This release comes with three major enhancements to the Service Cloud:
- Einstein Bots for Service
- Lightning Flow for Service
- Einstein Next Best Action
Einstein Bots for Service is providing the ability to easily configure chat bots that enable instant response to customers and a seamless handoff to customer service agents.
Lightning Flow for Service gives companies the ability to automate processes with contextual, step-by-step guidance for fulfilling requests and resolving issues, using a graphical interface.
Einstein Next Best Action is delivering intelligent recommendations and offers on any channel to increase customer satisfaction.
While Einstein Bots for Service and Lightning Flow for Service are in General Availability since July 11, 2018, Einstein Next Best Actions will remain in a Pilot phase for some more time. The reason for this is that Salesforce wants to be double sure that this functionality is reliable. It needs a good amount of data and a good training set. And Salesforce cannot look into the data.
The bots themselves do need to get trained and, once active, take feedback from the service agents.
All three features work hand-in-hand. Salesforce uses a credit card scenario to make this point. When a customer goes to the web site for help the chat bot takes over and gathers the necessary contextual information and then escalates the issue to a customer service agent who continues the chat at the position the chat bot exited with all information available. A Lightning workflow then guides customer and agent through the resolution process. This is also where Einstein Next Best Action comes into the picture. Based upon the customer’s history and the conversation Next Best Action provides the agent with a tool to suggest offers matching the customer’s profile.
According to Bobby Amezaga, Senior Director Salesforce Service Cloud Product Marketing, all three innovations are about helping Salesforce customers to “create the digital service experience they [their customers] are asking for” and to be able to provide a guided experience by connecting data.
The Bigger Picture
This triplet of interacting features intends to solve a dilemma facing companies.
Customers expect that they are known to the company and that their current interests are acted upon, using the communications channel of their choice, and across communications channels.
On the other hand employees need to become more productive while following a trusted process. In addition, with an increasing number of millenials in the workforce, it is increasingly necessary to also provide a better than just good user experience. This is especially true in service center scenarios where we often see young people and a high attrition rate.
Bot capabilities are limited and will continue to be for some time going forward. They are powered by rule based systems and narrow AIs. We do not see anything that is close to a general AI. Still, employees are in fear of AI as a technology. They fear that their jobs are moved to the machines.
As a consequence of the need to do more with less and the employees’ fears it is double necessary to have bots and human agents work hand in hand instead of in competition.
This is also true for customers that are exposed to the bots. It is still necessary for them to know whether they are interacting with a bot or with a human. This might change over time with humans becoming increasingly used to interacting with chat bots, but for now it is a matter of ‘etiquette’ to identify what is bot and what not. The recent fierce discussion about Google Duplex and its capabilities made that clear abundantly, one more time. After all, humans introduce themselves, too. So, why shouldn’t bots do the proper thing …
The narrow nature of the tasks that a single bot can perform as per now also makes it necessary to easily build and maintain bots, ass well as monitor and improve their performance. It needs swarms of bots that are interacting with each other and with human agents – and that continue to ‘learn’ on the job.
Last, but not least, there is the matter of training the AI. To deliver accurate results it needs a well performing training set. Which for time being makes companies rely on data scientists to create this data set. This is especially important for prescriptive scenarios like Next Best Action. These scenarios also need a lot of data that in all likelihood does not lie in a Salesforce database. It is here, where the acquisition of Mulesoft has a good chance of paying dividends. The Integration Cloud, as it is named now, enables customers to enrich Salesforce data with data coming from a plethora of different sources.
My PoV and Analysis
With this release Salesforce reinforces its strategy of embedding AI and Machine Learning directly into the application. This is also where it belongs and similar to the strategy that also Microsoft and SAP are pursuing. AI for AI sake is not a winning proposition. As Marco Casalaina, VP Product Marketing Einstein put it “AI alone does not bring your business forward”.
We see a seamless integration that provides a handover from bot to service agent, along with a sense of important details. The bot, for example, introduces itself as a bot. This helps building trust.
The bots itself being built using rules and Natural Language Processing (NLP) suggest an ability to not only escalate from bot to service agent but also an ability to hand-off from one bot to another. If not, this is surely something that I’d encourage Salesforce to look at, as this ability will further reduce the strain on service agents and help them focusing on the tough issues that will continue to need human to human interaction.
Additionally, and this is a bit more challenging, companies will benefit from pre-trained intelligences and from a transparent method that improves solution accuracy and broadens the covered scope on an ongoing basis. Right now Salesforce helps customers with a service offered by their own data scientists. Additionally, some partners offer pre-trained models. The bots are learning on an ongoing base only by feedback that is given to them by agents.
It should be interesting to see two measures being implemented.
Bots running as sidekicks to agents while they handle issues that are handed off to them. This will help in increasing accuracy and broadening a bot’s skill.
Secondly, I propose an engine that identifies a training set for resolving a pattern of issues and then trains an AI with it. This has the potential of building bots more efficiently.
The combination of these would help training a group of bots until they reliably reach a minimum accuracy and helps them maintain this accuracy in a changing world.
I’d love to see both of these in action, especially in combination.