Custom AI Receptionist and Client-Intake Assistant Development | Disclosed AI, Human Approval Gates, Voice and Text

Updated September 2026 · Written and maintained by the Progression Agency strategy team

Progression Agency builds custom AI receptionist and client-intake assistants for professional-services firms: a disclosed AI that explains services and published ranges, qualifies prospects, prepares a written intake brief the visitor confirms, and hands qualified matters to a human, in text and voice, with the firm keeping ownership of the knowledge, identity data, actions and audit trail.

What does AI receptionist development include?

Design of the decision modes (answer, guide, act, escalate, human approval) so model text never silently triggers an action; a grounded knowledge base built from your approved public information, with the assistant saying when the knowledge does not support an answer; a progressive identity model from anonymous visitor to verified client, where typing a name is never proof; disclosure classes that keep matter data and prospect intelligence out of public chat; a visitor-confirmed intake brief; text and voice with transcripts kept as separate records; handoff to email, CRM and calendar; an action register and audit log; and staging with release gates before anything answers a real visitor.

The assistant lives independently of the website front end. The browser calls same-origin endpoints you own; model keys stay server-side; the core runs as lightweight APIs, queues and cron workers that work on shared PHP and MySQL hosting such as Hostinger and can move to a VPS or cloud later without a rebuild. Replacing the website theme or uploading a new build must not require retraining or re-integration, and it does not; see AI integration services.

The model provider is replaceable. Prompts, knowledge, identity state, relationships, actions and audit history belong to the firm, exported on request, so no chatbot platform becomes the source of truth; see AI chatbot development for simpler grounded chatbots and AI agent development for agents that execute workflows.

AI receptionist builds we deliver

Projects we are most often asked to build.

Website intake assistant

A text assistant on the firm’s site that explains services and published ranges, qualifies the visitor, and prepares a confirmed intake brief for a named person; the entry-level build and the one most firms start with.

Voice and text client assistant

Voice in, text out, text in, voice out, or voice both ways, with text always available and no autoplay; original audio, verbatim transcript, reply and structured summary stored as separate records.

AI receptionist for law firms

Conflict-aware intake that never gives legal advice, keeps matter details out of public chat, routes sensitive documents to secure intake and books consultations through the firm’s calendar rules.

AI receptionist for medical and dental offices

HIPAA-conscious scheduling and triage that collects only what the practice permits, with a business associate agreement and audit logging; see healthcare software development.

Multi-channel assistant with WhatsApp and email

Website, voice, email and the official WhatsApp Business API resolving to one relationship record: person, organisation, session, opportunity, matter, action.

Smart-link and outreach intake

Prospect-specific links that carry limited public-safe context through opaque server-side tokens, so a targeted outreach email lands the prospect in a conversation that already knows the topic without any personal data in the URL.

Website, voice, email and the official WhatsApp Business API resolving to one relationship record: person, organisation, session, opportunity, matter, action.

Key features and components

What separates a useful build from a costly one.

Common integrations

Where the value usually comes from.

Controls that keep the firm human-led

An assistant that can quote or commit is a liability; one that prepares a person to do so is an asset.

• The assistant identifies itself as AI in every channel and never claims to be a named person, following the disclosure principles in the NIST AI Risk Management Framework.
• Secrets stay server-side; actions and webhooks are HMAC-signed, rate-limited and idempotent; sessions and CSRF are protected; prompt-injection controls follow OWASP’s Top 10 for LLM applications.
• Personal data is handled to the privacy law that applies to the firm, for example HIPAA for US practices or POPIA for South African firms, with an operator or business-associate agreement where required.
• Production release follows staging tests of security, public and private separation, session continuity, brief confirmation, webhook behaviour, retries, idempotency, website-update independence and human-control boundaries.
• Backups and restores of assistant records are documented separately from the hosting backup; export, configuration access and an exit path are written into the contract.

Our AI receptionist development process

We start by writing the rules with you, build the knowledge base from approved content, design identity and disclosure tiers, build the core as same-origin APIs with a replaceable model provider, add voice and channels, connect handoff to email, CRM and calendar, test on staging against release gates, then launch with monitoring and cost controls.

How much does an AI receptionist cost to build?

As US planning figures: a grounded text intake assistant $10,000-$30,000 over 3-6 weeks; a voice-and-text assistant with progressive identity, confirmed intake briefs and CRM and calendar integration $30,000-$80,000 over 2-4 months; a multi-channel build with WhatsApp, email, smart links and matter workflow $80,000-$150,000 over 4-6 months. Running costs are $150-$1,500 a month for model, voice and hosting usage, depending on volume. AI receptionist software subscriptions run $50-$500 a month and suit firms whose rules fit the product; a custom build is for firms whose rules do not. Prices are confirmed after a paid discovery phase.

Technology we typically use

Mainstream, maintainable choices.

Mistakes to avoid

Common causes of overruns and low adoption.

• Letting the model decide when to act instead of an explicit decision mode.
• Treating a typed name or email as proof of identity.
• Putting confidential matter data into the same retrieval index as public knowledge.
• Locking the firm into a chatbot platform that owns the knowledge and the history.
• A website rebuild that breaks the assistant because it was wired into the theme.
• Launching without staging gates, cost caps or a documented backup and restore.

Why choose Progression Agency for AI receptionist development?

Engineering, design and go-to-market in one team.

• We build AI systems and the marketing sites they sit on, so the assistant is designed around how enquiries actually arrive.
• Rules, gates and audit are designed first and coded second; you review them in writing before the build.
• Provider-neutral architecture that runs on shared hosting and migrates later without a rebuild.
• Clients worldwide under New York contracts with milestone prepayment; paid discovery first.
• You own the code, data and accounts.
• Headquartered in New York City, working with clients nationwide and worldwide.

Related services

See AI chatbot development, AI agent development, AI integration services and AI automation cost.

Frequently asked questions