads and 100,000 call attempts, put the difference between a 5 minute response and a 30 minute one at roughly 100 times more likely to make contact and 21 times more likely to qualify. The same body of work found the average company taking about 42 hours to answer at all, in a Harvard Business Review audit of 2,241 firms.
Now put that against how the industry actually reports its own behaviour. In the National Association of Realtors 2025 profile, 67% of first time buyers and 76% of repeat buyers interviewed only 1 agent before deciding who to work with. So the race is not to be the best agent they spoke to. It is to be the one they spoke to.
2. The database nobody worked
This is the expensive one and it is invisible, because nothing about it shows up as a loss. Every brokerage is sitting on thousands of past clients, old inquiries, open house sign-ins and dead portal leads. Each of those was paid for once. Reaching them again costs nothing but the reaching, and a past client who moves every 7 to 10 years is not a cold lead, they are a warm one on a slow clock. Most databases are worked as a mailing list, which is not the same thing as being worked at all: a mail to 4,000 people is 1 event, while a check-in that arrives when a specific person's mortgage is 5 years old is 4,000 events.
3. The listing appointment nobody followed up
A seller lead is the highest value conversation in the building and it usually arrives before the seller is ready. Somebody asks what their home is worth 6 months before they list. The valuation goes out, the conversation ends, and nothing happens for 6 months, at which point the seller lists with whoever happened to be in front of them that week. The follow-up that wins that listing is not clever, it is only long: it has to survive half a year of nothing happening, which is exactly the kind of patience a person with a full pipeline does not have and a system does not mind.
What the automation must never be allowed to do
This industry is licensed, and the parts of the job that carry the licence are the parts to keep away from the machine. Written into the configuration before anything goes live:
- No valuation, ever. The assistant never puts a number on a specific property, not even a range, not even one it read off an automated estimate. It offers the appointment where a licensed agent does that.
- No advice on condition, disclosure or negotiation. Those questions get a person, immediately, and the assistant says so plainly rather than deflecting.
- It says what it is when asked. An assistant that dodges the question is a brand problem and, in a growing number of states, a compliance one.
- Fair housing is not a prompt, it is a hard stop. The assistant never asks about or responds to anything touching a protected class, including apparently harmless questions about schools, neighbourhood character or who lives nearby. Route those to a person who has had the training.
- It never contradicts what is on the listing. If it cannot read the current status from your system, it does not guess at it.
What this looks like on a real inquiry
Before
A buyer fills in the form on a 3 bedroom listing at 9:40 on a Sunday evening. The lead lands in the shared inbox and in the customer relationship manager. The agent on duty sees it at 8:15 on Monday morning, calls at 8:40, and gets voicemail. They try again Tuesday. By then the buyer has been on 2 showings with the agent who called them back at 9:47 on Sunday night, and 76 times in every 100 they never speak to a second agent at all.
After
The same form fires an assistant inside 60 seconds. It texts first, because a text at 9:41 on a Sunday is welcome and a phone call is not, and it asks 3 things: is this the right number, are you buying or selling, and are you looking in the next 30 days or later this year. The buyer answers 2 of the 3 in the next 4 minutes. The assistant offers 2 real showing slots read live out of the agent's calendar, books one, and puts a task on the agent's phone with the whole conversation attached. Nobody was woken up, and Monday morning starts with an appointment rather than a voicemail.
The right build for each part of the brokerage
Speed to lead on new inquiries
- Text inside 60 seconds, call inside 5 minutes, on every source including the portals
- Qualifies buying or selling, timeline and financing status, nothing more
- Books straight into the agent's live calendar
- Escalates to a person the moment the lead asks for one
Database reactivation
- Works past clients, old inquiries and open house sign-ins as individuals rather than as a mailing list
- Triggers on the date that matters to that person, not on your campaign calendar
- Hands anything that answers with intent to an agent the same hour
- Costs nothing per contact, because you already paid for them
Seller lead follow-up
- Survives the 6 to 12 months between the valuation request and the listing
- Checks in on a schedule that thins out rather than stops
- Never quotes a price, only offers the appointment
- Wakes the agent up the week the seller starts answering differently
Under contract and after closing
- Milestone updates to both sides so nobody rings to ask where things stand
- Review request at the moment the client is happiest, which is the day of closing
- Anniversary and equity check-ins that keep a past client warm for the next 7 years
- Referral ask made once, at the right time, rather than never
The tools doing the work
| What it does | Tools | Monthly cost | Setup |
|---|---|---|---|
| Customer relationship manager built for real estate, holding the database and the routing rules | Follow Up Boss, Sierra Interactive, BoldTrail, Lofty | $70 to $500 and up by seat | Low |
| AI conversation layer that texts and calls new leads inside the window and qualifies them | Structurely, Ylopo, the AI assistant inside Lofty or Sierra | $300 to $1,000 | Low |
| Answering service for overflow and after hours on the main line | Smith.ai, Ruby, Goodcall | $150 to $600 | Low |
| Database reactivation campaigns triggered on the person's own dates | The automation already inside your customer relationship manager, plus Structurely for the replies | Usually included | Low |
| Reviews and referral asks at closing | Birdeye, Podium, NiceJob | $75 to $300 | Low |
| Custom intake and reactivation agent across voice, text, your customer relationship manager and the listing feed | Built by OpsJuice on Retell, n8n and Follow Up Boss or Sierra | Project based | Managed |
Those are the ranges we see in the market in August 2026, not quotes. Confirm current pricing with each vendor before you budget, because per seat pricing in this category moves.
Build or buy
Buy first, and buy the cheap thing. Nearly every brokerage that believes it needs a custom build has an automation module and a lead routing rule sitting switched off inside a customer relationship manager it already pays for. Those cost nothing to turn on and they recover real money inside a month, which is also the fastest way to find out whether your agents will actually work what the system hands them. If they will not, no amount of building fixes it, and better to learn that for 0 dollars.
A custom build earns its place for a specific reason rather than as an upgrade: several offices with different routing and split rules, a lead source your customer relationship manager cannot ingest without somebody retyping it, or a referral and relocation book that needs rules of its own. If none of those describe you, the off the shelf stack is the correct answer and it is the cheaper one.
The first 30 days, in order
- Days 1 to 3, measure the leak before you fix it. Pull the last 90 days of leads out of your customer relationship manager and read 2 numbers: the median time to first contact, and the share never contacted at all. Almost nobody knows these before they look, and the second one is usually the shock.
- Days 4 to 10, close the window on new leads. Turn on instant text and a 5 minute call attempt on every source, including the portals that route around your forms. This is the change that pays for everything after it.
- Days 11 to 20, work the database you already own. Segment past clients, dead inquiries and open house sign-ins, and start the check-ins that trigger on their dates rather than yours. Nothing here costs per contact, so the only limit is how well the list is tagged.
- Days 21 to 30, catch the seller leads on the long clock. Put every valuation request from the last 12 months into a follow-up that survives 6 months of silence, and set the review and referral asks to fire at closing.
Only after those 4 are running does a custom build make sense, and it makes sense for a specific reason rather than as an upgrade: several offices with different rules, a lead source nothing off the shelf will ingest, or a referral book that needs its own logic.
Where to go next
- AI for real estate brokerages, by city, which is the same guide read against the local shape of 25 metros
- AI for law firms, the other business where the first responder usually wins the client
- AI for home services and AI for contractors
- AI for med spas, AI for dental practices, AI for veterinary practices and AI for auto repair shops
- AI for insurance agencies