Estate Agent Software With AI: Questions to Put to Your Vendor

Before switching to estate agency software with AI features, check five things: whether the AI is built in or a bolt-on, where your data goes and whether it trains models, whether you can export everything, the true per-user price once AI is included, and whether each feature works on a demo script built from your own files.

Most comparison pages rank tools. That is the wrong question to start with, because a ranking cannot know your listings, your lettings split or your team size. A sheet of questions and a test script can, and they are cheap to prepare in an afternoon.

Follow me on Instagram@sagnikteaches

Built-in assistant or bolt-on: decide which you are buying

An AI feature reaches an agency in one of two ways. It is part of the CRM, the system that holds your applicants, properties and diary. Or it is a separate tool that connects to the CRM and reads or writes records through an integration.

Connect on LinkedInSagnik Bhattacharya

Reapit is a useful example of the first kind. Its AI and automation page, read in November 2026, lists AI property descriptions that agents can edit and add to the property record, automated applicant pre-qualification in its Bookings product, viewing slots offered to fit agents' travel and diaries, a home dashboard that organises each agent's day, AI-driven insights on agency data, and a chatbot assistant, Fi, that answers questions about the platform. It links to the announcement of an assistant called Reapit AI (RAI) and says AI agents are in development. The same page also promises more AI through partner integrations, which shows how blurred the line between built-in and bolt-on can be. The site publishes no price list (pricing runs through a demo request), and the page does not map every feature to a plan, so both are questions for the vendor, not assumptions. Treat Reapit as one example of the type, and ask every vendor the same things.

Subscribe on YouTube@sagnikteaches

Neither wins on its face. Built-in is easier to keep tidy. A bolt-on can beat it on a single task, and it keeps its value if you change CRM later. The questions below work for both.

The question list, grouped by what can hurt you

Data and privacy

• Which AI provider processes our data, and where? Ask for the sub-processor list in writing.
• Is our data, including applicant details and instructions, used to train models, yours or anyone's? Is that the default, and can it be switched off for the whole account?
• How long are prompts and outputs kept, and can we delete them on request?
• Does the assistant see all records, or only the ones the signed-in user may see? An assistant that ignores permission levels can leak a colleague's confidential notes.
• What is the data processing agreement, and can we read it before signing?

Accuracy and control

• When the AI writes something, who must approve it before it is saved or sent? Can we make approval mandatory?
• Does it show what it used to produce an answer, for example the fields or notes behind a summary?
• Can we turn individual AI features off for the whole agency or for specific users?
• How do we report a wrong output, and what is the response?

Leaving and lock-in

• Can we export contacts, applicants, properties, notes, documents and activity history? In what format, at what cost, and how long does it take?
• Is the AI-generated content, such as descriptions and summaries, included in the export?
• What notice and charges apply if we leave mid-contract?

Cost

• Is AI included in the per-user price, an add-on per user, or metered by usage?
• What counts as a user? Does a part-time administrator or a branch manager need a licence?
• Are there set-up, training or migration fees, and what happens to the price at renewal?
• Is any feature on a higher tier than the one we were quoted?

Support and roadmap

• What can we use today, as opposed to what is on the roadmap? Ask for a list split into live and planned, with dates for the planned items in writing.
• Who do we call when it breaks, and are those hours useful to us?

Print the list, take it into every demo, and write the answers down. Vendors who answer in writing, with page references to their own terms, deserve a longer look than those who say "we'll get back to you".

A per-user price sum built from your own written quotes

Some agency CRMs quote on request rather than publishing a price list; Reapit, for one, routes pricing through a demo request. So the comparison starts with written quotes, and the sum below is what you run once they arrive. The headline price is rarely the real cost, so work the sum for a whole year using your own headcount.

Every figure in the table is a placeholder chosen to make the arithmetic easy to follow. None is a vendor price, none comes from Reapit or any other supplier, and none tells you what to expect to pay. Replace each one with the number on your own quote. The layout assumes a twelve-person agency.

On those placeholder numbers Quote A comes out $2,500 lower in year one; on your real quotes the gap could be larger, smaller or the other way round. The method is what matters: count only the licences people truly need, add AI charges, add one-off costs, and run it again with the quote's renewal terms. Then ask what you are getting for the difference. If B's bolt-on does one job far better than A's built-in feature, a gap of that size may be a fair price. If not, it is a tax on complexity.

One more division makes quotes comparable with any headline price a salesperson mentions: first-year cost, divided by licensed users, divided by twelve. On the placeholders, Quote A works out at $13,380 / 11 / 12, about $101 per user per month once migration is spread over the year, against $15,880 / 12 / 12, about $110, for Quote B. Run the same division on your own quotes. If a vendor's real figure lands well above the per-user price you were shown in the demo, the difference is sitting in add-ons, set-up fees or licence counting, and the question list above will find it.

Build a demo test script from your own files

Demos are rehearsed. Replace their script with yours. Before the meeting, pick five real, anonymised situations from your last month and ask the vendor to run them live, in your presence, in the trial account or demo environment.

1. Property description. Give the assistant the details of a real instruction with one awkward feature, for example a flat with a shared entrance and a restrictive lease clause. Check whether it states the awkward fact or glosses over it. See how estate agents use AI to write property descriptions for what to look for.
2. Applicant matching. Describe a buyer with a hard requirement, such as step-free access. Check the shortlist for properties that fail the requirement.
3. Follow-up. Ask for a message to a vendor after a viewing with mixed feedback. Is the tone right, is anything promised that you did not say, and who must approve it before it goes? The buyer and vendor follow-up tutorial lists what a good result contains.
4. Permissions. Sign in as a junior user and ask the assistant about a record that user should not see. A correct answer is a refusal.
5. Export. Ask them to export ten properties and ten contacts with notes, now, and open the file.

Score each out of five on a sheet and write one line on what went wrong. The point is not a clever result. It is whether the feature handles an awkward case, shows its working and respects your rules.

What a failed test looks like

Say the description test returns "a bright, sunny home with generous storage" for a flat whose notes mention a north-facing aspect and a loft hatch only. That is a failure of accuracy: the output invented two selling points. A vendor who responds by saying "agents edit descriptions anyway" has told you the control is yours, so ask whether you can require review before the text publishes to a portal. The guide to AI listing mistakes that can mislead buyers shows how these slip out.

Test for lock-in before you sign, not after

The cheapest protection against a bad switch is an export you have actually opened. During the trial, import a copy of a subset of your data and export it back. Look for: missing notes, lost attachments, applicants collapsed together, activity history that no longer links to its property. Ask a colleague who did not run the test to open the file and tell you whether they could find a particular vendor's history from last spring.

Also read the termination clause with this in mind: how long after leaving can you still get your data out, and at what cost? A system where leaving is hard is a system where you cannot walk away from poor AI features either.

Check what the AI does with calls and enquiries

Some agencies add AI at the front door first: a website chat or a phone assistant that takes enquiries out of hours. If that is your goal, a bolt-on may serve you better than switching systems. The questions are the same, with two additions: does the assistant hand over to a person cleanly when asked, and does it write what it learned to the CRM record in a form your team can use? See the tutorial on booking valuations with a chatbot out of hours for the practical limits, and what to leave to people.

A shortlist method that takes one week

1. Day 1. List your top three jobs that waste time, for example descriptions, follow-ups and diary clashes. Write what a good result looks like for each.
2. Day 2. Ask your current vendor, in writing, which of the three it can do today, on which plan, at what price. You may find the answer is already yes.
3. Days 3 and 4. Book at most two other demos. Send them your five test situations beforehand and your question sheet, and ask them to follow it.
4. Day 5. Fill in the cost sheet for each, score the tests, and decide with two colleagues who will actually use the system.

If your current vendor can do the three jobs, the decision is made and you have saved a migration. If nobody can, the best answer may be to wait and re-ask in three months, since features change quickly and a vendor's roadmap list should be dated.

Two agencies, two different answers

Take a sales-led agency with eight staff. Its pain is slow follow-up after viewings and descriptions that take a negotiator twenty minutes each. Its current CRM vendor, asked in writing, says a description assistant is available on the plan it is already on, and follow-up drafting is on the roadmap. The sensible move is to switch the description feature on, run the demo script on it, and revisit follow-up in three months. No migration, no new contract.

Now take a lettings-heavy agency with fifteen staff, where most hours go on maintenance requests and tenant queries. Its CRM has nothing for either. A bolt-on that triages maintenance messages and answers tenant questions could pay off, but it must write back to the CRM record. Here the integration question outweighs the price: ask which fields it writes to, what happens when it cannot connect, and who is notified of a failed write. Switching the entire CRM for that one job would cost far more than the problem is worth, unless the CRM was already failing for other reasons.

The same checklist gives two opposite answers, which is a sign it is asking about your situation rather than the market in general.

Reading vendor answers: good, thin and evasive

Vendors rarely say no. They say things that sound like yes. Here is how three illustrative replies to the same question compare, "Is our data used to train models?"

Send the question list by email before the demo and ask for written answers within a week. A reply that arrives late and vague tells you what support will feel like after you have signed.

An email to send before any demo

Hi [first name], we are comparing systems and would like to see yours on our own scenarios. Could you answer the attached questions in writing before the demo, and confirm which AI features are live today, on which plan, and at what per-user price? We will send five anonymised test situations a few days before we meet. Please also confirm the export formats available for contacts, properties, notes and documents. Thanks, [your name]

This one message sorts vendors quickly. The ones who engage with it are the ones who will engage with your support requests.

A trial scoring sheet your team can fill in

During the trial, give each tester one sheet and ask them to score the same items. Use a plain scale of one to five, and ask for one sentence for any score below four. A small sheet beats a long one, because people complete it:

• Did the AI feature save time on a real task, in minutes, against doing it by hand?
• Was the output accurate without edits, with small edits or with heavy edits?
• Did it show what it used, or did you have to guess?
• Did it respect what this user is allowed to see?
• Would you use it again tomorrow without being asked?

The last question is the most honest. A feature that scores well but that nobody wants to use again will not survive the first busy month. Add up the scores per vendor, compare with the cost sheet, and keep the testers' sentences, which often decide the choice more than the totals do.

Getting the team ready if you do switch

If the decision is to move, protect the first fortnight. Agree a short list of AI features that will be used from day one, and leave the rest alone until people are comfortable. Name one person per branch as the person who answers "how do I" questions. Keep the old system read-only for a set period so history can be checked, and agree a date after which it is switched off. And decide the rule for AI-written text before anyone starts typing: for example, nothing generated goes to a vendor, buyer or portal without a person reading it first.

Ask what happens when the AI is wrong or unavailable

Every AI feature fails sometimes: the provider behind it has an outage, a model update changes the tone of its drafts, or a record is read wrongly. Ask the vendor three practical questions. What does a user see when the AI is down, and can work carry on without it? Are you told in advance when the underlying model changes, so you can retest your script? And is there a record of what the assistant wrote or changed, so a bad description or message can be traced and fixed? A vendor with sensible answers has thought about operations. One that has not will leave you finding out on a Saturday morning with a full diary.

Red flags worth walking away from

• No written answer on whether your data trains models.
• An export that exists only as screenshots or a PDF.
• Features sold as live that appear only on a roadmap slide.
• Pricing that arrives only after you have signed a pilot.
• An assistant that cannot be limited by user role.
• Pressure to sign this week for a discount that was available last week.

If you want to compare the same discipline on a different sector's software, evaluating AI features in recruitment software applies the same method, and if you would rather work through your own shortlist with someone, my AI implementation consultation is built for that.

Questions agency owners raise before switching

Should I switch CRM just to get AI features?

Rarely. Switching costs migration time, retraining and a risk of lost history. If your current vendor ships the AI features you want, or a bolt-on covers them, that usually costs less. Switch for core reasons, and treat AI as a tie-breaker.

Is an AI add-on in my CRM safer than a separate tool?

Often simpler, because the data stays in one system and one contract covers it. It is not automatically safer. Ask the same data questions of both: where the data goes, whether it trains models, and how you delete it.

How many agents should join the trial?

Two or three people who do different jobs, for example a negotiator, a valuer and an administrator. A trial with one enthusiast tells you about the enthusiast.

What should I never accept in a contract?

Any clause that stops you exporting your own contacts, instructions and history in a usable format, or that lets the vendor change per-user charges without notice.

Further reads

• AI or Human Answering Service: Which Suits a Property Manager? — Another buy-or-bolt-on decision, for calls.
• How Estate Agents Use AI to Prepare Valuation Appointments — A feature worth adding to your test script.
• AI for Letting Agents: Handle Tenant Queries Without Extra Staff — If lettings is a big part of your business.
• How Property Managers Use AI to Triage Maintenance Requests — Maintenance triage as a test case.
• AI Screening: Bolt It Onto Your ATS or Switch Systems? — The same bolt-on versus switch choice, in recruitment.
• AI Tools for Estate Agents: A Stack by Job, Valuation to Completion — A stage-by-stage AI stack for a small estate agency: valuation, listing, viewings, offers and progression, with costs and the checks each needs.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Reapit's AI and automation page and site navigation (features listed; a demo request in place of a price list), read 4 November 2026; general vendor-evaluation practice. Every cost figure in this tutorial is a placeholder, not a vendor price. Verify every feature and price with the vendor in writing.