This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.
Decoding Dialogflow: Integrating CCAI with Contact Centers
Although Google hasn’t announced when Contact Center AI will become generally available, I fully expect certain parts of CCAI to be GA by the end of the year. At that time, the floodgates will open as companies will begin using CCAI to implement self-service customer experiences on the Google platform. Any company wishing to use CCAI capabilities will need to purchase them through one of Google’s CCAI contact center partners, pictured above.
In this article, the ninth in a series, I’ll examine how two of these partners integrate with CCAI and how they intend taking CCAI to market.
When it comes to using CCAI with Genesys contact centers, Genesys plans to provide a curated experience. A curator, as initially defined in Old English, is a person charged with the care of souls. Thus, Genesys intends to provide professional services such as technical guidance, design assistance, and deployment oversight for customers that adopt CCAI as part of their Genesys contact center strategy.
Genesys hasn’t announced CCAI pricing; however, CCAI will be purchased on Genesys paper and orchestrated by Genesys’s customer care organization. In most cases, pricing will be close to Google’s pricing for Dialogflow interactions, but Genesys will bundle in all required elements, including the CCAI connectors and the agent assist knowledge base. Although Google prices Dialogflow on a per-interaction basis, Genesys will price CCAI on a per-agent seat, per-month basis.
CCAI has several key capabilities, including virtual agent, agent assist, sentiment analysis, and conversational topic modeler. Genesys will create packages that include one or more of these functionalities based on customer need.
Genesys looks at bots and intelligent virtual agents as part of the routing experience, as opposed to them being separate from traditional contact center routing. A simple four-step architectural block diagram illustrates how Genesys contact centers integrate with CCAI.
Inbound voice calls are anchored in Genesys’s telephony front end. The system can also accept text via messaging apps or integrations. For this discussion, I’ll focus on voice calls.
Prior to the call even being established, Genesys is monitoring customer activity if the person is browsing on an instrumented webpage or using an instrumented messaging app; monitoring may also include gathering data from CRM and marketing automation systems. The Genesys system is examining the customer journey, trying to identify characteristics about that customer. Characteristics may be previous interactions, recent purchases, replies to a marketing campaign, responses to an IoT sensor, etc.
Multiple Bot Entry Points
Using this customer information, the business logic determines how to route the interaction to one of a number of entry points that a bot can handle. Multiple entry points means that the customer’s initial experience with the bot is contextually aware; this can be the difference between a bot greeting the customer with an open-ended question like “How may I help you?” or a more directed response such as “It looks like you were comparing pricing plans, I can assist with that.”
When the bot is invoked, the caller’s voice is digitized and streamed in real-time to Dialogflow using the Google gRPC protocol I discussed in a previous article. Interactions between the bot and the customer can occur within Dialogflow until it deduces the caller’s intent. Once the bot has determined the intent, then the system needs to determine how to fulfill what the user wants.
The Fulfillment Decision
The contact center owner and its bot designers and developers will need to decide where fulfillment will occur. The Genesys team anticipates that most customer fulfillment activities will occur within its contact center system, because it already has a variety of integrations with third-party systems. Reimplementing them within Dialogflow may be a bit like reinventing the wheel if your organization already uses any of these integrations, Genesys reasons. The Google Dialogflow team, on the other hand, believes that many CCAI customers will choose to use Webhook functions launched from within Dialogflow to fulfill a customer’s needs.
Where fulfillment takes place doesn’t really matter; however, there are some things to consider. As already noted, integrations are one factor; another is regulatory compliance. If you go outside of the Genesys system, do you remain HIPAA, PCI, or SOX compliant?
Once fulfillment occurs, then the Genesys system can complete typical contact center functions, shown above in the “Post Processing” step. These may be actions such as creating post call notes, logging the customer intent into a database, summarizing and/or transcribing the conversation, surveying the customer for Net Promoter Score ratings, etc.
This description is somewhat simplistic in terms of what happens in a Genesys-Dialogflow/CCAI bot interaction. Lots of things can go on under the hood; nevertheless, this describes the interaction at a high level.
However, there’s one item not yet covered: passing a call from Dialogflow to a live agent.
Passing a Call to a Live Agent
One of the advantages of using CCAI over just using Dialogflow is that the new CCAI APIs allow more granular control over Dialogflow’s processing during an interaction, making it much easier and more natural for the bot to transfer control to a live agent. Should the bot fail to understand the user’s intent, it can bring the live agent in, and along with transferring the call, the bot sends the agent a transcription of the interaction between the customer and Dialogflow so that the agent can become familiar with what the customer was trying to say or accomplish by interacting with the bot.
Once a live agent receives the call, agent assist, the other Genesys-supported CCAI capability, is available to provide context and suggestions allowing the agent to better respond to the customer’s need. In the example below, agent assist is prompting a live agent with real-time information on how to respond or help the customer purchase a new pair of shoes.
Which Genesys Contact Centers Work with CCAI
Genesys customers can expect to see CCAI available on all three of the company’s contact center systems — PureCloud, PureConnect, and PureEngage — in the near future. But when CCAI goes GA, Genesys will launch its first CCAI integration with PureEngage. Two CCAI use cases will be available: creating virtual agents and agent assist.
Genesys has made changes to all three of its platforms to enable streaming of audio to Google CCAI. Depending on the particular product line and some deployment decisions, an additional component may be required to stream the audio from an on-premises platform to the Google Cloud.
Designer: The Genesys Virtual Agent Development Environment
Genesys has a development tool called Designer you can use to implement your contact center’s business logic. One of Designer’s advantages is that it can interface with Dialogflow and import intents and entities created in Dialogflow. From Designer, you can launch different entry points into Dialogflow based on pre-routing logic, decide how to fulfill intents, and make the Dialogflow/CCAI capabilities integrate smoothly with the rest of your contact center application.
Cisco’s first CCAI-integrated solution, Cisco Virtual Assistant (CVA), will be included in its forthcoming 12.5 Customer Voice Portal release in Q1 2020, shortly after Google formally announces CCAI’s general availability.
Cisco Answers is a second Google-based solution Cisco will offer. This solution will use CCAI’s agent assist capability. Cisco Answers will be available sometime after CVA.
Cisco hasn’t finalized commercial terms for either CVA or Cisco Answers.
Architecturally, CVA will leverage Dialogflow much like that shown above for the Genesys integration. I’ll focus on Cisco Answers for the rest of the architectural discussion.
Upon receiving a call or text-based communication, the Cisco contact center will apply pre-routing business logic to the interaction. This will involve identifying who’s calling or sending the message using caller ID, app authentication, or cookies. Based on this information, the system will retrieve customer journey information, including recent interactions, purchases, etc. The interaction would then hit the IVR, for routing to an appropriate agent.
When the IVR routes a call or text message to a live agent, Cisco Answers will fork the interaction to Dialogflow/CCAI. In the case of an audio call, Dialogflow’s speech-to-text capability will convert the caller’s voice to text. CCAI’s agent assist functionality will evaluate the text stream using natural language understanding to predict what information the customer is seeking. Based on this prediction, CCAI will search the customer-specific knowledge base for articles or information appropriate to the interaction. This information will pop up on Cisco Finesse, a dashboard agents use for interacting with callers. Agents will be able to examine the information Cisco Answers provides and respond more rapidly or more completely to the caller’s question than they would be able to do without the information. CCAI/Cisco Answers will continuously update the information sent to the agent dashboard as it follows along with the conversation.
Once the call or interaction ends, the system will complete typical contact center functions in a Post Processing step.
Using Click-Throughs to Train the AI Model
Clicks on information Cisco Answers presents in the Finesse dashboard get sent to CCAI to refine machine learning predictions on which information is most useful to customers. In future interactions on the same topic, CCAI will give more weight to articles and information that have received clicks than to those that don’t receive clicks, implying that the information surfaced and clicked on was of value to the agent. Click-throughs from more experienced agents will receive more weight than clicks from newer agents, which assumes that the experience from these more skilled agents will cause them to progressively gravitate to the more relevant information. Over time, Cisco Answers will get better and better at predicting the information to surface because the agents will be continuously providing feedback on which information on a given topic is more valuable via this click-to-train method for updating the AI algorithm.
Which Cisco Contact Centers Work with CCAI
When CVA and Cisco Answers become generally available, they’ll initially be offered only on Cisco Contact Center Enterprise. The roadmap calls for Cisco to develop CCAI capabilities for Webex Contact Center next, and then for Cisco Contact Center Express.
I’ve covered only two of the CCAI partner contact center integrations in this article. Writeups of other partner integrations will occur in future articles in this series.
- Although the CCAI capabilities provided by Google are the same for each CCAI partner, how each partner chooses to use these and surface the results in their contact center offerings will vary widely.
- Each contact center vendor has its own development environment for implementing the business logic in its contact center; how the vendor invokes CCAI capabilities from within its own environment will be unique.
- A contact center partner can create mechanisms to automatically tune the machine learning parameters so that agent assist can surface the most applicable information.
- Developers can incorporate business logic plus agent assist in scripts to make interactions easier for the customer and for the live agent.
In the next article, I’ll discuss issues around bots knowing when it’s time to hand off an interaction to a human agent and some of the considerations when testing an intelligent virtual agent and going live with it.
Learn more about AI in the contact center at Enterprise Connect 2020! Register today using the code NOJITTER to save $200 off current rate.