An AI voice agent development for a UK mortgage broker handles the inbound call types that do not require FCA-regulated mortgage advice: initial enquiry qualification (property value, loan amount, employed vs self-employed, credit history flags), appointment booking, callback scheduling, document checklist reminders, and case status updates for existing clients. For a busy mortgage broker receiving 80-200 inbound calls per week, an AI voice agent handles 55-65% of these calls, freeing advisers to focus on advice-giving and case progression. Implementation costs £2,500-£6,000 and deploys in 3-6 weeks. Softomate Solutions builds FCA-aware AI voice agents for UK mortgage brokers using VAPI and ElevenLabs infrastructure.
Last updated: 18 May 2026
Published 18 May 2026The Financial Conduct Authority regulates mortgage advice through the Mortgage Conduct of Business Sourcebook (MCOB). Before deploying any automated voice technology in a mortgage brokerage, it is essential to understand precisely where the regulatory boundary sits - and where an AI voice agent operates safely on the non-advice side of that boundary.
MCOB 4 covers pre-application disclosure requirements. When a consumer contacts a mortgage broker for the first time, the firm must disclose whether it provides regulated mortgage advice or information only, and confirm its permissions. An AI voice agent can deliver this disclosure verbatim from an approved script at the start of every call. This is not advice - it is factual disclosure, which any automated system can handle. The key requirement is that the disclosure must be given consistently, which is something AI voice agents do better than human receptionists who often skip or abbreviate it under call pressure.
MCOB 5 covers responsible lending and the requirement for firms to assess whether a mortgage is suitable for a customer. This is the core advice provision. No AI voice agent should perform an affordability assessment or recommend a mortgage product. The system must be configured to deflect any question that crosses into advice territory with a consistent response such as: "That is a great question for your mortgage adviser. I will book you in for a conversation where they can go through the figures with you."
MCOB 9 covers advised sales. The rule is clear: a mortgage recommendation must be made by an FCA-authorised individual with the appropriate permissions. AI voice agents are not FCA-authorised individuals. Any statement that could be interpreted as a recommendation - including phrases like "based on what you have told me, a 5-year fixed rate might suit you" - is prohibited territory for an AI system.
The following table sets out clearly what an AI voice agent can and cannot do within an FCA-authorised mortgage brokerage:
| Permitted AI Voice Agent Actions | Requires FCA-Authorised Adviser |
|---|---|
| Deliver pre-application FCA disclosure script | Recommend a specific mortgage product |
| Collect property value, purchase price, loan amount | Assess affordability or debt-to-income ratio |
| Record employment status (employed, self-employed, contractor) | Advise on suitability of repayment vs interest-only |
| Flag yes/no credit history questions (CCJs, missed payments) | Quote a rate as appropriate for this client |
| Book adviser appointments and send confirmation | Discuss the key features illustration (KFI) in detail |
| Send document checklists (payslips, P60, bank statements) | Explain the implications of early repayment charges |
| Provide case status updates (valuation booked, offer received) | Advise on product transfer vs remortgage decision |
| Confirm completion date and next-step logistics | Discuss government schemes (Help to Buy, Shared Ownership eligibility) |
| Take inbound enquiry calls outside business hours | Handle vulnerable customer disclosures substantively |
| Escalate to adviser with warm transfer or callback booking | Discuss negative equity or arrears resolution |
One area that frequently causes confusion is the collection of initial fact-find information. The FCA's position is that collecting factual information from a client - their income, their outstanding mortgage balance, their property value - is not advice. It is data collection. A mortgage broker's receptionist does this every day without holding mortgage permissions. An AI voice agent performing the same function is operating in exactly the same space. The critical discipline is ensuring the system never interprets that data or makes recommendations based on it.
Softomate's implementation process includes an MCOB compliance review of every call script before deployment. Scripts are reviewed against MCOB 4, 5, and 9, and a compliance sign-off document is produced for the firm's FCA compliance file. This is not legal advice - we recommend all clients run the final script past their compliance officer or appointed representative principal - but it demonstrates the level of regulatory awareness that most AI voice agent providers completely ignore.
Vulnerable customer handling is a specific FCA priority under the Consumer Duty (PS22/9, effective July 2023). The AI voice agent must be configured to recognise distress signals - unusual pauses, confusion, requests to repeat everything, mentions of bereavement or financial difficulty - and escalate to a human adviser immediately. Softomate builds this escalation trigger into every mortgage broker deployment as a non-negotiable feature.
The average UK mortgage broker takes 4.7 hours to respond to a new online enquiry, according to research by Twenty7Tec. By that point, the prospect has typically contacted two or three other brokers and is moving towards whichever firm made contact first. An AI voice agent changes this equation entirely: the system calls back within 90 seconds of a web form submission, 24 hours a day, 7 days a week.
The pre-qualification conversation captures everything an adviser needs to start the case properly. A well-designed mortgage broker AI voice agent covers the following data points in a natural 4-7 minute conversation:
This data is pushed directly to the broker's CRM - GoHighLevel automation, HubSpot, or Salesforce depending on the firm's setup - before the adviser makes contact. When the adviser calls the client, they open the case record with a fully populated pre-application data set. There is no need to ask the same questions again, which clients frequently find frustrating. The adviser can move directly into advice-level conversation: explaining the process, discussing product options in principle terms, and building the relationship.
| Without AI Voice Agent | With AI Voice Agent |
|---|---|
| Average response time: 4.7 hours | Average response time: 90 seconds |
| Adviser spends 10-15 minutes on initial data collection every call | Adviser starts call with pre-populated fact-find data |
| Out-of-hours enquiries wait until next morning | Out-of-hours enquiries called back within 90 seconds |
| No qualification filter - adviser speaks to unqualified leads | Unqualified leads (outside lending criteria) flagged before adviser time spent |
| Conversion from enquiry to appointment: industry average 28% | Conversion from enquiry to appointment: 40-55% in Softomate implementations |
| Adviser capacity: 6-8 meaningful advice calls per day | Adviser capacity: 9-12 meaningful advice calls per day |
The qualification filter is particularly valuable for brokers operating in specialist markets. A commercial mortgage broker, for example, can configure the AI to detect enquiries that fall outside minimum loan thresholds or security types, and route them to a referral partner or decline politely - before the adviser has invested time in the conversation. A buy-to-let specialist broker can filter for minimum rental yield requirements, minimum property value, or portfolio size thresholds.
Softomate configures the CRM integration so that the AI voice agent's call transcript and extracted data fields populate the lead record automatically. The adviser sees not just the data points but the exact words the prospect used - which often reveals context that structured fields miss. A prospect who says "we are trying to get our children into the catchment area before September" is indicating urgency. That context, captured in the transcript, shapes how the adviser approaches the first conversation.
A mortgage case from application submission to completion typically takes 8-12 weeks. During that period, clients are anxious. They want to know what is happening with their case, and when they do not hear from their broker, they call in. These inbound status enquiry calls - "where are we with the application?", "has the valuation come back yet?", "when will we get the mortgage offer?" - represent 25-40% of all inbound calls at a typical mortgage brokerage. They are necessary from the client's perspective, but they do not require adviser expertise. An AI voice agent handles them entirely.
The outbound case update workflow operates on CRM triggers. When the case status changes in the broker's CRM or case management system, the AI voice agent initiates a proactive outbound call to the client. The client receives a call from the broker's number with a natural-sounding update, delivered in the same voice they have already heard during the qualification process. This consistency builds trust and reduces the "who is calling me?" friction of an unknown number.
Key trigger points and corresponding AI voice agent call scripts:
Brokers who implement this outbound update system report a 60-70% reduction in inbound status enquiry calls within the first month. The client experience improves significantly - they feel informed and looked after without needing to chase. This drives referral and review behaviour: clients who feel proactively communicated with are significantly more likely to leave a Google review and recommend the broker to friends and family.
The system also handles the case where the valuation comes back below the purchase price, or where the lender declines the application. These sensitive situations are flagged for immediate adviser escalation - the AI does not deliver bad news autonomously. Instead, it sends an alert to the adviser with a prompt to call the client within the hour, and books a callback slot if the client picks up an outbound call before the adviser is available.
Ask any mortgage broker what takes the most time in their day and the answer is almost always document chasing. Clients submit an application and then go quiet. Payslips arrive late. P60s are lost. Bank statements are three months old when the lender needs six months. ID documents expire. Proof of address is a utility bill from 18 months ago that does not meet the lender's requirements. Every missing document adds days to the case and creates compliance risk if the case file is incomplete for FCA supervision purposes.
Document chase automation using an AI voice agent removes this burden entirely from the adviser and the admin team. The workflow is as follows: when the broker's CRM flags a document as missing or outstanding, the AI voice agent initiates a multi-channel chase sequence. This is not a simple email reminder - it is a voice-first sequence with email and SMS as supporting channels:
The result is a 3-5 day reduction in average case processing time, driven entirely by faster document receipt. Lenders have specific response time requirements - some require case completion within 30 days of application or they close the case. For clients with complex income structures (self-employed directors, contractors, those with multiple income streams), document requirements are more extensive and the chase cycle would previously consume significant adviser time. The AI handles it end to end.
Integration with specialist mortgage software is handled via CRM tags. For brokers using Mortgage Brain, the CRM tag triggers correspond to case status fields that Mortgage Brain populates. For Iress and Intelliflo users, the integration is via webhook from the CRM layer (typically GoHighLevel or a bespoke integration). Softomate's team handles the integration mapping as part of the implementation project.
One detail that is often overlooked: the AI voice agent's document chase calls are recorded and logged with timestamps. This creates an audit trail that demonstrates to the FCA (if asked) that the firm took reasonable steps to obtain required documentation within required timeframes. For firms that have experienced FCA supervision visits, this audit trail capability alone is worth the implementation cost.
Choosing the right infrastructure platform for an AI voice agent in a regulated financial services context is not simply a matter of picking the most capable voice or the cheapest cost per minute. UK mortgage brokers are FCA-authorised firms. They have GDPR obligations. Their clients are sharing sensitive financial and personal data over the phone. Platform selection must account for data residency, call recording obligations, GDPR consent handling, and voice quality in trust-sensitive financial conversations.
| Feature | VAPI | ElevenLabs (with Twilio) | Bland.ai |
|---|---|---|---|
| EU/UK data residency option | Yes (EU servers available) | Yes (ElevenLabs EU region) | US-only at present |
| GDPR call recording consent handling | Configurable pre-call consent prompt | Configurable via Twilio | Limited native GDPR tooling |
| Voice naturalness (financial trust context) | High (ElevenLabs voices integrated) | Excellent (native ElevenLabs voices) | Good but detectably synthetic |
| Latency (UK callers) | Low (200-400ms with EU routing) | Low-medium (300-500ms) | Medium (400-700ms, US-routed) |
| Cost per minute (inbound) | £0.05-£0.12 | £0.08-£0.18 (combined stack) | £0.04-£0.09 |
| Call transcript storage location | Configurable (EU bucket) | Configurable (EU bucket via Twilio) | US storage by default |
| CRM webhook integration | Native + custom webhooks | Via Twilio Functions or middleware | Native webhooks |
| Warm transfer to human adviser | Yes (native) | Yes (via Twilio Transfer) | Yes (native) |
| FCA compliance documentation support | None (Softomate provides) | None (Softomate provides) | None (Softomate provides) |
For UK mortgage brokers, Softomate's recommended configuration is VAPI with EU data residency routing and ElevenLabs voice synthesis. This combination delivers the best balance of voice naturalness (critical for financial conversations where synthetic voice reduces trust), data residency compliance, and cost efficiency. The VAPI platform provides the call orchestration, CRM webhook integration, and warm transfer capability. ElevenLabs provides the voice layer - clients consistently report that ElevenLabs voices are the most natural-sounding they have encountered, which matters when asking someone to share their income and credit history over the phone.
Bland.ai's US-only data residency is a significant issue for FCA-authorised firms. GDPR requires that personal data processed by or on behalf of a UK firm is either held in the UK/EEA or transferred under appropriate safeguards (Standard Contractual Clauses with a Transfer Impact Assessment). For sensitive financial data including income, credit history flags, and property details, most FCA compliance officers will advise against US-only storage without a formal data transfer assessment. Bland.ai is not disqualified by this - SCCs can be implemented - but it creates additional compliance overhead that VAPI with EU routing avoids.
Call recording is a specific FCA requirement for mortgage brokers. MCOB requires firms to retain records of advice given. For AI voice agents, the recording serves a dual purpose: FCA compliance (demonstrating that the system stayed within permitted boundaries) and quality assurance (reviewing calls to identify where the script needs refinement). Softomate configures call recording storage in a GDPR-compliant EU bucket with a defined retention period (typically 5 years for FCA mortgage records, matching MCOB record-keeping requirements).
The GDPR consent requirement for call recording is handled by the AI voice agent itself. At the start of every call, the system delivers the following disclosure: "This call may be recorded for quality and compliance purposes in line with our data protection obligations. By continuing, you consent to this recording. If you prefer not to be recorded, please say 'no recording' and I will connect you to a member of our team." This consent capture is logged with a timestamp for the firm's GDPR records.
Softomate Solutions implements AI voice agents for UK mortgage brokers through a structured 5-stage process designed specifically for regulated financial services firms. The process is designed so that the broker's compliance officer can review and sign off every stage before the system goes live.
Stage 1: Compliance and call script review (Week 1-2)
Softomate's team reviews all proposed call scripts against MCOB 4, 5, and 9. Every script is checked for prohibited advice language. Vulnerable customer escalation triggers are defined. The GDPR consent and recording disclosure is drafted and approved. A compliance documentation pack is produced for the firm's FCA file.
Stage 2: Platform configuration and voice selection (Week 2-3)
VAPI account configured with EU data residency. ElevenLabs voice selected and customised with the broker's brand name and tone. Call flows built in VAPI: inbound qualification, outbound case updates, document chase sequences. GDPR consent capture configured. Call recording configured to EU storage bucket with 5-year retention.
Stage 3: CRM integration (Week 2-4)
GoHighLevel, HubSpot, or Salesforce integration built. Webhook triggers mapped to CRM case status fields. Lead data extraction fields mapped to CRM lead record schema. Call transcript posting configured. Test data flows validated end-to-end.
Stage 4: Adviser handoff workflow (Week 3-4)
Warm transfer configuration for live calls. Callback booking integration (typically Calendly or the broker's existing booking system). Adviser alert SMS/email for priority escalations (vulnerable customer detection, declined application, urgent decision required).
Stage 5: Pilot launch and 30-day review (Week 4-6)
Soft launch with a subset of inbound calls (typically 20-30% initially). Daily call quality review for the first week. Script refinements based on real call data. Full rollout at end of pilot. 30-day post-launch review meeting: call handling rates, conversion metrics, adviser time saved, client satisfaction scores.
Investment and ongoing costs:
For a 3-adviser mortgage brokerage handling 120 inbound calls per week, the AI voice agent manages approximately 70-80 of those calls (60-65%). At an estimated adviser cost of £35/hour and 8 minutes per call, that represents approximately £2,100/month in adviser time freed - every month. The system typically pays back its implementation cost within 2-3 months.
No. Quoting a specific mortgage rate as suitable for a caller constitutes regulated mortgage advice under MCOB 9 and requires FCA authorisation. The AI voice agent is configured to deflect rate enquiries with a consistent response: "Your adviser will be best placed to discuss rates with you once they have reviewed your full circumstances. I can book you in for a call - shall we do that now?" The system is designed to move callers toward an adviser appointment, not to provide advice itself.
No. The AI voice agent is a software tool used by an FCA-authorised firm. It is analogous to a telephone system or a CRM - the tool does not require FCA authorisation, but the firm using it must be authorised and must ensure the tool operates within regulatory boundaries. The key obligation is on the broker to ensure their AI system does not cross into regulated activity. Softomate's MCOB compliance review and script sign-off process is specifically designed to support this obligation.
Yes, all calls are recorded for FCA compliance purposes. GDPR consent is obtained at the start of every call via a standard disclosure. Recordings are stored in an EU-based storage bucket with a 5-year retention period, matching MCOB record-keeping requirements. Clients can request access to their call recording under GDPR subject access rights. Softomate configures all of this as part of the implementation, and a GDPR data processing addendum is included in the service agreement.
Sole trader brokers with lower call volumes (20-50 calls per week) typically fall into the lower end of the implementation range (£2,500-£3,500) and the lower VAPI infrastructure tier (£80-£120/month). The business case is slightly longer to pay back but remains compelling: even at 25 calls per week handled by the AI, that represents approximately 200 minutes of adviser time per week freed for advice-giving activity. Softomate offers a scaled implementation package specifically for sole trader and two-adviser firms.
Yes. The call flow branches based on the property type identified at the start of the conversation. Buy-to-let enquiries trigger a different qualification track: rental yield estimates, portfolio size, existing BTL mortgage count, limited company or personal name ownership. The system can also apply different qualification thresholds - for example, routing BTL enquiries below a minimum property value to a referral partner automatically. Brokers with mixed residential and BTL books frequently use this branching to route enquiries to the relevant specialist adviser rather than a general inbox.
Vulnerable customer detection is built into every Softomate mortgage broker deployment as a non-negotiable feature, in line with the FCA's Consumer Duty requirements. The AI voice agent monitors for distress signals: mentions of bereavement, job loss, relationship breakdown, financial difficulty, or behavioural cues (confusion, distress, requests to repeat). When these signals are detected, the system immediately escalates: it apologises for the automated system, explains that a member of the team will call within the next hour, and triggers an urgent alert to the adviser. The system does not attempt to handle vulnerable customer situations autonomously.
AI receptionists handle unlimited simultaneous calls - there is no queue, no hold music, and no missed calls during peak periods. A UK trade business receiving 40 calls simultaneously on a Monday morning (post-weekend enquiry surge) will have all 40 answered within 2 seconds by the AI. Human receptionists typically handle 1-2 simultaneous calls. For UK businesses where peak call volumes cause missed enquiries (trades, healthcare, estate agents, hospitality), the simultaneous handling capability alone justifies AI receptionist investment: each missed call in a UK trade business represents £150-500 average job value.
AI voice agents built for FCA compliance are now a practical infrastructure layer for UK mortgage brokers. The technology handles 55-65% of inbound calls - qualification, case updates, document chase - without stepping into regulated advice territory. Two facts worth holding: the UK mortgage market processed 1.5 million new mortgages in 2024 (UK Finance), and brokers who respond to enquiries within 5 minutes are 21 times more likely to qualify the lead than those who respond after 30 minutes (Harvard Business Review). Speed and compliance are not in tension when the infrastructure is designed correctly from the start.
Handle 55-65% of calls without tying up your advisers. Explore Softomate's AI Voice Agent for Mortgage Brokers or book a free compliance consultation.
Written by Rakesh Patel, AI Automation Consultant at Softomate Solutions, Barking, East London.More from the blog
Let us help
Talk to our London-based team about how we can build the AI software, automation, or bespoke development tailored to your needs.
Softomate Solutions Limited
Deen Dayal Yadav
Online
