Automate buyer and vendor follow-up by setting CRM triggers for the moments that matter (after a viewing, after an offer, weekly for every vendor, and when an applicant goes quiet), then letting AI draft each message from your notes and feedback. Buyer nudges can send automatically; vendor updates and anything about offers wait for a negotiator's approval. Start with post-viewing feedback.
Start there because viewing feedback feeds both sides at once: the buyer's next step and the vendor's weekly update. Most agencies' follow-up doesn't fail on wording; it fails on data. If feedback lives in a negotiator's head or a scribbled diary note, AI has nothing to personalise, and the "automated" messages come out as generic as the ones they replaced. Fix capture first and the rest gets easy.
Map every follow-up moment before automating one
List the moments where someone should hear from you, who it is, and how much a mistake would cost. That third column decides whether AI sends, drafts for approval, or stays out. A filled-in map for a typical sales agency:
| Trigger | Recipient | AI's role | Send mode |
|---|---|---|---|
| Viewing booked | Buyer | Confirmation with time, address, parking note | Automatic |
| Two hours after viewing | Buyer | Feedback request, referencing the property | Automatic |
| No feedback after 24 hours | Buyer | One polite chaser, then stop | Automatic |
| Feedback logged | Vendor | Tactful summary of the viewing | Negotiator approves |
| Every Friday | Vendor | Weekly marketing update | Negotiator approves |
| New instruction matches criteria | Applicants | Match email explaining why it fits | Automatic, with consent |
| Applicant silent for 21 days | Applicant | Re-check what they are looking for | Automatic, once |
| Offer received or rejected | Both | None beyond logging | Negotiator only |
| Sale agreed, weekly | Buyer and vendor | Progression status from milestone notes | Negotiator approves |
| Sale falls through | Vendor | None | Negotiator phones |
Offers stay human because every word can move a negotiation, and fall-throughs stay human because a vendor who has just lost their buyer should hear a voice, not read a template. Everything else can have AI in it somewhere.
Capture viewing feedback in sixty seconds
The single highest-value change is making feedback capture effortless for the negotiator. Instead of typing notes back at the office, they record a voice note in the car straight after the viewing. Transcription is built into most phones and many CRMs' mobile apps, or you can use an AI note-taker. The transcript then goes through a prompt that structures it into the fields your CRM holds.
From this voice note after a viewing, fill in these fields. Use only
what the note says; write "not mentioned" for anything missing.
Property ref:
Buyer name(s) as said:
Interest level (1-5, with the phrase it's based on):
Liked:
Concerns:
Comments on price (quote exactly, or "not mentioned"):
Buyer's position (chain, mortgage, cash) if mentioned:
Next step agreed:
Anything the vendor must NOT be told:
Voice note: [paste]An illustrative result, from a 40-second voice note about a flat viewing:
Interest level: 4 ("they asked about the service charge twice and
measured the second bedroom")
Liked: light in the living room, the size of the second bedroom
Concerns: kitchen feels dated; worried about noise from the main road
Comments on price: "they said it's a bit toppy compared to the one on
the next street"
Buyer's position: first-time buyers, mortgage agreed in principle
Next step agreed: second viewing Saturday with a parent
Anything the vendor must NOT be told: they'd go to asking if the
vendor leaves the white goodsWhat the negotiator checks: that the interest level matches their own read (the AI gives a 4 because of the measuring, which is fair), and that the last field is right, because the whole approval step exists to stop that line reaching the vendor. The "quote exactly" instruction on price matters too. "A bit toppy" is useful to the negotiator; paraphrased by AI as "the buyers felt the price was too high", it becomes a stronger statement than the buyer made.
Vendor updates that are honest without being brutal
Vendors care about two things: is anything happening, and what are people saying. AI is good at turning raw feedback into something a vendor can hear. A before and after, using the feedback above. What a rushed negotiator might forward:
Viewing feedback: kitchen dated, road noise, think it's overpriced vs the one on the next street.
The AI draft, after a prompt asking for a tactful, specific and honest summary with no negotiating information:
Saturday's viewers, first-time buyers with a mortgage agreed in principle, were positive about the living room light and the size of the second bedroom, and they've booked a second viewing this Saturday. Their reservations were the kitchen's age and some traffic noise from the main road. They also compared the asking price with another flat nearby, which is worth discussing when we speak on Monday.
Same facts, no spin, and the price comment is raised as a conversation rather than dropped as a verdict. The white-goods remark is absent, as it should be. The negotiator's approval takes thirty seconds: read, check the "must not" field hasn't leaked, send.
The weekly vendor update works the same way. Feed it the week's viewings, feedback summaries, enquiry numbers from your CRM or portal reports if you have them, and any marketing changes. Ask for five short sentences: activity this week, what viewers said, what we're doing next, one thing we'd like to discuss, and the next update date. When the honest message is that the price needs revisiting, the AI can prepare the data, but that conversation happens on the phone.
An illustrative Friday draft for the same flat, and the one line the negotiator had to change:
This week: three viewings (Tuesday, Thursday and Saturday) and 40%
more portal enquiries than last week.
What viewers said: all three liked the living room light; two raised
the kitchen's age, and one mentioned road noise.
Next: the Saturday couple return for a second viewing with a parent.
To discuss: how the asking price compares with the flat on the next
street, which two viewers mentioned.
Next update: Friday, as usual.The "40% more enquiries" was wrong. The AI had divided a seven-day portal figure by a figure covering the four days since the listing went live. A vendor who hears their enquiries rose 40% will quote it back to you when they fall. Either paste numbers for matching periods, or tell the prompt to report raw counts and never calculate percentages. Raw counts ("11 enquiries this week, 8 last week") are harder to misread anyway.
Keeping buyers warm without flooding their inbox
Applicants go cold because agencies either contact them too little or bombard them with every new listing. The first message they get after a viewing sets the tone, so keep it short and easy to answer. An illustrative feedback request, sent two hours after the viewing:
Subject: 14 Elm Road viewing today
Hi [first name], thanks for viewing 14 Elm Road this afternoon.
A one-word reply is plenty:
1 - We'd like a second viewing
2 - Still thinking; happy to chat
3 - Not for us (a word on why helps us find the right one)The AI's first draft of this template opened with "We hope you loved the stunning open-plan kitchen!", about a buyer whose note said the kitchen felt dated. Keep property features out of the automatic request altogether; the negotiator's voice-note summary is where the buyer's own view lives. One edge case needs a rule too. A buyer who views three properties on a Saturday shouldn't get three separate requests at two-hour intervals. Group viewings by applicant and day, and send one message listing each address with the same three options.
Two automations then fix most of the cold-applicant problem:
- Match emails that explain the match. When a new instruction fits an applicant's saved criteria, AI writes one sentence on why: "You mentioned needing parking for two cars; this one has a driveway and a garage." Ten precise matches beat forty loose ones. If your listing copy comes from a proper fact sheet, as in writing property descriptions with AI, the match sentence is only as accurate as that sheet.
- A re-check after silence. After 21 days with no response, one message asking whether their search has changed, with three quick options to reply with. Their answer updates the criteria, which improves every later match.
The re-check only pays off if each reply changes something in the CRM. An illustrative mapping for a three-option message:
| Reply | What the applicant means | What the automation does |
|---|---|---|
| "1" or "still looking" | Same search, keep sending | Resets the 21-day timer; no other change |
| "2" or "changed" | New area, budget or size | Creates a call task for a negotiator within one working day |
| "3" or "bought / paused" | Stop for now | Sets status to inactive and stops all match emails |
| Anything else | Unclear | AI summarises the reply into the record; a person decides |
The last row matters more than it looks. Replies such as "we've had an offer accepted on ours, so we're serious now" are the best leads you'll get all week, and a rule that only reads "1", "2" or "3" would file them as unclear and forget them.
Set a frequency cap (for example, no more than two automated messages a week per applicant) and honour opt-outs immediately. Marketing emails to applicants need their permission under data-protection law in most places, so check your consent records before switching anything on, and tidy the records first if they are messy; cleaning customer records before adding AI covers the de-duplication that stops one applicant getting three copies of every email.
A realistic mistake worth designing against: an automation sends a "new match" email for a property that went under offer that morning, because the status change hadn't synced yet. Three applicants call to book viewings that can't happen. The fix is a short delay (send match emails at 5pm, not instantly) and a status check as the automation's last step before sending.
Sales progression chasers after the offer is agreed
Between sale agreed and completion, both sides mostly want to know that someone is watching. Keep a milestone list per sale (memorandum sent, searches ordered, mortgage valuation, survey, mortgage offer, enquiries raised and answered, exchange date proposed) and have the negotiator or progressor tick them off. Each Friday, AI turns the ticks and any notes into a short status update for the buyer and a separate one for the vendor, drafted for approval. It can also draft chasers to the buyer's and seller's conveyancers when a milestone has been stuck for more than a set number of days. For how conveyancers are automating their side, see the further reading below.
Keep the chasers factual: which milestone, since when, what you need. AI drafts tend to become apologetic or pushy depending on the example you gave them, so include a sample chaser in your tone in the prompt. A before and after for a search that has been outstanding for 16 days:
- AI draft: "So sorry to bother you again, and we completely understand how busy things are, but we just wondered if there might be any update at all on the searches when you get a moment?"
- Edited: "Searches for 14 Elm Road were ordered on 2 September. Could you confirm whether they're back, or the expected date? The vendor has asked for a timeline by Friday."
The edited version gives the milestone, the date and the reason for asking, so the conveyancer can reply in one line.
Wiring it together: CRM first, then Zapier or Make
Most agency CRMs have their own triggers for viewings, feedback and vendor reports. Use those before anything else, because they already know your properties, applicants and statuses. Add a general automation platform only for the gaps, typically the AI drafting step and anything that crosses between systems.
- Zapier: Professional costs $19.99 a month on annual billing for 750 tasks. Each successful action step is a task; triggers and filters are free. An AI by Zapier step uses 1, 3 or 5 tasks per run depending on the model tier you pick.
- Make: free up to 1,000 credits a month, with paid tiers starting around $9 monthly; routers and filtered-out bundles use no credits.
- HubSpot: if you run your applicant marketing there, note that the full workflow editor needs a Professional plan; Starter has simpler automations attached to forms and emails.
Count tasks before you choose a plan, because AI steps multiply them. Take one viewing's journey through Zapier: an AI step to structure the voice note (1, 3 or 5 tasks depending on the tier), a step writing the fields into the CRM (1 task) and a step creating the vendor draft (1 task). That's 3 to 7 tasks per viewing. At 45 viewings a week, roughly 195 a month, the viewing flow alone uses about 585 tasks on the cheapest AI tier and about 1,365 on the most expensive. The first fits inside Professional's 750; the second doesn't, before you've added a single match email. Test whether the cheapest tier structures your voice notes well enough before paying for more tasks.
The safest pattern for approval-required messages is "draft into the inbox": the automation writes the email and saves it as a draft in the negotiator's Outlook or Gmail, and nothing sends until a person presses send. For buyer-facing automatic messages, run them as drafts for the first two weeks anyway, until you trust them. The general method in automating sales follow-ups without being pushy applies here too, and updating your CRM automatically after calls shows the voice-note-to-fields step in more detail.
An illustration: two offices, 60 live instructions
A hypothetical sales agency with two offices, six negotiators and about 60 properties on the market holds around 45 viewings a week. Before automating, the office manager checks a month of records and finds feedback logged for fewer than half of viewings, and vendor updates sent on no fixed day. Those two numbers become the baseline.
Month one: voice-note feedback capture and the buyer feedback request go live. Month two: vendor summaries drafted for approval, and the Friday update. Month three: applicant match emails and the 21-day re-check. Costs: a Zapier Professional plan and a business AI plan, well under $100 a month between them for this volume, with the CRM's own automations doing most of the triggering.
What to track, rather than assume: the share of viewings with feedback logged within 24 hours; whether every vendor gets their Friday update on Friday; buyer reply rates to feedback requests; the number of vendor calls asking "any news?" (which should fall); and unsubscribes from applicant emails (which should stay low). Offers per viewing is the number everyone wants, but it depends on price and market far more than on follow-up, so treat it as context, not proof.
Four follow-up errors that surface in the first weeks
- Wrong property, right person. Messages referencing the wrong address are the most common automation error, usually from a viewing rebooked on another property. Include the address in the draft subject line so it is visible at approval.
- Leaked negotiating information. Read every vendor summary for anything from the "must not" field. One leak can cost more than a year of saved time.
- Tone drift. After a few weeks, drafts may creep towards estate-agent clichés. Refresh the prompt's examples with messages your team actually wrote.
- Silent failures. An automation that stops sending doesn't announce it. Check each one's history weekly for the first month, then monthly.
Further reads
- How Much Does It Cost to Automate Lead Follow-Up With AI? — What these automations cost to build and run each month.
- How Estate Agents Use AI to Prepare Valuation Appointments — The follow-up that starts before a vendor even instructs you.
- How Conveyancers Use AI to Cut Admin on Each Transaction — How the other side of sales progression is using AI.
- How to Set Up Human Review for AI Work Without Slowing Down — Set up approvals for vendor updates without slowing them down.
- Signs Your Business Is Ready to Automate Sales Follow-Ups With AI — Check your data and process are ready before you switch anything on.
- WhatsApp Business App vs API: Which Do You Need for AI Replies? — If applicants prefer WhatsApp, which set-up AI replies need.
- Can an AI Chatbot Book Valuations for Estate Agents Out of Hours? — What an out-of-hours valuation bot needs: live diary access, eight qualifying questions, written diary rules and a list of things it must never say.
- AI Virtual Staging for Property Listings: Costs and Disclosure — Per-photo and per-listing costs of AI virtual staging, the edits a label can't excuse, label wording to copy, and a pre-publish check for every image.
- AI Listing Mistakes That Can Mislead Buyers, and How to Check — Check room counts, measurements, parking rights and altered photographs against evidence before publishing an AI-written property listing.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Zapier pricing and task-counting help pages; Make pricing page; HubSpot knowledge base on workflows by plan.