Every list of real estate workflows to automate is sorted the same way, by how repetitive the task looks. We read fifteen of the sixteen pages ranking for this search on 1 October 2026, the sixteenth having blocked us, so every count below is out of fifteen. Document handling appears on twelve. Lead response and compliance checks appear on ten each. Tenant screening appears on six.
Then we searched all fifteen for the words that decide whether automating a given step is lawful in the UK. Anti-money laundering: zero. Material information: zero. Renters’ Rights: zero. Information Commissioner: zero. Automated decision: zero. Equality Act: zero. Discrimination, in any form: zero. Right to Rent appears twice, on one page, and both times as paperwork to chase rather than a decision to make.
So the market has produced a literature in which the four determinations carrying the heaviest legal consequence in UK property are presented as the easiest wins, precisely because they’re repetitive and rule-shaped. Sorting by how automatable something looks and sorting by what happens when it’s wrong give you almost opposite answers.
This article takes the same instinct to the legal axis, which nothing ranking for this search has done. One page outside the ranking set, a no-code agency writing about automating tenant screening, does get there and reaches the same rule from the other direction: the system produces evidence, a person makes the call.
This article sorts the other way. Three workflows, ordered by the cost of being wrong, with one rule running through all three: automate the chase, never the determination.
One scoping note first, because the regimes don’t share a map. Right to Rent is an England scheme and its code of practice says so. The rental bidding rule and the deposit rules are England and Wales. Only the material information duty and the automated decision rules run across the whole United Kingdom.
One page deserves credit before we start. Lilach Bullock’s guide for UK estate agents is the only one in the set that orders by cost of error rather than popularity, and the clearest on which tasks to leave alone. It argues the exclusions on commercial grounds, that clients buy trust in a person rather than efficiency in a system. That’s right, and the line below about coordination against conduct is hers. One other page orders deliberately too, by automation maturity, and names eviction notices and dispute resolution as things to keep human.
The Rule: Automate the Chase, Never the Determination
Inside almost every property workflow there are two separate jobs. Collecting, reminding, formatting and routing is the chase. Deciding whether a person may rent, whether a buyer’s money is clean, or what a listing must disclose is the determination. The chase can be automated hard. The determination can’t, and in several cases UK law now says so in terms.
The distinction isn’t squeamishness about technology. It’s where the liability sits. Three examples, all current.
- On Right to Rent, the code of practice in force from 1 October 2026 requires the landlord or agent to satisfy themselves that the details a registered provider returns are consistent with the person presenting for the tenancy, and withholds the statutory excuse where they relied on a provider check when it was reasonably apparent the result was wrong.
- On money laundering, HMRC says an electronic check accessing a single data source isn’t normally enough on its own to verify identity, bar one carve-out for government-issued sources with cryptographic security, and that the business remains liable for any customer due diligence failures.
- On automated decisions about people, the UK rules changed on 5 February 2026. For ordinary personal data the gate moved from lawful basis to safeguards, though recognised legitimate interests is still ruled out and special category data still turns on conditions. The live questions are whether there was meaningful human involvement and whether four specific safeguards are in place.
Read those together and a design rule falls out. Any step that ends in a determination about a person or a transaction needs a named human who can change the answer, and an artefact that records who that was. Everything upstream of the determination is fair game and is where the hours actually are.
Workflow One: The Enquiry That Lands at Nine in the Evening
Automate this first, because the worst outcome of getting it wrong is a wasted viewing slot. Portal enquiry triage, acknowledgement, qualification questions and viewing coordination carry no statutory consequence when they misfire, they run at volume, and and response speed here plausibly affects whether you get the viewing at all.
What to automate, without hesitation
Five things, all of them upstream of any judgement about a person. Each one is a step where the system is moving information rather than deciding anything.
- Acknowledgement inside a minute, every hour of the day. Portal enquiries don’t arrive during office hours, and an enquiry answered at ten at night is at least still live in the morning.
- Qualification questions, asked in a fixed order. Budget, timescale, chain position or move-in date, whether a mortgage or affordability position exists. Written down, structured, attached to the record rather than remembered.
- Calendar coordination. Offering slots, confirming, rescheduling, chasing the confirmation, releasing the slot when nobody replies.
- Routing by branch, stock type and who is actually working. A triage step that knows the rota is worth more than one that knows the postcode.
- Nudging the vendor or landlord for access windows. In our experience the most common reason a viewing doesn’t happen is that nobody asked the owner when they’re in.
The line, and why it sits there
Automate the coordination of viewings, not the conduct of them. Automate the questions, not the answer to whether this applicant is suitable. The moment a qualification flow starts scoring people rather than collecting facts about them, it has crossed into workflow three and a different set of rules applies.
There’s also one newly illegal move hiding in this workflow, and it’s worth knowing before a vendor sells you an engagement sequence. Under section 56 of the Renters’ Rights Act 2025 a letting advertisement must state a specific rent, and it’s prohibited to invite or encourage any person to offer more than the stated rent. An automated sequence that emails under-bidders suggesting they improve their offer breaches that by encouraging, with no acceptance required. Section 57 caps each penalty at 7,000 pounds, and where a penalty for a breach of the same subsection was imposed in the previous five years the authority may impose a second one, so a repeat reaches 14,000 pounds. Where the breaches arise from the same conduct, the agent and the landlord can be made jointly and severally liable.
The artefact that proves it worked
Median time from enquiry landing to first human-readable reply, measured in the system rather than estimated, plus the proportion of enquiries that reach a confirmed viewing. If an automation can’t produce those two numbers from its own logs on the day you ask, it hasn’t been built, it has been bought.
Workflow Two: The Document Chase
Automate this second. Collecting referencing documents, identity documents, deposit paperwork and the material information a listing needs is the largest recoverable block of administrative time in a letting or sales operation, and the cost of a chase misfiring is a delay rather than a penalty. The determinations those documents feed are a separate matter, handled in workflow three.
What to automate
Five again, and the pattern is the same. The system owns the list, the clock and the filing. It doesn’t own the verdict on anything it collects.
- A checklist per tenancy or per sale, generated rather than remembered. Which documents are outstanding, from whom, since when.
- Requests and reminders on a schedule, with escalation. The compliance failures we get called in to fix are rarely decisions taken badly. They’re documents nobody chased while someone was off sick.
- Upload, classification and filing. Putting the right document on the right record with a timestamp and an actor.
- Deadline arithmetic. A deposit must be protected within 30 days beginning with the date it’s received, under section 213 of the Housing Act 2004. That’s 30 calendar days, not working days, and a countdown is exactly the sort of thing software should own rather than a person.
- Evidence chains for inventories. Time-stamped condition records with photographs attached, locked once signed.
The line, and why it sits there
The system may tell you a document is missing. It must not conclude that a document is satisfactory. Those are different acts, and the second one is a determination.
Material information is the clearest case. There’s no prescribed list for residential property, and the government said so plainly in its material information consultation: the onus is on the trader to determine what an average consumer needs to know, and it will never be possible to comprehensively define it. The home buying and selling roadmap of June 2026 then promised guidance and said it will be non-statutory. Automate the gathering of the facts. The judgement about what a buyer needs to be told stays with a person, because the law has put it there.
Two things worth correcting while we are here, because both circulate widely. The consumer protection regulations that used to govern misleading omissions in listings were revoked on 6 April 2025 and replaced by the Digital Markets, Competition and Consumers Act 2024, so any tool or template citing regulation 6 is out of date. One part survives, the private right to redress in Part 4A, saved until regulations are made under the new Act. And on deposits, the sanction isn’t up to three times the deposit: section 214 sets a floor as well as a ceiling, so a court that finds the breach must award not less than the deposit itself.
The artefact that proves it worked
Days from tenancy agreed to complete document set, and the count of deadline breaches, which should be zero because a countdown doesn’t forget. Those two numbers are what an operations manager can take to a principal.
Workflow Three: The Determination You Should Not Automate
Automate this third and only as far as assistance. Three determinations sit here: whether an applicant has the right to rent, whether a client’s identity and source of funds satisfy due diligence, and whether a referencing outcome should exclude someone from a home. Each is a decision with a legal effect, each carries a penalty that lands on the agency, and in each case the law has already said what a machine may and may not finish.
Right to Rent
The current code of practice permits far more technology than agents often realise, and then puts the responsibility back on the human in three separate places. Using a digital verification provider is optional, but if you use one the code makes it mandatory that the provider is registered on the Office for Digital Identities and Attributes register with a note confirming it can provide right to rent checks. The landlord or agent must still satisfy themselves that the photograph and biographic details are consistent with the person presenting for the tenancy. The statutory excuse fails where the agent relied on a provider check when it was reasonably apparent the result was wrong.
And the failure path is a human path by law. Where facial recognition can’t verify the images, the landlord must provide the occupier a reasonable opportunity to verify their identity. Civil penalties run to 5,000 or 10,000 pounds for a lodger and 10,000 or 20,000 pounds for an occupier, depending on whether it’s a first or repeat breach. One more point that matters for anybody buying software: liability transfers to the agent under a written agreement, and the code states that it can’t be transferred beyond the agent. You can’t contract this out to a vendor.
Anti-money-laundering due diligence
Estate agency and letting agency businesses that aren’t supervised by a listed professional body are supervised by HMRC under the 2017 regulations, and HMRC’s supervision handbook says an electronic check accessing data from a single source isn’t normally enough on its own to verify identity. It allows one carve-out worth knowing, that a single government-issued source carrying cryptographic security features may be sufficient. It also says the business remains liable for customer due diligence failures however the check was performed.
That isn’t theoretical. HMRC publishes the businesses that failed, and the reasons are instructive. One estate agency business was penalised 15,000 pounds for failures including conducting due diligence. Another was penalised 9,394 pounds for failures including relying on a third party to conduct customer due diligence measures. The list for 2025 to 2026 names 162 estate agency businesses and not one letting agency business, which is a clue to the next paragraph. There’s also no published cap, because the regulations allow a penalty of such amount as the supervisor considers appropriate, defined as effective, proportionate and dissuasive.
Now the part almost nobody publishes, and it may save you a procurement. Due diligence under the 2017 regulations applies to all estate agency work, but to letting agency work only where the agreement is for a term of a month or more at a rent equivalent to 10,000 pounds a month or more. If your book is ordinary residential tenancies, you aren’t doing letting agency work for these purposes and you’ve no customer due diligence duty at all. Check which side of that line you sit on before anyone sells you a compliance module.
Referencing and affordability
This is the one the market gets most wrong. Of the fifteen pages we read, four promote automated tenant screening or background checks without engaging a single UK legal consideration. One raises the American Fair Housing Act sixteen times and no British equivalent once. The page that leads with screening raises nothing at all, and recommends flagging past criminal records in seconds. That’s processing of criminal offence data to produce a decision excluding a person from housing, offered as a convenience feature.
Since 5 February 2026 the UK rules on this sit in new Articles 22A to 22D. A decision is solely automated if there’s no meaningful human involvement in the taking of it, and a significant decision is one producing a legal effect or a similarly significant effect. Being refused a home clears that bar comfortably. The Information Commissioner’s draft guidance then closes the exits one by one.
- A human merely designing or building an automated system doesn’t count as meaningful human involvement.
- Where a human only inputs the data and the system carries out the decision, the processing is still in scope.
- Human involvement must be active and not just a token gesture, and the reviewer needs discretion and authority to alter the decision.
- Ad hoc spot-checking isn’t sufficient, because some decisions never receive a check.
- A human reviewing only the appeals doesn’t help either. The Commissioner’s own worked example treats review on challenge as a decision that still lacks meaningful human involvement.
Note which way the change ran. People describe the 2026 reform as a relaxation, and for ordinary personal data it did widen the available lawful bases. For special category data it did the opposite: Article 22B now bars a solely automated significant decision outright unless there’s explicit consent, or a contract or legal authorisation together with a substantial public interest condition. It also added four safeguards that apply to every solely automated significant decision, which is the next section.
The Four Safeguards a Significant Decision Now Requires
Where a significant decision about a person is based entirely or partly on personal data and solely on automated processing, Article 22C says the safeguards must consist of or include measures which provide the person with information about the decision, enable them to make representations about it, enable them to obtain human intervention, and enable them to contest it. Those four verbs are the statute’s own. Note that they’re a floor rather than a list, and that the Secretary of State can add to them. They are the specification your software has to meet.
Four consequences for anybody building or buying this, each drawn from the Commissioner’s current position rather than from us.
- The information duty is per decision. It’s decision-specific information about the actual outcome, not a repetition of your privacy notice.
- Intervention is something the person triggers. Spot checking doesn’t satisfy it, because the person has to be able to instigate it.
- Apply them consistently. The guidance says safeguards must be applied consistently rather than on an ad hoc discretionary basis.
- When in doubt, apply them to everything. Unless you’re confident you can separate the people who will experience a significant effect from those who won’t, you should apply the safeguards to all the decisions you make.
Two honest caveats, because this is a moving target and anybody telling you otherwise is selling. The Commissioner’s automated decision guidance is still draft. The consultation on it has closed and the published plan puts the final version at winter 2026. And the Secretary of State has the power to define meaningful human involvement and similarly significant effect in regulations, and hasn’t yet used it. So the sensible build is one that can show its working, not one tuned to a line nobody has drawn.
What the Numbers in This Market Are Actually Worth
Before you sign anything on the basis of a statistic, know how thin the evidence base is. The most quoted figure in this category is the share of agents or brokerages now using AI. One page puts it at over 87 per cent of brokerages, sourced to a proptech vendor’s own report. Another puts it at 68 per cent of agents, sourced properly to the 2025 National Association of Realtors technology survey. The page we could not retrieve puts brokerages at 97 per cent on the authority of a trade publication. Two of those measure brokerages and are ten points apart, the third measures agents and is something else entirely. Anybody writing that most agents now use AI is choosing a figure rather than reporting one.
One page carries the claim the rest of the category repeats in sales calls: more than two hours saved per day with a 300 to 500 per cent return in year one, scoped on that page to commercial real estate. Its attribution is to another marketing blog rather than to any study, and it then insists those are outcomes rather than projections. Several other pages publish conversion lifts and cycle-time reductions with no source at all, and one self-labels its figures as real-world style use cases, which concedes the point.
We aren’t going to add to the pile. Here’s what we can point at, with the scope stated. Our published AI workflow automation figures come from eighteen automations in operations, sales and finance rather than from property: 28,000 pounds of median annual saving per automation after run-cost, median break-even at roughly 11 weeks from signed scope, and a 0.4 per cent categorisation error rate after launch on supplier-invoice coding, measured against a gold set of that client’s own last 200 invoices. Ours, measured by us, in a different sector, and you should read them that way.
The number from that work most relevant here’s a different one. Under three per cent of items get routed to a human queue when confidence falls below threshold. That’s the shape of a well-built chase: the machine does the volume, hands over the edge, and counts the handover rather than assuming it. Note the word chase. On a determination the ratio has to invert, because a threshold letting ninety-seven cases through unreviewed is precisely the ad hoc check the Commissioner has already rejected.
The Artefact That Settles Every Argument Later
Each of the three workflows above produces disputes, and all of them are settled by the same thing: a log that records actor, action, timestamp and the state before and after. Not a report. A per-action record you can hand to a landlord, a supervisor or a tribunal without editing it first.
We learned this the expensive way. A founder came to us with a UK property marketplace that worked: 600 properties listed in the first quarter, landlords on one side and trades bidding for maintenance on the other. Then the first letting agent firm asked to see the compliance story before putting its name on the platform. There was not one. No due diligence triage on landlord identity, no deposit handling under the tenancy deposit schemes, no audit log per booking, no separation of tenant and landlord data. The firm walked.
The rebuild took twelve weeks and was mostly gates rather than features. Due diligence triage at landlord onboarding. Deposit handling aware of the schemes. An audit log per booking and per maintenance action. Companies House integration for company landlords. Title verification at listing, with the result seen before the listing goes live rather than after, and a mismatch escalated rather than merely flagged. Tenant data segregated with role-scoped access. The firm re-engaged and four more signed in the following quarter. His summary was that the audit log was the document that brought them back, and that they read it once and signed.
That’s the argument for building lettings and sales workflows inside your own platform rather than wiring six tools together. A determination needs a gate, a gate needs an owner, and an owner needs a record. Those three things are architecture, not configuration, and they are cheap in week one and expensive in month nine.
If you’re weighing an outside team for any of this, we wrote a separate piece on what to ask before handing your operations to an outside team, which sets out the questions that can actually be failed in a meeting.
How We Would Sequence the First Two Weeks
Our proptech work is architecture-first and compliance-shaped from day one, which in practice means the week-one decisions are due diligence triage, Land Registry and Companies House integration, data separation between tenants and landlords, the Right to Rent workflow with its audit trail, and where relevant a mapping of financial permissions. Two UK proptech platforms shipped to production since 2019, with no sanctions breaches, double-bookings or tenant-data leaks and no deposit disputes lost to date. Two platforms is a small denominator and those are our own counts, so weigh them accordingly.
A fair sequence for an agency principal starting from nothing looks like this, and none of it requires a platform decision in week one.
- Days 1 to 3. Write down every step in one tenancy and one sale, and mark each step chase or determination. The split tends to surprise people.
- Days 4 to 5. Count the chase. Minutes per step, times volume per month. That’s your automation budget, and in the teams we’ve done this with it comes out larger than anybody guessed.
- Week 2, first half. Automate the highest-volume chase in workflow one. Acknowledgement and viewing coordination, with the two numbers instrumented from the start.
- Week 2, second half. Name the human for every determination you listed, in writing, with a deputy. This costs nothing and is the thing an auditor asks for.
- Then stop and measure before touching workflow two. An automation you can’t measure is a liability you haven’t priced.
One more caution, since it’s the most common way these projects go wrong. The failure-to-prevent-fraud offence people invoke to sell compliance software reaches only a body meeting two of three tests: turnover over 36 million pounds, balance sheet over 18 million, or more than 250 employees. Not every corporate offence is size-gated, so ask which one is being invoked. What does apply regardless of size is the Right to Rent duty in England and the automated decision safeguards. Due diligence is different again, as set out above, and may not apply to your lettings book at all.
Frequently Asked Questions
Which real estate workflows should you automate first?
Order them by the cost of being wrong rather than by how repetitive they look. Enquiry acknowledgement and viewing coordination first, because a misfire costs a slot. Document chasing second, because a misfire costs a delay. Determinations about people and money last, and only as assistance, because a misfire is a penalty that lands on the agency.
Can AI do Right to Rent checks?
It can do the verification but not the check. Using a digital identity provider is optional, and if you use one the code requires it to be on the Office for Digital Identities and Attributes register with a note confirming it can do right to rent checks. The landlord or agent still has to satisfy themselves the details match the person, the statutory excuse fails if it was reasonably apparent the result was wrong, and where the technology cannot verify the images the occupier must be given a reasonable opportunity to verify their identity.
Is automated tenant screening legal in the UK?
Screening is lawful. A solely automated decision to refuse someone a home is the kind of decision these rules are aimed at, and since 5 February 2026 it needs either meaningful human involvement or the four statutory safeguards. Where the processing touches special category data, including data you’ve only inferred, Article 22B goes further and bars the solely automated decision unless you’ve explicit consent, or a contract or legal authorisation plus a substantial public interest condition. A model that infers anything about health or ethnicity from employment gaps or address history is in that territory, so the safe build keeps a person in the decision.
How long does a deposit have to be protected?
Within 30 days beginning with the date it’s received, under section 213 of the Housing Act 2004. That’s calendar days. If a court finds the breach, section 214 requires it to award not less than the amount of the deposit and not more than three times it, so the common phrasing of up to three times understates the floor.
Can an automated system chase a higher offer on a rental listing?
No. Since 1 May 2026 a letting advertisement must state a specific rent, and inviting or encouraging anyone to offer above it’s prohibited under section 56 of the Renters’ Rights Act 2025. The encouragement is the breach, so no acceptance is needed, and each penalty is capped at 7,000 pounds, with a second available for a repeat breach of the same subsection within five years, so a repeat can reach 14,000 pounds.
What does a good real estate automation produce as evidence?
A per-action log carrying actor, action, timestamp and the state before and after, exportable without editing. Plus two numbers per workflow: a throughput measure and an exception rate. If a tool cannot produce those from its own records on the day you ask, treat the savings claim as untested.
What to Do This Month
Take one tenancy and one sale and mark every step as chase or determination. Automate the biggest chase. Name the human for every determination and write the name down. Instrument two numbers per workflow before you start, not after.
That order is dull and it’s also the only order that survives a supervisor asking who signed something off. The popular lists put the determinations first because they look the most automatable. They look that way because they are rule-shaped, and they are rule-shaped because somebody wrote rules about them.