AI should be useful, bounded and visible.

This is JMB’s public baseline. Every material AI call journey still needs a use-case review covering disclosure, lawful basis, permitted data, tools, failure routes, provider boundaries and human escalation.

1. Our approach

CallPlatform uses artificial intelligence and machine-learning services to help answer, route, understand and improve business communications. We design these features to work inside dependable telephony, not to hide uncertainty or remove human accountability.

We favour the least open-ended method that can solve the caller’s problem: fixed audio for a fixed message, deterministic routing for a known decision, structured speech for a short choice, and a conversational AI Agent where a genuine multi-turn conversation adds value.

2. Distinct AI and speech products

AI Voice Agents

Conversational agents can follow configured instructions, use approved knowledge, collect details, call permitted tools and transfer to a person or service. They are probabilistic systems and can misunderstand, hallucinate or take an unexpected conversational path.

Speech-enabled Flow Builder

Speech Input recognises a short English response and maps it to configured intents; Speak Text generates a dynamic message. These nodes are not open-ended agents. They use explicit retry, confirmation, timeout, error and fallback routes.

Call Intelligence and assistance

Where enabled, AI can help analyse calls, identify themes or sentiment, apply a configured quality rubric, summarise information and assist an author with an initial flow or instruction set. These outputs support human review; they are not automatically objective facts.

3. Where AI may be used

• Conversation: generating and understanding turns in an AI Agent interaction.
• Speech: generating audio from text and transcribing short caller input or a recording.
• Knowledge retrieval: finding relevant passages from content a customer has approved for an agent.
• Author assistance: proposing an initial call-flow structure, instructions or analysis from a user’s brief.
• Post-call processing: transcription, summarisation, tagging, sentiment, quality and coaching workflows where enabled.
• Diagnostics: assisting an authorised user to understand reports or technical call evidence.

Ordinary routing rules, billing, account ownership and security decisions should remain deterministic unless a customer has deliberately designed and reviewed a different approach.

4. Providers, models and data flows

CallPlatform can use specialist third-party providers for speech generation, speech recognition and language models. Current product configurations may include ElevenLabs and selected language-model providers. The exact model, voice, region, retention option and data path depend on the feature, estate and customer configuration.

A dedicated CallPlatform estate isolates the core platform components; it does not automatically make an external carrier, AI provider, email service or customer API part of that estate. Those processors and transfers must be identified in the solution design and contract.

JMB does not operate a public foundation model trained on customer calls. Third-party provider handling is governed by the provider configuration, its terms and the customer/JMB agreement. Customers should not submit data to an AI feature unless the intended provider and data use are appropriate.

5. Controls and safeguards

Depending on the feature, CallPlatform provides controls such as:

• approved instructions, knowledge sources, tools and account-scoped permissions;
• visual routes for human transfer, queueing, voicemail or another deterministic service;
• explicit no-input, no-match, attempts-exhausted, timeout and error outcomes;
• confirmation before a sensitive structured speech choice;
• simulation, test calls, service version history and restore;
• call history, journey events, outcome reporting and quality review; and
• feature entitlements and estate-level enablement.

Routine speech-journey reporting records operational outcome and timing rather than presenting the caller’s short transcript. Recording and full transcription are separate features with their own configuration and retention decisions.

6. Caller disclosure and human access

Customers are responsible for giving any disclosure required by law, contract or reasonable caller expectation, including when a caller is interacting with an automated agent, when a call is recorded and how personal data will be used.

Important journeys should provide a practical route to a person or another safe service. The right fallback depends on operating hours, risk, caller needs and the action being automated.

7. Limitations

AI output can be inaccurate, incomplete, biased, delayed or inconsistent. Speech recognition varies with accent, language, codec, background noise, mobile coverage and prompt design. External providers and network connections can fail.

CallPlatform AI must not be treated as emergency service access, a human identity check, or professional medical, legal, financial or safety advice. Sensitive or irreversible actions require an appropriate deterministic check, confirmation, human approval or provider-supported idempotency design.

8. Customer responsibilities

Customers decide the purpose of their agents and call flows. They are responsible for lawful data collection, caller notices, content accuracy, tool permissions, connected systems, retention, testing, monitoring and sector-specific requirements.

JMB can help design a narrower, measurable first use case and review the technical failure paths. This does not replace the customer’s legal, compliance or professional advice.

9. Questions and concerns

Email: [email protected]
Privacy: [email protected]
Address: JMB Technology Ltd, 20 Wenlock Road, London, England, N1 7GU, United Kingdom