
Build
An AI voice assistant for an inbound worker-support line
A production AI voice assistant for Liquid Friday’s inbound worker-support calls: callers verified, answers grounded in their own platform, and a safe human handoff the moment a call needs one.
Context
Liquid Friday provides payroll, invoice finance and workforce compliance services to contractors, recruiters and hirers across the UK. They are a regulated, accredited business in a sector where compliance failures are expensive.
Their inbound worker-support line carries high call volume, and much of it is people asking for information their own platform already holds.
The problem
A voice assistant handling payroll and compliance questions cannot guess. Every answer has to come from the authoritative system, and anything sensitive or uncertain has to reach a person quickly.
It also had to fit an established estate rather than introduce a technology island alongside it.
What we built
A voice assistant that verifies the caller, interprets the request, retrieves approved information from Liquid Friday’s own systems through a controlled API boundary, explains it clearly, and escalates to a human whenever the request is sensitive, unsupported or uncertain.
First release covers general support assistance, three to five read-only worker intents, caller verification, audit, monitoring and human handoff.
Built with their grain: a Node.js and TypeScript voice service integrating with their existing Express, PostgreSQL, Redis and Heroku estate, rather than a re-platform.
A provider-adapter architecture isolates telephony, speech and model providers, so any of them can be replaced without rewriting the platform integration.
Outcome
Structured as a fixed-price implementation of GBP 100,000, with a GBP 1,000 per month managed service covering monitoring, incident response and updates across a 36-month term.
Write-actions such as logging support tickets and scheduling callbacks are scoped separately, to be added once the read-only journey is proven.
Technical and security
Policy sits outside the model. Authentication, authorisation, data minimisation and allowlists are enforced by the platform, not by the assistant.
The model has no direct database access, no unrestricted internal API access, and no authority to make payroll or compliance decisions. Liquid Friday’s platform remains the single source of truth.
Every interaction is audited, with 24/7 monitoring and monthly service reporting under the managed service.
Building something similar?
Tell us what you are trying to achieve and we will tell you which engagement model actually fits.


