Conversational AI & Chatbots

Automating the Full Lead Lifecycle for HMO Room Rentals

A deployed, zero-touch pipeline that captures rental enquiries, qualifies tenants over WhatsApp and voice, books viewings, and chases property managers to a resolved outcome.

August 21, 2026

OUTLINE

ENGAGEMENT SNAPSHOT

Figure 1 - Key figures from this engagement, at a glance.

EXECUTIVE SUMMARY

Our client's HMO (House in Multiple Occupation) letting operation ran its entire lead lifecycle manually: SpareRoom enquiries, WhatsApp threads, outbound qualification calls, Property Manager chasing, and daily status reporting were all handled by a person, which capped how many leads the operation could work in parallel.

Pfactorial Technologies built and deployed the HMO AI Agent to replace that manual loop end to end: every SpareRoom enquiry is captured automatically, tenants are qualified over WhatsApp with an AI voice call fallback, viewings are booked directly against live Calendly availability, and Property Managers are chased through a timed escalation sequence - all without a person driving each step.

The system closes each day with a consolidated End-of-Day report, delivered automatically over WhatsApp and email at 6 PM UK time, giving the operation a single, reliable summary of every lead, call, viewing, and follow-up without anyone compiling it by hand.

Why this engagement is representative This engagement demonstrates Pfactorial's approach to full lifecycle automation: replacing a fixed-throughput manual process with a coordinated set of services that capture, qualify, book, and escalate without a person driving each step, while keeping every architectural boundary deliberate and documented.

THE CHALLENGE

The manual lead process capped how much volume the letting operation could work in parallel, and introduced delay and inconsistency at every handoff. Four problems needed solving together.

1. Lead capture was entirely manual

SpareRoom enquiries had to be read and logged by a person, and SpareRoom exposes no API for automated access - every enquiry represented a manual step before qualification could even begin.

2. Qualification depended on a person being available

Move-in date, budget, employment status, and house-rule fit all had to be established through a manual conversation, which meant qualification speed was bounded by how many conversations one person could run at once.

3. Property Manager follow-up had no consistent cadence

Confirming a viewing with a Property Manager depended on someone remembering to chase them, with no consistent, timed structure ensuring a booked viewing was never left unconfirmed indefinitely.

4. Daily status reporting was compiled by hand

Pulling together the day's leads, calls, viewings, and outcomes into a status report was manual work repeated every single day, with no automatic aggregation across the systems involved.

The real brief Not a chatbot bolted onto one step, but a coordinated system that runs the entire lead lifecycle - capture, qualification, booking, and PM follow-up - with zero manual intervention.

THE SOLUTION

Pfactorial built the system as five coordinated layers: a lead source, a capture layer, a core backend owning business logic, an automation layer owning every third-party integration, and a daily reporting layer.

Figure 1 - System architecture: lead source through capture, FastAPI core, the Make.com automation layer, and the daily report.

Architectural principles

FastAPI owns logic, Make.com owns integrations - the backend never calls Google Sheets, ClickUp, or WhatsApp directly - it fires a webhook event to Make.com and returns in under 10 seconds, isolating core business logic from third-party API quirks and rate limits.
Silence is a signal, not a failure - a 30-minute window with no WhatsApp reply automatically triggers a VAPI outbound qualification call rather than leaving the lead stalled, treating non-response as an expected branch in the flow rather than an edge case.
Escalation is timed and bounded, never open-ended - Property Manager confirmation runs on a four-step, time-boxed escalation (immediate, +2h, +4h call, +6h human handoff), ensuring a viewing is never left unconfirmed indefinitely while still giving PMs a fair window to respond.
PostgreSQL is the single source of truth - leads, conversation state, call logs, and booking status all live in one relational database that the core backend owns, with ClickUp and Google Sheets kept as synchronized views rather than competing systems of record.

CAPABILITIES DELIVERED

The deployed system covers the full lead journey, from first enquiry to a resolved viewing outcome and same-day reporting.

CAPABILITY	WHAT IT DOES
Automated Lead Capture	A scheduled Playwright script polls the SpareRoom enquiry inbox every 5-10 minutes, parsing and storing every new enquiry with zero manual reading.
WhatsApp-First Qualification with Voice Fallback	Five core qualification questions run over WhatsApp first, falling back to an AI voice call after 30 minutes of silence.
Live Calendly Booking	Viewings are booked directly against real-time Calendly availability, verbally offered during qualification calls or sent as a one-time link over WhatsApp.
Timed PM Escalation	A four-step escalation sequence chases Property Manager confirmation, ending in human handoff if unresolved after six hours.
Structured Voice-to-Data Extraction	Every call transcript is passed to an LLM immediately after the call ends, converting free-form conversation into structured qualification data.
Automated End-of-Day Reporting	A consolidated daily summary of leads, calls, viewings, and PM follow-ups is generated and delivered over WhatsApp and email at 6 PM UK time.

Figure 2 - Lead lifecycle: from SpareRoom enquiry through qualification, booking, PM follow-up, and same-day reporting.

Design note SpareRoom offers no official API, so lead capture depends on a scheduled Playwright polling script (with Selenium documented as a fallback framework) rather than a direct integration - a deliberate, documented constraint rather than an oversight.

ENGINEERING FOR SCALE AND RELIABILITY

Several engineering decisions shape how the system holds up under real tenant and Property Manager behavior, not just a clean demo flow.

Two purpose-built voice assistants, not one general one

the Tenant Qualifier (180-second budget, warm UK-accented tone, barge-in enabled) and the PM Chaser (60-second budget, direct and brief) are separately configured, each with its own LLM, dynamic variables, and toolset matched to its specific job.

Barge-in is handled natively, not scripted

VAPI handles conversational barge-in natively, letting a tenant or PM interrupt the assistant mid-sentence the way they would a human caller, rather than forcing a rigid turn-taking script.

Retry logic is bounded and spaced

unanswered qualification calls are retried up to three times, two hours apart, before a lead is flagged unreachable - persistent enough to reach real people, bounded enough not to harass them.

A dedicated LLM is used specifically for report quality

Claude Sonnet is applied specifically to End-of-Day report generation, where concise, well-structured summarization was prioritized, while GPT-4.1 handles the qualification and PM-chasing conversations - a deliberate model split rather than one LLM used everywhere.

Third-party coupling is architecturally barred

FastAPI is deliberately barred from calling Sheets, ClickUp, or WhatsApp directly; the alternative of direct calls was evaluated and explicitly rejected in favor of a dedicated integration layer.

Speech is handled by specialized, tuned components

Deepgram Nova-2 performs speech-to-text and ElevenLabs 2.5 Turbo performs text-to-speech, specifically tuned for fast, natural UK-accented conversation rather than generic voice defaults.

DELIVERY APPROACH

The engagement moved from a fully manual lead process to a deployed, zero-touch pipeline in five phases.

1. Build lead capture - built the Playwright-based SpareRoom polling script and PostgreSQL system of record, with automatic ClickUp and Google Sheets synchronization.

2. Build the qualification flow - implemented the WhatsApp-first, five-question qualification state machine with a 30-minute silence trigger for AI voice call fallback.

3. Build voice orchestration - configured the Tenant Qualifier and PM Chaser VAPI assistants, with Deepgram and ElevenLabs tuned for natural UK-accented conversation.

4. Build booking and PM escalation - wired live Calendly availability into the qualification flow and built the four-step, time-boxed Property Manager escalation sequence.

5. Build daily reporting and deploy - implemented the 6 PM UK End-of-Day aggregation and delivery, and deployed the full pipeline to production for the client's live SpareRoom-sourced letting operation.

RESULTS AND IMPACT

Figure - Key outcomes from this engagement.

In production use, 100% of SpareRoom leads are captured and stored automatically, with every enquiry reflected in ClickUp and Google Sheets within 10 minutes of arriving - a latency and consistency bar a manual process could not sustain across variable enquiry volume.

The full lead lifecycle - qualification, viewing booking, Property Manager follow-up, and outcome resolution - now runs without a person driving each step, closing out each day with an automatically compiled End-of-Day report delivered on schedule at 6 PM UK time.

What it enabled commercially

By automating lead capture, qualification, booking, and PM escalation end to end, the client's letting operation can work SpareRoom enquiries at whatever volume arrives, rather than being capped by how many conversations one person can run in parallel.

WHY PFACTORIAL

This engagement draws on Pfactorial's AI product engineering capability: coordinating voice, chat, and workflow automation into a single deployed system with deliberate architectural boundaries - not a chatbot bolted onto one step of a larger manual process.

Figure - Service lines this engagement draws on.

Engagement enquiries Pfactorial Technologies works with property and letting operations looking to automate lead capture and qualification without losing the natural, human feel of the conversation. If you are evaluating whether a lead lifecycle is a good fit for automation, we are happy to give you an honest read on scope, cost, and risk before anyone commits to anything. · pfactorial.ai

APPENDIX A - TECHNOLOGY STACK

The technology stack underpinning the system, grouped by the layer it serves.

Result and Analysis

ENGAGEMENT SNAPSHOT

A deployed, zero-touch pipeline that captures rental enquiries, qualifies tenants over WhatsApp and voice, books viewings, and chases property managers to a resolved outcome.