Voicebots & chatbots (bot flows)

Genesys Cloud bot flows put conversational self-service directly inside your contact centre platform. Because the bot is built in Architect alongside your call flows, inbound message flows and queues, there is no separate vendor to integrate, no second set of credentials and no gap in reporting between what the bot handled and what agents handled. For many common journeys – checking a balance, booking or changing an appointment, tracking an order, resetting access – a well-designed native bot resolves the request quickly and frees agents for conversations that need judgement.

The gap between a bot that helps and one that frustrates is almost entirely in the design. Customers abandon bots that misunderstand them, ask the same question twice or trap them in loops; they accept, and often prefer, bots that understand quickly, confirm what matters and hand over without fuss. QVCCS builds from that standpoint. In Discover we analyse contact reasons, volumes and handle times to find the journeys where a voicebot or chatbot can deliver a complete outcome, then capture the success criteria for each in a Business Requirements Document before any intent is written.

Genesys offers two native options, both built in Architect. Genesys Dialog Engine Bot Flows are integrated into Architect call, chat and message flows, and Genesys documents them as PCI DSS compliant for use in secure call flows. Genesys Digital Bot Flows are integrated into inbound message flows, where turns are typed, conversations can pause and resume, and rich elements such as quick replies, cards and carousels are available depending on the channel. The two have different licensing prerequisites and metering, so we confirm both early. Organisations still on the legacy Genesys Dialog Engine should note that Genesys removed it in 2023 in favour of these bot flows.

At the heart of each bot is the natural language model. Intents describe what a customer wants, each trained with example utterances; slots capture the details needed to act, using built-in types, custom lists, dynamic lists or regular expressions. Confirmations decide when the bot checks before acting, and learning surfaces real utterances for review. With the Genesys Virtual Agent upgrade, generative AI can propose intents and utterances from a description, AI-powered slot types handle free-form input, digital bots can search knowledge articles, and bot conversations can be summarised with a wrap-up code applied. Intent Miner analyses transcripts to suggest intents you may have missed.

Voice and text are different channels and need different designs. A voicebot must keep prompts short, confirm critical details aloud, cope with background noise and accents, and offer keypad input when speech struggles. A chatbot can show quick replies, accept a pasted reference and let a customer return later. Our consultants write a channel-specific dialogue design specification and prompt list rather than reusing one script for both, and decide where a menu or button is clearer than an open question. The result is a bot that feels natural in each channel instead of an IVR in disguise.

Intent design is where most bot projects go wrong. Intents that overlap confuse the model, and intents that mirror an organisation chart miss what people actually ask. We build intent sets from anonymised transcripts, mined intents and search logs, keep them distinct, and use intent health findings to fix weak training. Slots are specified with validation and confirmation rules, and we define what happens on every failure: a second attempt, a clarifying question, a different input method or a handover. Each rule is recorded in the Requirements Traceability Matrix, so every path has a test case and no path ends in a loop.

Escalation is part of the bot, not an afterthought. Genesys bot flows include agent escalation that detects a request for a person without a dedicated intent, confirms it and returns the exit reason AgentRequestedByUser to the calling flow. We design what follows: intent, captured slots, authentication status and a readable summary written to participant data, routing to the right queue with the right priority, and the agent script or workspace view that presents that context. We also plan for systems being slow or unavailable, so a failed data action produces a graceful message and a handover rather than silence.

QVCCS treats a bot as a product delivered by a team, so we muster one from our own bench in which conversation design carries as much weight as configuration. A Senior Business Consultant / Business Analyst leads discovery, mines intents from real transcripts and writes the dialogue design specification. A Solution Architect owns the High-Level Design for intents, slots, data actions, routing and reporting. Senior Developers build the bot flows, surrounding Architect flows and data actions to the Low-Level Design under four-eyes peer review. A Systems Integration Tester builds the utterance test sets that become the regression suite, and a User Acceptance Test Lead involves the agents who receive escalations. The Senior Platform Practice Lead reviews the design.

Testing a bot means testing language as well as logic. Our test strategy combines Architect bot testing, utterance test sets drawn from real phrasing to measure recognition, and a SIT pack that exercises every slot with valid, invalid and ambiguous input and every data action with good, empty and failed responses. Negative testing covers silence, no-match limits, timeouts and missing data. Voicebots are tested on real telephony with varied speakers. UAT scenario scripts involve the agents who receive escalations, so the context they see is proven useful. Only after the go / no-go gate do we launch to a controlled share of traffic.

A bot is a product, not a project, and it needs ongoing attention. QVCCS uses Flow Insights, the Optimization dashboard, the Virtual Agent performance dashboard where licensed, and Architect replay mode to see containment, recognition, slot failures, escalation reasons and drop-off points, then reviews missed utterances with your team on a regular cycle. New utterances are added through learning, confusing prompts are rewritten and new intents introduced as needs change, with the SIT pack rerun as a regression suite before every publish. Containment alone can mislead, so we read it alongside repeat contacts and escalation reasons. When you are ready, we extend the bots with AI Guides and agentic virtual agents in AI Studio.