Answer capsule
BetterUp's May network guide says AI-coaching voice depends on OpenAI, Deepgram, and LiveKit; missing safelist access can produce errors or silent feature failure; and TCP fallbacks may reduce quality or increase latency when preferred UDP paths are unavailable. A successful office demo does not establish participant access. The system buyer should test representative networks, regions, devices, accessibility needs, failure signals, text fallback, privacy, and recovery before launch.
What the source establishes
- BetterUp's guide, updated May 2026, says network configuration should be completed at least two weeks before launch and warns that skipped sections can cause features to fail silently.
- The page says AI-coaching voice relies on OpenAI for voice processing, Deepgram for streaming speech-to-text, and LiveKit for real-time audio connections and turn taking.
- It says OpenAI real-time voice prefers UDP port 3478 with TCP 443 fallback that may affect quality and latency; LiveKit also documents TCP and UDP fallback paths.
- The guide distinguishes US and EEA instances, says all three services must be accessible for normal voice behavior, and identifies unavailable microphones, load failures, and partial audio failure as troubleshooting cases.
Define the coaching task under degradation
Name the population, coaching mode, voice-enabled activity, device and browser support, network locations, regions, languages, accessibility needs, and consequences of interruption. For a role-play or reflective conversation, define what must remain intelligible, private, interruptible, correctable, and recoverable when audio is delayed, clipped, mistranscribed, duplicated, or lost. Decide whether the participant may continue in text, restart, reschedule, or request human help, and what happens to a partial transcript or derived insight. A connected microphone icon is not evidence that the complete coaching interaction works acceptably.
Test the real dependency paths
Use representative corporate, home, mobile, VPN, proxied, TLS-inspected, restrictive, US, and EEA paths. Test allowed and blocked OpenAI, Deepgram, and LiveKit endpoints; UDP success; TCP fallback; high latency; packet loss; device permission denial; backgrounding; reconnection; and one-service-only availability. Preserve endpoint and protocol results, region, device, browser, timestamps, errors, audio and transcript quality measures, and user-observed behavior without retaining unnecessary coaching content. Confirm that a partial service set fails visibly and that support can distinguish configuration from provider outage, device failure, and application defect.
Protect participant choice and confidentiality
Explain before use that spoken audio crosses named external services, which region applies, what is transcribed or stored, who can access content and metadata, and how the participant can decline voice without losing an equivalent coaching path. Verify microphone state, recording or transcription indicators, mute and stop behavior, correction, deletion, and accessibility alternatives. Do not infer engagement, capability, emotion, or coaching progress from connection quality, speech recognition, interruptions, or a participant's choice to use text. Network and device differences can otherwise become hidden eligibility filters or misleading behavioral signals.
Accept recovery, not just connectivity
Set thresholds for connection success, latency, intelligibility, transcription accuracy where used, interruption recovery, failed-session visibility, support response, and safe fallback. Track failures by network, region, device, disability or accommodation need where lawful and appropriate, and activity without exposing private coaching content. Recheck after endpoint, IP range, subprocessor, protocol, app, network policy, or regional configuration changes. BetterUp's guide gives unusually specific architecture evidence; it does not establish a buyer's configured path, participant experience, coaching quality, accessibility, privacy compliance, or outcome.
Turn this source into a reviewable decision
For AI Coaching Platforms for Leadership Development, use this briefing as a dated decision record rather than a substitute for the source. Preserve Network Configuration Guide: Email, Video, & AI Coaching, the exact URL, the September 14, 2026 review date, the supported facts above, the editorial interpretation, the limitations, and any buyer-specific evidence. Link that record to the decisions most directly affected: Workflow and identity integration; Accessibility, language, culture, and user control; Data flow and confidentiality; Safety, boundaries, and escalation. State whether the source changes the scope, evidence requirement, control, sequence, or only the language used to describe the decision.
Before action, name the accountable owner, affected population and workflow, exact offering or configuration, source data and rights, human decision point, exception and appeal path, complete cost, expected benefit, failure and stop conditions, retained evidence, and next review date. Keep official facts, provider statements, buyer observations, representative tests, measured outcomes, editorial inferences, and unknowns visibly separate. Reopen the record when the source, offer, model, integration, data, policy, population, responsible person, or measured result changes.
Limitations and unknowns
BetterUp is the provider source. Its May 2026 technical guide documents network prerequisites, regions, named external services, ports, protocols, fallbacks, and possible failure behavior, but does not establish a buyer's firewall state, endpoint availability, observed audio or transcript quality, participant notice and consent, accessibility, data retention, incident performance, coaching validity, or outcome. Endpoint and IP details can change and require current verification. Configured network and regional evidence, representative device and degraded-path tests, participant notices and choices, support and incident records, and qualified IT, security, privacy, accessibility, procurement, coaching-program, works-council, and legal review control.
Decision test
Ask whether the source changes the decision itself, the evidence required, the implementation sequence, or only the language used to describe an existing capability. Record which claims are directly supported, which are provider statements, which require an independent test, and which remain unknown. A source-linked review should make uncertainty easier to see, not bury it inside a blended score.
Questions to take into review
- Where does coaching appear and what data or actions flow through HRIS, collaboration, calendar, email, and identity systems?
- Can the intended population understand, use, contest, pause, correct, export, and leave the experience?
- What does the system ingest, infer, retain, share, and expose to coaches or administrators?
- How does the system respond when coaching is unsuitable or a person discloses harm, crisis, discrimination, legal, medical, or employment issues?
The publication supports research and executive decision preparation. It does not provide legal, financial, accounting, employment, clinical, cybersecurity, investment, procurement, or implementation advice.