Case Study
A London property management group with 1,800 units across Greater London deployed a Softomate-built AI voice agent on Vapi + GPT-4o, handling 1,200 inbound tenant and contractor calls per month, capturing 91% of after-hours calls into qualified callbacks, and lifting daytime call answer rate from 67% to 99%.
1 min read By Deen Dayal Yadav, Founder & AI Automation Director
London property management group, 47 staff, 1,800 managed units across Greater London
A London property management group with 1,800 units across Greater London deployed a Softomate-built AI voice agent on Vapi + GPT-4o, handling 1,200 inbound tenant and contractor calls per month, capturing 91% of after-hours calls into qualified callbacks, and lifting daytime call answer rate from 67% to 99%.
London property management group, 47 staff, 1,800 managed units across Greater London
A London property management group with 1,800 units across Greater London deployed a Softomate-built AI voice agent on Vapi + GPT-4o, handling 1,200 inbound tenant and contractor calls per month, capturing 91% of after-hours calls into qualified callbacks, and lifting daytime call answer rate from 67% to 99%.
A London property management group with 47 staff and a portfolio of 1,800 residential units spread across Zones 2 to 4 of Greater London had reached a clear operational limit on its inbound phone handling. The firm operated a centralised tenant services line that handled all inbound calls from tenants (reporting maintenance issues, asking about deposits, requesting documents, etc.) and from contractors (confirming attendance, asking for access codes, providing quotes). The line was staffed by three coordinators during weekday office hours, and rolled to a voicemail service outside of those hours.
The firm's call-handling data, pulled from the telephony provider, showed an unambiguous picture. Daytime inbound call volume averaged 240 calls per day across the working week, peaking at over 320 on Mondays. The three coordinators could realistically handle around 160 calls per day at the level of attention each call required. The unanswered calls (roughly 33% of total volume) either dropped into voicemail, were retried by the caller later, or were lost entirely. The operations director had measured, by sampling, that approximately 14% of unanswered tenant calls related to genuinely urgent maintenance issues (water leaks, lock failures, heating outages) that escalated into larger problems because they were not handled at first contact.
The after-hours position was worse. The firm received approximately 80 calls per evening on weekdays and 110 per weekend day, all routed to a voicemail service that produced transcripts the next working day. Around 4% of those after-hours calls were genuine emergencies (security incidents, gas leaks, severe water ingress) that required out-of-hours engineer dispatch through the firm's emergency contractor panel. The current process for triaging emergency calls from a voicemail backlog the next morning was, the operations director acknowledged, structurally inadequate. The firm had absorbed two material insurance claims in the previous twelve months that the loss adjuster had explicitly linked to delayed emergency response.
Adding human capacity was not commercially viable. Three additional daytime coordinators would have cost roughly ?108,000 fully loaded, and an out-of-hours human dispatch desk would have added a further ?90,000+ annually. The firm's competitive position in the London letting market did not support absorbing those costs into management fees, and clients had made clear in renewal conversations that fee increases were not acceptable.
The brief to Softomate was direct. Handle the daytime overflow so the three coordinators could focus on calls that genuinely needed human judgment. Triage after-hours calls so genuine emergencies were dispatched to the emergency contractor within minutes rather than the next morning, and so non-emergency calls were qualified and queued for the appropriate coordinator to action on next working day. Capture every call into the firm's existing property-management system (Reapit) so no information was lost. Sound like a competent member of the team, not an obvious automated system.
Softomate built the AI voice agent on Vapi as the telephony orchestration layer and GPT-4o as the language model, with retrieval-augmented access to the firm's property database (Reapit), the contractor panel directory, and a structured knowledge base of property-management policy (deposit handling, repair categorisation, emergency-vs-routine criteria, statutory landlord obligations). The agent was reachable on the firm's existing main tenant services number, with calls routed first to the agent and escalated to a human coordinator at any point the caller asked or the agent's escalation logic determined.
The conversation design was the highest-effort part of the build. Property-management calls cover a wide range of intents (maintenance report, deposit query, document request, contractor confirmation, tenancy query, emergency report, complaint, payment query) and each requires a different response pattern, different data lookups, and different downstream actions. Softomate built explicit intent classification at the opening of each call, with the agent asking a single clarifying question if intent was ambiguous (?Is this about a maintenance issue or something else??). Once intent was classified, the conversation was routed to the appropriate sub-flow.
The emergency classification was the most safety-critical sub-flow. The agent was trained to recognise specific signals indicating a genuine emergency: keywords (gas, smoke, fire, flood, break-in, locked-out, no-heating-elderly), tone signals from the caller, and explicit statements. On recognising an emergency, the agent immediately initiated a parallel action: it confirmed the property address with the caller, looked up the property's assigned emergency contractor in Reapit, dispatched the contractor via SMS to the relevant on-call mobile, and notified the firm's on-call duty manager via SMS and a phone call. The caller stayed on the line throughout this process and was given a confirmed timeline for engineer arrival before the call ended.
The maintenance-report sub-flow was the highest-volume intent. The agent gathered structured information (property address, specific issue, tenant contact details, photographic evidence if the caller could send WhatsApp), wrote a fully formed maintenance ticket into Reapit categorised by urgency level, and either dispatched to the relevant routine-maintenance contractor for category-2 or category-3 issues, or queued for human coordinator review for ambiguous or complex issues. The agent gave the caller a clear expectation of next steps and a reference number for follow-up.
Brand voice calibration used the firm's existing call recordings from its top three highest-performing coordinators (with appropriate consent). Softomate analysed 240 sample calls to identify the tonal patterns that worked: when to apologise, when to reassure, when to be efficient and brief, when to slow down for elderly or distressed callers. The model was calibrated against these patterns, and the operations director reviewed 100 generated call transcripts before launch to sign off on tone consistency.
The deployment was phased. Week one was a closed pilot with internal staff calling the agent and reporting transcripts. Week two opened to 10% of inbound traffic, with the rest routing normally. Week three rose to 30%, then 60%, then 100% by week five. Every call during the first six weeks was reviewed daily by Softomate and the operations director jointly, with 34 specific refinements made to intent classification, contractor-dispatch logic, and tonal calibration during that window.
Within the first 90 days of full deployment, the agent was handling approximately 1,200 calls per month directly to completion (resolution, dispatch, or scheduled callback) without human coordinator involvement. The daytime call answer rate rose from 67% (with significant volume going to voicemail or unreached) to 99%, with the remaining 1% being calls the agent escalated immediately to a human coordinator because intent was ambiguous or the caller asked for a person.
The after-hours emergency response time was the outcome the operations director valued most. Before deployment, average response time on after-hours emergency reports (from caller report to engineer dispatch) was over 11 hours, because reports sat in voicemail until the next morning. After deployment, average response time fell to 7 minutes, measured across 43 confirmed after-hours emergencies in the first 90 days. The firm's insurance broker explicitly cited the improved out-of-hours response in renewing the firm's professional indemnity policy at a 4% reduced premium.
The three human coordinators reported, in a structured interview at the 60-day mark, that their working day had changed in character. Routine maintenance reports, deposit queries, and document requests (which had previously consumed the bulk of their time) were now handled by the agent. Their time was concentrated on complex tenancy queries, complaints, multi-property landlord-side queries, and the harder cases the agent escalated. Two of the three coordinators specifically commented that the work was more satisfying and that they felt their senior judgment was being used rather than spent on data lookups.
The cost comparison was unambiguous. The agent handled volume that would have required at least three additional daytime coordinators plus an out-of-hours dispatch desk to service at the same quality level. Total avoided headcount cost was estimated at over ?198,000 annually. Softomate's engagement cost was recovered within the first 5 months of go-live, calculated against avoided hire cost alone.
The unexpected outcome was data quality. Because every call was structured into a formal Reapit ticket with consistent fields, the operations director gained for the first time a clear view of which properties generated the most maintenance contact volume. A pattern emerged showing that 14 specific properties (representing under 1% of the portfolio) generated approximately 11% of all maintenance calls, which led to a targeted capital expenditure review of those properties and a series of landlord recommendations that reduced ongoing contact volume from those properties by an estimated 60% over the following quarter.
The agent has since been extended to handle outbound calls in two specific workflows: confirmation calls to tenants the day before a scheduled engineer visit, and follow-up calls after engineer attendance to confirm satisfaction with the work completed. Both extensions were delivered as configuration changes rather than new builds and are expected to add a further 200-300 hours of operational saving per quarter.
Related service:AI Voice Agent Development London. Further reading:AI Receptionist for London Estate Agents, Bland AI vs Vapi vs Retell AI UK and AI Receptionist vs Human. Related case study:AI Chatbot for London Letting Agency.
Anonymised client engagement. Identifying details modified for confidentiality. Outcome ranges reflect typical results from similar projects.
Names withheld to preserve confidentiality.
Names withheld to preserve confidentiality.
Work with us
Every project we take on has a measurable outcome. Talk to our London team and we will show you exactly how we would approach your challenge.
Softomate Solutions Limited
Deen Dayal Yadav
Online
We use essential cookies to keep the site running. With your permission, we also use analytics cookies to understand how visitors use our site so we can improve it. No data is sold. Privacy Policy