The Lettings Process Map

How to read it

Three measurable properties decide what AI can do with a process — and a fourth, the verdict, falls out of them.

• Volume — events per year per 100 managed properties (the portfolio-agnostic base; ×1.5 for our 150-property model book, ×2.9 for Glenn Flegg’s ~290). High volume = where labour and cost concentrate.
• Risk — the cost of being wrong: 🔴 High · 🟠 Medium · 🟢 Low, tagged by flavour (safety / financial / regulatory / reputation) and whether the harm is recoverable or irreversible.
• Verifiability (τ) — time to ground truth. 0 = knowable now (a rule, an authoritative gate, or a self-evidencing output); slow = knowable eventually; ∞ = never cleanly knowable (contested judgement).
• Verdict: 🟢 A — AI delivers the outcome (τ ≈ 0, recoverable). 🔵 B — AI is a context-complete copilot; a human owns the outcome. ⚫ Physical — local human presence; AI coordinates but never performs.
• Outsource — who performs the work and who bears its cost. 🏠 In-house (central AI brain; near-zero cost, sitting in the management fee) · 👤 Part-time field staff (physical attendance needing no certificate) · 🔧 Outsourced specialist who carries the liability (signing trades, inventory clerk, accountant, solicitor, referencing bureau). Cost is then recharged to the landlord, absorbed in the management fee, or carried as firm overhead.

Rule of thumb: low τ + recoverable → A. Irreversible or contested → B. Must be at the property → Physical. Many processes split across all three.

Two added layers this revision: each area carries a second table — the process spec (trigger, data required, output, and the verification path: the concrete signal that it’s done) — and every regulated process’s licensing, rules and operating parameters are compiled in the Regulatory appendix.

Executive summary — the process index

Forty-four operating processes across eleven areas. Volume is events/yr per 100 managed properties. Reg marks a statutory/regulated process (28 of 44). Outsource: 🏠 in-house (AI) · 👤 part-time field · 🔧 outsourced specialist.

*AML applies conditionally — see On AML. For a pure sub-€10k/month lettings book it is not a statutory duty at all. (40: PRS Database not yet live — Phase 2.)

The map

For each area: the analysis (volume · risk · τ · verdict & build) followed by the process spec (trigger · data required · output · verification path).

1 · Landlord onboarding

2 · Tenant finding

3 · Rent collection — the heartland

4 · Arrears management

5 · Maintenance

6 · Compliance — high risk, low volume; the safety floor

7 · Inspections

8 · Tenancy management

9 · Financial

10 · Regulatory

11 · Disputes — where the copilot earns its keep

Disaggregation — one process is many τ

The method is not to label a whole process; it is to split it and treat each part on its own terms. A Section 8 possession:

• Draft the notice — grounds and dates are rules. τ ≈ 0 → 🟢 A, self-verifying.
• Serve and file — the court’s acceptance is the gate. τ ≈ 0 → 🟢 A.
• Win at the hearing — a world-outcome months away. τ = months → 🔵 B (human + our private possession data).

Same logic for compliance (track = A; certify = physical/outsourced; cert-on-file = a τ ≈ 0 gate) and arrears (reminders = A; the payment-plan conversation with a struggling tenant = B). Disaggregation is where the alpha is: you buy a process the market prices as human-only and re-engineer the τ ≈ 0 parts to A, leaving humans only the irreducible remainder. The verification path column above is the operational handle on this: it names the exact τ ≈ 0 gate (cert on file, scheme receipt, reconcile-to-zero, court acceptance) that lets the machine close the loop without a human.

Column B is not where AI stops — it is where its leverage is highest

A complaint, a vulnerable tenant, a possession call, a deposit dispute — these are hard not because the reasoning is beyond AI, but because the right answer depends on everything: this tenant’s full history, every prior message, the tenancy terms, the landlord’s instructions, the arrears ledger, the maintenance log, the relevant regulation and our own precedent. A traditional agency scatters that across a CRM, an inbox, a filing cabinet and someone’s memory — so the human operates half-blind, and the “system” adds friction rather than judgement.

The Labs-app pattern inverts this. One unified schema — wikis and structured data and full history — connected to a single AI instance that can read all of it. On a column-B task the human does not get a chatbot that “handles” the case; they get a copilot that has already read the entire file, surfaced the governing clause and the closest precedent, drafted the response, and flagged the two or three things that actually matter — so the human applies taste and judgement, and owns the outcome, with total context.

That is the difference between B done badly — CRM theatre, fields nobody fills in — and B done our way: context-complete augmentation. It is also why B is not a cost centre. A human with a context-complete copilot handles several times the caseload of one working half-blind, at higher quality, with a complete audit trail. The accountability stays human; the preparation, recall and drafting do not.

What the map tells us

• By volume, the work is automatable. The overwhelming majority of events — rent collection, payments and statements (~3,600/yr per 100), tenant comms (800), maintenance intake (400) — are τ ≈ 0, anywhere-knowledge → A. This is the labour, and therefore the cost, that the central brain absorbs.
• By risk, the danger concentrates in low-volume work — and splits cleanly: safety compliance is high-stakes but τ ≈ 0 to track (A) with the liability outsourced to the signing professional (physical); the genuinely contested work — disputes, complaints, possession strategy — is low-volume → B copilot.
• By cost, outsourcing follows liability — and most of it is recharged, not borne. The physical, certifiable work (gas, EICR, EPC, Legionella, inventory/check-in & check-out, emergency call-outs) is 🔧 outsourced to the signing professional and recharged to the landlord — it barely touches our P&L. Physical attendance needing no certificate (viewings, inspections) is 👤 part-time field staff, absorbed in the fee. The high-volume heartland — rent, payments, statements, comms, arrears, reconciliation, compliance tracking — is 🏠 in-house to the AI, near-zero marginal cost inside the fee. What remains as genuine firm overhead is small and non-property: the company's own accounts & corporation tax (outsourced), VAT review, AML supervision where in scope, legislative monitoring. The structural result: the cost of doing the work is overwhelmingly recharged to landlords or near-zero in-house — very little is fixed overhead we carry.
• Very few processes are both high-volume and high-irreducible-risk. That is precisely why the economics work: automate the volume, procedural-gate or outsource the safety, put a context-complete human on the contested remainder.
• The residual is exactly section 8 of The Intelligent Letting Agency: physical presence, accountability, and judgement.

On AML — and why we do not run AML on tenants

A point that materially changes the compliance burden, verified against HMRC guidance (June 2026): letting-agency work is only within the Money Laundering Regulations 2017 where the monthly rent is €10,000 or more. Below that — i.e. essentially the entire normal residential book — a pure lettings/management firm has no AML registration and no customer-due-diligence duty at all, on landlords or tenants.

• AML on landlords — included in the map, but conditional. It bites only if the firm does estate-agency (sales) work (in scope since 2017, no threshold) or any lettings ≥ €10k/month. Most high-street agents do some sales, which would bring the whole firm into scope. Action: confirm whether Glenn Flegg does sales — if so, AML registration + CDD on landlords (and on the counterparties to any sale) is mandatory.
• AML on tenants — deliberately excluded as a process. It is required only in the high-value (≥ €10k/month) letting case, where the agent must CDD both landlord and tenant. The tenant-side check we do run universally is Right to Rent (Immigration Act 2014) and a financial-sanctions screen — neither of which is AML. Carrying a tenant-AML process for a sub-€10k book would be inventing an obligation that does not exist.

Appendix — regulatory parameters (every regulated process)

Compiled from the process registry (public.labs_lettings_processes); legal facts verified against gov.uk / legislation.gov.uk, June 2026 (see Provenance). One row per regulated operating process (28 of 44).

Firm-level registrations — one-time / annual, not per-process

From public.labs_lettings_regulatory_reqs. These authorise the business to trade; the per-process table above governs day-to-day operation. The final column reads them through the lens of an asset purchase — a clean NewCo buying the book and assets rather than the company’s shares; the vendor’s authorisations do not transfer, so each is held in our own name before close, borrowed transitionally from the vendor (a TSA, until ours is granted), or deferred. On a share purchase the opposite holds — the company and its registrations come with the shares; only PI insurance must be re-placed.

The acquisition engine — noted

The same lens applies to how we buy (sourcing → desktop DD → compliance audit → valuation → SPA → TUPE → migration → onboarding). Most of it is τ ≈ 0 knowledge work the market charges £30–50k of professional fees for — A with Ed reviewing — leaving only relationship/negotiation and the office visit as B/physical. It deserves its own map; flagged here for completeness.