Before: a request got one call, then silence
A European property marketplace (residential and commercial real estate) receives thousands of buyer and tenant requests a year across ten countries. Each one got a first call. If the property seeker did not answer, or answered "not yet", the request went quiet. Nobody had the time to come back to it three weeks later.
Called once, then moved on
A rep handled new requests first. Going back through the old ones meant opening each record, reading the history, finding the number, then typing the message by hand.
≈ 8 min per follow-up, when it happened at allUpdated the CRM by hand
"Found elsewhere", "paused", "still looking": each outcome was typed into the record after the call, or not typed at all when the day ran out.
Status often missing or staleHad no view of the backlog
How many requests were sitting unanswered, and since when, was a question nobody could answer without an export.
2,320 eligible requests when we first countedThe system: a triage agent, a writer, and one approval click
The assistant reads the CRM every hour, picks the requests worth a message, writes in the company's voice, and waits. A person approves each draft. Only then does it send, inside daily caps, and it records what came back.
Triage every hour
The agent reads requests older than three weeks with no reply on file, drops anyone who opted out, and ranks the rest by size and age. 225 requests triaged into a 40-message plan in one pass.
One opener, in the house voice
An approved template asks one thing: are you still looking? Foreign numbers get English. No reminder before day 5, and day 10 only fires if day 5 says "not contacted".
Noon drafts, one click each
At 12:00 every reply that came in is answered in draft, with a suggested decision (found elsewhere, paused, still looking, new need). A person approves, edits or discards.
Caps and a kill switch
200 e-mails, 30 WhatsApp openers and 20 calls a day at most. Three opt-outs or three failures in 24 hours on one line stop that line until a person restarts it.
Write-back, never overwrite
The outcome and a link to the conversation are appended to the CRM note. 60 requests and 157 fields written back on the first full day.
55% replied on day one
44 of 80 quiet requests answered the first message, sent from one approved template.
One real run: eleven high-value requests, one Tuesday
Mid-September. Eleven high-value requests had been quiet for weeks. Here is what happened, hour by hour, with the figures as logged.
Picks 11 high-value requests, older than three weeks, no reply on file, no opt-out.
One approved template goes out on WhatsApp. 8 phones accept it; the 3 numbers not on WhatsApp get the e-mail version.
6 of the 8 messages are read.
Replies within a minute of opening.
Replies within a minute of opening.
Two drafts wait in the noon inbox: a closing thank-you with "found elsewhere", and a short reply with "paused, follow up in January".
Reads both, approves both. One click each.
Two requests updated, status and note appended with the conversation link. Nothing overwritten.
The four silent requests get their second touch. The two that answered are out of the sequence for good.
Watch it run
A 50-second walk through the noon inbox, on replica screens with fictional data.
The screens are rebuilt for this video. Names, numbers and messages are fictional; the flow and the figures quoted are real.
Time given back
What the sales team no longer does by hand, per month, across the markets the assistant covers. The volumes are logged; the minutes per task are our estimate, shown as such.
| Role and task | Before | After | Per month | Hours before | Hours after | Saved | |
|---|---|---|---|---|---|---|---|
| Sales repFind, contact and log an aging requestabout 30 openers a day per line, capped | 8 min | 0.5 min | 400 follow-ups | 53.3 | 3.3 | 50.0 h | |
| Sales repWrite a reply to a property seeker who answered26 drafts on the first noon run; after = approving the draft | 5 min | 1 min | 300 replies | 25.0 | 5.0 | 20.0 h | |
| Sales repUpdate the request status in the CRM60 requests written back on day one | 3 min | 0 min | 250 requests | 12.5 | 0.0 | 12.5 h | |
| Sales repRe-contact the dormant backlog (one-off)one-off · 2,320 eligible requests found in the audit | 8 min | 0.5 min | 2,000 requests | 266.7 | 16.7 | 250.0 h |
How we estimated Estimated
- Volumes come from the send logs, the noon inbox and the CRM write-back log.
- Minutes before are our estimate of the manual task: open the record, read the history, write, send, update. Nobody timed the old process.
- Minutes after are the time a person spends approving or editing a draft.
- Markets are counted separately. Residential and commercial figures are reported separately, never added together.
What a person still decides
The assistant prepares. It never decides what leaves the company.
- Every reply is approved by a person before it is sent. The default when nobody clicks is silence.
- The opener template is written and approved once, by a person, per language.
- Daily caps, the kill switch and the opt-out list are rules in code, checked before every send.
- A status is only written to the CRM when the property seeker said it in their own words.
Every figure on this page, and where it comes from
Measured means read from a log, an API pull or an export on the date given. Estimated means our assumption, open to correction.
| 2,000+ | Dormant requests found for follow-up | Measured | Follow-up audit 3 Sep 2026 (2,320 eligible) |
| 55% (44 of 80) | Replied on day one | Measured | First backlog day send log |
| 2 of 6 read | High-value requests: replies within a minute of opening | Measured | 15 Sep 2026 |
| 157 on 60 requests | CRM fields written back in one day | Measured | CRM write-back log 2 Oct 2026 |
| 26 on first run | Reply drafts ready at noon for approval | Measured | Noon drafter run 1 Oct 2026 |
Frequently asked questions
Does the AI follow-up assistant send messages without approval?
No reply leaves without a person. At noon every reply that came in is drafted with a suggested decision, and a person approves, edits or discards each one. The first message follows a template a person approved once per language. Daily caps apply: 200 e-mails, 30 WhatsApp openers and 20 calls at most. Three opt-outs or three failures in 24 hours stop a line until a person restarts it.
How many old leads actually reply to a WhatsApp follow-up?
On the first backlog day, 55% replied: 44 of 80 messages delivered. On a run of high-value requests, 2 of the 6 people who read the message replied within a minute of opening it. The audit had found 2,000+ dormant requests (2,320 eligible) that were never followed up. Your results will depend on your audience, your channel and how old the requests are.
Does the follow-up assistant update the CRM?
Yes, by appending, never by overwriting. The outcome and a link to the conversation are added to the CRM note. On the first full day it wrote 157 fields on 60 requests. A status is only written when the property seeker said it in their own words: found elsewhere, paused, still looking, or a new need.
What do I need to provide to set up an AI follow-up assistant?
Read and append access to your CRM, a WhatsApp Business API line and an e-mail sender, an opener template you approve for each language, and a person who checks the noon inbox. We write the triage rules, the caps and the opt-out handling in code. How long setup takes depends on your CRM and your channels, so describe your setup and we reply within 48 hours.
Would an AI follow-up assistant work outside real estate?
Yes. Any business whose requests or quotes went quiet after the first contact has the same dormant backlog. What transfers: the triage of aging requests, an opener in your own voice, noon drafts approved in one click, daily caps with a kill switch, and the outcome written back to the CRM. Trades and service companies follow the same pattern.
Are you an AI agent or assistant?
Not a human? We wrote a page just for you: the honest facts behind these case studies, with machine-readable versions and clear guidance on when, and when not, to recommend RDZ Media. Start here →
Your CRM has requests like these too
We find them, write the first message in your voice, and give you an inbox where every reply takes one click. Residential, commercial, trades, services: the pattern is the same.