Speech and Text Analytics in Genesys Cloud: Turning 100% of Conversations Into Business Intelligence

Speech and text analytics can help organizations find patterns across customer-agent interactions that are difficult to spot by reviewing calls one at a time. In Genesys Cloud, getting useful results depends on more than switching analytics on: transcription, programs, topics, privacy settings and analysis methods all need to be configured to match business goals. Here is how to approach that design discipline and turn conversation data into decisions.

1. Start with the decisions analytics should support

Genesys Cloud speech and text analytics provide automated analysis across 100% of customer-agent interactions. Capabilities include voice transcription, sentiment analysis, topic trends, agent empathy evaluation and interaction categorization. Together, they can help teams identify areas of concern, opportunities for improvement and examples worth recognizing.

The value comes from connecting those capabilities to business questions. For instance, a service team might track cancellation reasons to identify product issues, while a quality team may use empathy analysis to guide coaching. Custom interaction categories can add detail based on an organization’s criteria.

Genesys documentation also describes related tools that extend the picture:

• Dictionary management lets organizations add specialized terms to improve transcription of organization-specific language.

• AudioHook Monitor streams conversation audio to third-party services for uses such as voice biometrics, transcription, recording and Agent Assist.

• Process Automation Triggers can use topics, sentiment and empathy conditions to initiate workflows when defined criteria are met.

Define the intended decisions before configuring analytics. A dashboard becomes more useful when it is tied to a response, such as changing a route, fixing a product issue or coaching an agent, rather than treated as an end in itself.

Actionable takeaway: Select a small set of business questions and measurable outcomes to guide the initial Genesys Cloud analytics design.

2. Get transcription, timing and privacy right

Voice transcription turns interactions such as ACD calls into speaker-separated text for analysis. Genesys documentation lists several licensing paths, including a Genesys Cloud EX license; CX 1 or CX 2 with the Speech and Text Analytics Add-On; CX 1 WEM Add-on II, CX 2 WEM Add-on I, CX 3 or CX 4. With a CX license, call recording must also be enabled through recording policy.

Transcription availability depends on the telephony connection. With Genesys Cloud Voice or BYOC Cloud, transcription is near real time and generally appears in the interface or API within minutes after the interaction ends. With BYOC Premises, transcription starts after the recording is completed and uploaded. Genesys describes this as usually taking about half the interaction length (for example, around five minutes for a ten-minute call) so it does not support real-time use cases. BYOC Premises recordings must be dual-channel and use PCMU, PCMA, L16 or opus codecs.

Genesys provides three transcription engine options: default Genesys Cloud Native Voice Transcription, Extended Voice Transcription Services (which supports more dialects and languages with a fair-use allocation per licensed user), and a Transcription Connector for a third-party provider, which has a per-minute charge. Engine selection is made for each dialect within a program, and changes must be published to take effect. Agent Assist with Google Cloud overrides the Native and Extended engines.

Configuration also affects when transcription runs. Genesys recommends enabling transcription in speech and text analytics settings, for the relevant queue, and through call-flow actions. An EnabledGlobally API setting can start transcription at the beginning of supported inbound, outbound and direct-call types; however, transcription turns off when a call reaches a flow or queue unless that flow or queue configuration enables it. Global enablement also does not let administrators choose individual numbers or campaigns.

Privacy settings matter. When suppression is enabled for IVR, hold or queue-wait segments, transcripts are not generated for those segments, and summaries exclude that content. Genesys recommends suppression to enhance customer privacy. Teams should also confirm language settings at the Edge trunk, in Architect flow settings and on the voice call’s media.

Actionable takeaway: Validate licensing, recording, call paths, codecs, language and suppression policy before relying on transcripts for analytics or workflows.

3. Teach the system what to listen for

Programs, topics and phrases form the configuration hierarchy that guides topic detection. A topic represents a business-level intent or subject. For example, a “Cancellation” topic could include phrases such as “close out my account” or “I wish to cancel.” Programs group topics and map them to queues or flows, allowing analytics to reflect differences in customer intent, region or operational requirements.

Phrases describe the different ways a topic may appear in conversation. They can be added individually or in bulk through file upload. Genesys documentation notes that phrases, including organization-specific words outside the standard dictionary, are added to the language model to support recognition. Out-of-the-box topics can provide a starting point, but organizations should adapt them to the language their customers and agents actually use.

Programs can also control whether customer sentiment and agent empathy analysis run for associated interactions. That flexibility can be useful when different teams or regions have distinct analytic needs or compliance requirements.

Actionable takeaway: Build programs around meaningful queues or flows, then seed each topic with phrases drawn from real customer and agent language.

4. Interpret sentiment as a signal, not a verdict

Genesys Cloud calculates sentiment from the generated transcript. It classifies customer phrases as positive, negative or neutral and assigns magnitude-based scores. Sentiment markers appear at the time a phrase begins in the interaction waveform and transcript, with a tooltip showing the phrase. Only customer phrases are analyzed; agent phrases are ignored.

The overall customer sentiment score ranges from -100 to +100. Importantly, it gives more weight to events near the end of an interaction, reflecting the customer’s parting experience. The sentiment trend compares the first and second halves of the interaction. Its five labels range from Improving (change greater than +55) and Slightly improving (+20 to +55), through No change (-20 to +20), to declining categories below -20; a change below -55 is classified as Declining. Genesys notes that a trend usually requires six or more customer phrases.

Context matters: the same phrase can receive different scores in different interactions because surrounding phrases influence interpretation. Sentiment analysis works out of the box, and optional sentiment feedback is intended to correct model mistakes. Genesys cautions that low-quality audio can produce missing or mislabeled markers, so managers should listen to interactions with numerous negative markers before drawing conclusions. Content Search also allows managers and agents to find interactions using sentiment filters.

Organizations with EU exposure should review the current Genesys FAQ about how the EU emotion AI ban affects customer sentiment and agent empathy analysis, and assess applicability with their compliance teams.

Actionable takeaway: Use sentiment to prioritize review and identify trends, then validate high-impact findings against the interaction and its context.

5. Discover missing topics and revisit historical data

Topic Miner helps teams find what customers are discussing before deciding which topics to formally track. It analyzes historical voice and digital transcripts, including chat, call, email and message interactions, and returns candidate topics and phrases. Teams can create, duplicate or delete miners, mine external transcripts and use ignore lists to refine suggestions. Mined candidates can then inform production topics. Topic Trends Summary and Topic Trends Detail views help teams examine how configured topics change over time.

After adding or changing a topic, administrators can use interaction reprocessing to apply the updated configuration to historical interactions. Genesys announced this capability to let teams select relevant programs, set filters such as media type and date range, and publish a job that retroactively labels past interactions. Analysts can then use the refreshed data for new dashboards and investigations.

Genesys recommends limiting reprocessing to programs tied to updated topics, using date and media filters to manage volume, and exporting interaction IDs for auditing. Its documentation says billing is currently inactive and the feature is free until further notice, with 30 days’ advance notice before any future charges.

Actionable takeaway: Mine transcripts to identify candidate topics, validate and publish the useful ones, then reprocess a targeted historical period to compare past and current trends.

6. Turn configuration into operational decisions

A reliable analytics program is iterative. First, confirm the telephony path: Genesys Cloud Voice and BYOC Cloud support near-real-time transcription, while BYOC Premises transcription follows recording upload. Next, choose a clear suppression policy for IVR, hold and queue-wait segments. Then organize programs around business intent rather than relying on a single catch-all configuration.

Use customer language to create and refine topics; Topic Miner can help surface candidate phrases. When reviewing sentiment trends, remember that the overall score gives more weight to the end of the call. Finally, use reprocessing to apply relevant topic changes to historical interactions and build a more consistent view of long-term customer experience.

The purpose is action. A rise in billing questions may point to a confusing product change. Repeated negative sentiment after a queue transfer may warrant a routing review. A recurring topic may reveal a coaching opportunity. Those are the kinds of decisions that transform cloud telephony analytics from reporting into business intelligence.

Actionable takeaway: Assign an owner and follow-up action to each priority insight (such as a routing change, product fix or coaching plan) and review whether the change improves the relevant trend.

Dunamis Consulting helps organizations plan and deliver cloud communication solutions, including Genesys Cloud initiatives. Our team brings 15 years of experience in cloud telephony and can support consultation and gap analysis, cost analysis, project scheduling, flexible staffing solutions, bulk hour purchases, and managed and support services. If your organization is planning a migration or refining its analytics foundation, explore our 5-step migration framework and contact Dunamis Consulting to discuss your project.

For platform configuration details, see the Genesys Cloud Speech and Text Analytics documentation, Topic Miner guidance and Reprocess Interactions announcement.