This week is all self-service. Two changes landed in the Virtual Agent (Bots & IVR) area, both small, both chipping away at the same problem: voicebots and digital bots failing in ways that frustrate customers and are hard for builders to control.
TL;DR
- Genesys' own native speech-to-text (STT) is now selectable as a bot transcription engine in Architect, alongside the Google-, Azure-, and AWS-backed "Genesys Enhanced" options.
- Native STT launches with en-US, es-US, en-GB, and en-AU dialects, and it's the same engine Genesys already uses for post-call transcription.
- Ask for Slot actions now support a configurable skip path, so a customer who doesn't have the requested information can move past the question instead of stalling the flow.
- Skip works with every slot type, including AI-powered slots, and with quick replies, carousels, and list pickers.
Self-Service and Automation
Native speech-to-text as a bot transcription engine
Until now, every transcription engine behind Genesys Dialog Engine Bot Flows was a third-party service wearing Genesys branding. Genesys Enhanced v1 is Google Cloud Speech-to-Text, v2 is Azure Cognitive Services (the current default), and v3 is AWS Transcribe. This release adds Genesys' own first-party STT to the lineup as Genesys Enhanced v4.
Why care about a fourth option? Support, mostly. When recognition quality goes sideways on a third-party engine, troubleshooting involves a vendor hop, and vendor hops add days. With the native engine the whole pipeline is Genesys', which should shorten the path from ticket to answer. Consistency is the other reason: if you already run Genesys native transcription for post-call analytics, putting your bots on the same engine gives you one recognition behavior to tune instead of two.
Genesys buried the most useful detail in the STT engine documentation: custom ASR dictionaries only work with Genesys Enhanced v3 and v4. If your bots need to recognize product names or account jargon, the native engine is one of only two choices that let you teach it those terms.
Dialect coverage at launch is en-US, es-US, en-GB, and en-AU. You can set the engine as the org default or override it per Ask for Slot action, so trial v4 on a single collection step before you bet a whole flow on it. Licensing requires Genesys Cloud AI Experience (or Genesys Cloud 4).
Skip paths in Ask for Slot actions
Anyone who's read bot transcripts knows the pattern. The bot asks for an order number, the customer doesn't have one, and the conversation grinds through retries until someone gives up. This release adds a proper escape hatch: a Skip Response Capture path on Ask for Slot actions. Overdue.
When it's enabled, the customer can skip by saying so in natural language or by tapping a skip button, and the flow continues down a dedicated skip output path instead of failing the step. It works with every slot type, including AI-powered slots, and with the rich inputs digital bots rely on: quick replies, carousels, and list pickers.
The design win is that skipping is an explicit branch. You decide what happens next. Route to an agent, look the customer up by a different identifier, or carry on with the slot empty and handle it downstream.
What this means for your contact center
Neither change forces action. Together, though, they raise the ceiling on self-service quality. STT engine choice has historically been set-and-forget, decided at implementation time and never looked at again. With a first-party engine in the mix and per-action overrides available, I'd revisit it. Recognition accuracy is workload-specific, so the only honest answer is to test with your own call audio and your own dialects. If you're outside the four launch dialects, v4 isn't an option yet.
The skip path is what I'd act on first. Retry loops on unanswerable questions are one of the most common causes of bot abandonment, and the workarounds until now were clumsy: global intents that hijack the flow, or exhausting no-match retries and catching the failure path. A first-class skip branch is cleaner to build and much cleaner to report on. Skips become a signal you can count, telling you which questions customers routinely can't answer. That's design feedback you were losing inside no-match noise.
One caution: don't blanket-enable skip on every slot. A skippable required field just moves the failure downstream. Save it for optional data, or for steps where a solid fallback branch exists.
What to check
- Check which STT engine your bot flows use today (org default vs. per-action overrides) and whether Genesys Enhanced v4 deserves a bake-off for en-US, es-US, en-GB, or en-AU traffic.
- If your bots struggle with domain-specific vocabulary, remember that custom ASR dictionaries only work with Genesys Enhanced v3 and v4.
- Confirm you have the licensing for native STT (Genesys Cloud AI Experience or Genesys Cloud 4) before planning a migration.
- Pull recent bot transcripts and find the Ask for Slot steps with the highest no-match and abandonment rates. Those are your first skip-path candidates.
- When you add a skip path, define the downstream handling explicitly (agent escalation, alternate lookup, or proceed without a value) and add tracking so skips show up in your flow analytics.
- If you use dynamic list slots, slot changes still require republishing the flow.
Full details are in the official Genesys Cloud release notes for May 4, 2026. Questions about what these changes mean for your deployment? Email me.