Kallin AI Transparency Statement
Kallin uses artificial intelligence to help businesses handle telephone calls, website chat and WhatsApp conversations. This statement explains where AI is used, how it supports a business, its limitations and how to ask for human help. The functions available in a particular interaction depend on the business's enabled service and instructions. Kallin is a trading name of Prospective Digital Ltd, a company registered in England and Wales under company number 15564702. Our registered office is Unit 7 Morston Court, Aisecome Way, Weston-Super-Mare, England, BS22 8NG. Contact: info@kallin.co.uk.
1 Knowing when you are interacting with AI
When a Kallin assistant handles your conversation, its replies are generated by AI. Its voice may sound natural and it may use a name chosen by the business, but it is an automated assistant.
We provide an AI disclosure at the start of an AI-handled call and a clear AI identity or opening notice for website and WhatsApp chat. The disclosure must be easy to understand and accessible in the relevant channel. Required wording must not be removed or contradicted by customer scripts or automated instruction updates. If a conversation transfers to a person, the handover should make that change clear.
This statement sits alongside those notices, the relevant business's privacy information and Kallin's Privacy Policy. People should receive the relevant information when interacting with the service, rather than having to discover a policy page for themselves.
2 What our assistants can do
Depending on the business's configuration, an assistant can answer questions using approved business information, capture enquiry details, prepare summaries, support bookings, route a call or message, and carry out supported actions through authorised integrations.
Kallin also supports automated notifications and enabled follow-ups after an interruption. These functions can use approved templates and event-based rules. Kallin Intelligence can review a customer's conversations and update that customer's assistant instructions, as explained below.
An assistant may act only within the authority and permissions set for that business. The standard service does not give it unrestricted access to other systems or permission to make new commercial commitments outside its approved instructions.
3 How a response is produced
Kallin combines third-party AI technology with business information, configured instructions, relevant conversation context and supported integrations. For telephone interactions, this can include speech recognition and a generated voice. For text channels, it includes interpreting messages and generating written replies.
Our voice and conversation platform uses ElevenLabs, including its hosted models and approved downstream AI processing where relevant. Primary responses, fallback, guardrails and evaluation can use different routes. Depending on the route, the service may use models associated with Google, OpenAI, Anthropic or other developers; the model developer is not necessarily the direct recipient of information. For example, a model can be supplied through an intermediary or a cloud provider. The applicable supplier register identifies the actual recipients and verification status. We may add, replace or upgrade models and approved providers, subject to our processing terms, required supplier-change notices and specific written commitments. A possible future provider is not represented as an active recipient merely because its technology is supported.
The system uses the information available in the interaction to generate a response or select a permitted action. Available information may include the business's services, opening hours, approved prices, booking rules and relevant records returned by an authorised connected system. Its access depends on the actual configuration; do not assume every answer has been checked against a current external source.
Replies are generally generated and sent automatically, without a person approving each response. An automated summary is also an AI output and may omit or misstate important details.
4 Accuracy and limitations
AI can misunderstand a request, mishear names or numbers, overlook context, produce an inaccurate summary or generate an answer that sounds confident but is incorrect. Performance can vary with language, accent, speech differences, background noise, incomplete business information and the availability of connected systems.
Please check important details such as dates, contact information, prices and booking requirements. If something appears wrong, correct the assistant or ask how to contact a person. The business should keep its source information current and review material errors. No accuracy percentage, equal performance across every situation or uninterrupted availability is promised by this statement.
Kallin is not an emergency response service or a substitute for medical, legal, financial or other regulated professional advice. Use an appropriate alternative route for urgent or sensitive matters. If an interaction is inaccessible or unsuitable for you, ask the business for another way to contact its team.
5 Human involvement and responsibility
The business using Kallin decides its approved information, permitted actions, escalation rules and contact arrangements. It remains responsible for its products, services and decisions about your enquiry. Kallin remains responsible for its own agreed service obligations and the controls it provides.
You can ask the assistant for the business's available human contact route. Depending on the setup, this may be a live transfer, a callback request, a telephone number, an email address or another support route. Immediate access to a person depends on the business's arrangements, opening hours and availability.
Authorised staff may review records for defined service, support or quality purposes under the relevant privacy arrangements. This does not mean every conversation is monitored live or checked by a person. Automated review by Kallin Intelligence also does not mean a person has approved each resulting instruction change.
6 Recordings transcripts and information used
Depending on the enabled service, Kallin may process audio, caller or sender identifiers, messages, transcripts, summaries, enquiry details, preferences and information from authorised integrations. Share only information needed for your enquiry. Do not provide passwords, authentication codes or payment-card security details in ordinary conversations.
Real-time speech processing, stored audio, retained transcripts and summaries are different activities. Switching off stored audio does not necessarily switch off transcription or other speech processing. The opening notice must accurately explain the activities taking place and provide an appropriate route to privacy information.
A disclosure that a call is recorded or transcribed is not automatic consent. The responsible business must establish the appropriate lawful basis, and any legally required consent must be obtained before the relevant processing begins. Ask the business about its settings and available alternatives if you do not want to use the enabled recording or transcription route.
For customer conversations, the business is normally the data controller and Kallin its processor. Kallin acts as controller for its own activities described in our Privacy Policy, including account administration and our own sales demonstrations. The applicable privacy information explains purposes, recipients, retention, overseas processing and individual rights.
7 Private Kallin Intelligence
Kallin Intelligence is live. Where included and enabled, it reviews eligible conversations within the customer's workspace to identify useful patterns or handling issues. It can generate or apply changes to that customer's assistant instructions on a daily cycle or another disclosed schedule. This improves the instructions used for that business's future enquiries; it does not guarantee an improvement after every review or the removal of all errors.
Private updates stay within approved business information, permissions and escalation rules. They must not remove required notices, expand marketing permissions, change prices or spending limits, enable sensitive processing or give the assistant new authority to take consequential actions. Sensitive personal details and unnecessary identifiers must not be copied into reusable learning instructions. Private instruction updates do not, by themselves, train the underlying general-purpose model or enable wider sharing.
Call Intelligence requires call recording to be enabled. Switching recording off disables recording-dependent call review and stops pending affected review jobs before processing. Switching Intelligence off stops future private reviews and automatic updates, including pending jobs still under our control and new wider contributions dependent on those reviews. Live speech processing, retained transcripts and text channels follow their own disclosed settings.
Applied changes are recorded and remain subject to configured approval requirements. Customers can request correction, removal or a supported restoration. Switching off does not automatically erase earlier records or reverse applied instructions. The business must explain applicable review in its privacy information and establish the appropriate lawful basis and additional conditions for source information and review.
8 Optional wider sharing and model improvement
A separate optional wider-sharing control lets an authorised business administrator choose whether eligible learning examples may help improve models operated by or for Kallin. It is off unless separately enabled after the relevant explanation. It is not required for ordinary AI handling or private Intelligence and does not activate a supplier's own independent training programme. We record the scope, date and relevant notice version of the choice.
Where an applicable model or improvement process is operating, accepted examples can be used for training, fine-tuning, evaluation or testing, and to improve general response instructions, safety and reliability. They can describe a general answer pattern, a handling mistake and correction, or an aggregated error category. Wider improvements can benefit other businesses without exposing the original customer's identifiable records or confidential knowledge.
Contributions must be genuinely anonymous before entering the wider training, fine-tuning or evaluation dataset. Personal identifiers, identifying context, raw audio and identifiable voices, sensitive personal information, confidential customer information and trade secrets must be excluded. This includes names, contact details, account or booking identifiers, health and other special-category information, criminal-offence information, passwords, authentication secrets and payment-card details. An example may be generalised or replaced with synthetic details; it must be rejected if it cannot meet the anonymisation and exclusion standard. Removing names alone or calling a record non-sensitive is not enough.
Preparation and checking can still involve personal data before it is anonymous. That processing must satisfy the relevant lawful basis, transparency, rights and any additional legal conditions. Where Kallin determines the separate preparation purpose and means, it acts as controller as explained in our Privacy Policy; ordinary customer-specific review remains within the customer processor arrangements. An account administrator's preference is not a caller's consent, and supplier or source restrictions must still be respected.
Contributions normally concern eligible conversations collected after wider sharing is enabled. Older records are not included automatically. Any proposed inclusion requires a specific disclosed selection and the relevant lawful permissions and information. Turning wider sharing off stops new contributions and cancels preparation or submission jobs still under our control before submission. Private Intelligence may continue if separately enabled.
Genuinely anonymous examples already incorporated into datasets, and learning already derived from them, may continue to be used. Switching off does not promise to reverse completed training. Identifiable information remains subject to individual rights and retention. If an accepted example is later found to contain personal or excluded confidential information, we must investigate, stop inappropriate use and take appropriate corrective, deletion and model-remediation steps where required.
Under our standard service, identifiable customer conversations and sensitive personal information are excluded from wider model-training, fine-tuning and evaluation datasets. Appointed suppliers must not use identifiable conversations for their own independent general-purpose training. Their contracts, account settings, retention and downstream routes must support this restriction; an affected route must be restricted or stopped if it cannot do so. Necessary operational, support or security processing is separate and remains subject to its lawful limits.
This does not mean that personal or sensitive information can never be processed during an interaction. Necessary content may be processed to answer an enquiry, prepare an authorised summary or perform a lawful private review. The business and Kallin must apply their relevant legal conditions and safeguards. The Privacy Policy, customer instructions and Data Processing Agreement govern retention, deletion and identifiable derived records.
9 Bookings and other external actions
An assistant may request a booking, update or other action only through its authorised workflow. The connected system must confirm success before the assistant presents an action as completed. Pending or failed actions should be described accurately, with a suitable fallback or escalation route.
An AI statement alone is not evidence that a calendar, CRM or other system accepted a change. If the status is unclear, confirm it with the business. AI involvement does not remove rights or responsibilities arising from a business's representations, bookings or transactions.
10 Automated notifications and interruption follow-ups
Staff alerts use generic event notices. Their content, previews and payloads must not include caller names, telephone numbers, conversation text, summaries or sensitive enquiry details. Authorised staff open detailed records through the protected workspace. Delivery still requires a recipient address or device endpoint and limited metadata. Optional staff notifications have their own control, separate from recovery follow-ups. Necessary account and security notices, and authorised messages to an End User such as a booking confirmation, may contain the limited information needed for their distinct purpose.
An enabled recovery feature may send a limited automated message after restoration concerning a recent attempted enquiry affected by an interruption. Its purpose is to help continue that enquiry.
The business controls whether recovery messaging is enabled. Sends remain subject to eligible contact-event information, timing limits, an appropriate lawful basis, channel permissions and individual preferences. Unless the person requests further contact, the standard feature allows no more than one initial recovery follow-up per destination per incident. It cannot recover events that were not reliably captured or guarantee successful delivery.
Optional follow-ups must provide a working way to stop them, including STOP where replies are supported. Turning recovery off cancels messages still queued under our control; a message already handed to a provider may still arrive. Re-enabling the feature must preserve applicable individual opt-outs. An enquiry or a reply is not blanket permission for unrelated marketing, and a telephone call does not establish WhatsApp opt-in.
11 Significant decisions and sensitive deployments
The standard service supports business communications and authorised administrative actions. It is not intended to act as an independent decision-maker for matters such as credit eligibility, employment selection or clinical decisions. A deployment involving legally significant or similarly significant decisions requires its own assessment, appropriate information, lawful conditions and applicable safeguards.
Where the applicable law provides safeguards for a significant decision based solely on automated processing, they can include information about the decision, the opportunity to make representations, human intervention and the ability to challenge the decision. Contact the responsible business about a decision it controls, or Kallin about processing for which we are responsible.
Deliberate sensitive-data processing, children's information at scale, or additional biometric or emotion-related functions require an appropriate assessment and any necessary separate arrangements and disclosures before use. A general transparency statement does not authorise a new purpose or deployment.
12 Questions corrections and concerns
Contact the business you were trying to reach about its enquiry handling, business information, booking or decision. You can ask it to correct an error or provide a human contact route. Contact Kallin at info@kallin.co.uk about our technology, transparency information or an issue involving our own conduct.
When reporting an issue, provide the business name, channel, approximate date and time, and a brief explanation. Share only the information reasonably needed to investigate. We may need to verify identity or authority before accessing or disclosing personal records.
Our Privacy Policy explains individual rights and privacy complaint routes. Where we process information for a customer, we assist the responsible business with relevant requests. Publishing this statement or accepting the Terms does not replace an individual's required permission or remove a legal right.