1. You are interacting with AI
The simulated customer in a DIRI AI training session is artificial. The interface and spoken/text opening must make this clear no later than the first direct interaction. Synthetic speech, generated replies, transcripts, scores, reports and coaching suggestions may be produced or assisted by AI.
A simulation is not a real customer's statement and must not be represented as one. Exported synthetic audio or content must retain an appropriate AI/simulation disclosure where context could reasonably mislead a person.
2. What the system does
Depending on the selected function, DIRI AI may:
- transcribe a trainee's speech;
- generate the simulated customer's next response;
- render a synthetic voice;
- analyse the conversation against coaching criteria;
- calculate or generate scores, summaries and suggested improvements; and
- aggregate workspace training activity for authorised administrators.
Model and voice processing routes are described in the Model Provider Disclosure. A provider's model language list is not a promise that DIRI AI offers or quality-approves every language.
3. Limitations
AI output is probabilistic. It can misunderstand accents, noise, irony or context; invent facts; apply criteria inconsistently; omit important conduct; or produce a plausible but wrong suggestion. Voice naturalness varies by selected voice and its native language/accent. A numerical score is not an objective measurement of a person's worth, intent, emotion, protected trait or future job performance.
DIRI AI does not promise a sales result, compliance outcome or improvement. Users must review important feedback and compare it with the actual transcript/audio where available.
4. Human oversight and challenges
Scores and reports are coaching aids. They must not be the sole basis for hiring, firing, discipline, pay, promotion, access to an essential service or another legal or similarly significant decision. An organisation must designate a qualified human who can understand limitations, review context, disregard an output and correct consequences.
A user may flag an inaccurate transcript, score or report and request human review through support@diriai.com. DIRI AI may correct the record, explain the general factors, rerun a report where appropriate or mark the output disputed. We may not disclose system secrets or security controls, but that does not remove applicable explanation or data-protection rights.
5. Uses DIRI AI does not provide
DIRI AI does not provide functions that:
- recognise emotions in a workplace or educational institution;
- perform biometric identification, voiceprint matching or biometric categorisation;
- infer race, health, disability, religion, politics, sexual orientation or other protected traits;
- make a solely automated employment or other high-stakes decision;
- secretly monitor a worker or real customer; or
- diagnose, provide emergency services or replace regulated professional judgement.
Customers must not convert a general coaching tool into a prohibited or high-risk system through their scenario, integration or decision process.
6. Data and model improvement
DIRI AI sends only the information needed for the requested model function to a listed provider. Customer Content is not used to train general-purpose models unless a separate, voluntary, specific choice and lawful basis exist. Applicable provider retention, processing locations and settings are recorded and disclosed as described in the Model Provider Disclosure and Subprocessor List.
Controlled quality review should use fictional, minimised or de-identified examples wherever possible. Optional use of identifiable recordings requires an appropriate notice and legal basis.
7. AI literacy and organisational responsibility
DIRI AI provides this Notice, product guidance and limitations so users and administrators can exercise informed oversight. Organisations remain responsible for role-appropriate training, workplace notice, correct interpretation, escalation and any AI-risk or impact assessment required for their use.
8. Contact and version
Questions, disputed output and accessibility needs: support@diriai.com. Privacy questions: privacy@diriai.com. Version: 1.2.0. Date: 30 August 2026.