不要将各种人工智能技术混为一谈。它们之间存在差异，本文将以服务中心技术为例进行说明。

我们看到服务中心外包市场发生了一些戏剧性的变化，这是由于客户期望的巨大变化和技术的指数级进步所致。不久之前，服务中心管理人员主要侧重于降低成本和提高生产力。这种思维很快发生了改变，原因如下：

l   客户人口组成正在改变。千禧一代现在拥有强大的购买力，他们希望在与品牌交互时能够获得非常不同的体验。他们希望跨实体商店、移动应用程序和Facebook Messenger获得无缝、全方位的体验。（译者注：Millennials，千禧一代，1984-2000年出生，差不多伴随着个人电脑和互联网[万维网]一同成长。）

l   品牌切换成本非常低，通过移动应用程序从一个品牌切换到另一个品牌非常容易。公司正试图通过创建个性化的体验——在正确的时间和正确的地点为客户提供正确的产品和服务——找到提升客户粘度的方法。

l   对话式人工智能（AI）已准备就绪。在过去的一年里，能够理解人类意图并进行类似人类交谈的技术有了显著改善。服务中心主管指出，对话式AI正在迅速超越原型并进入产品生产阶段。最近的ISG研究强调了这一点：在未来两年内，虚拟代理和聊天机器人的使用将增加一倍以上。

我们看到，企业买家对聊天机器人、虚拟代理和智能问答的兴趣显著增加，我们将这些技术统称为“智能助手”。同时，我们看到，人们对它们之间的差异颇为困惑。由于技术和服务供应商以不同的方式使用这些术语，因此很难确定哪家供应商可以解决手头的特定问题（这听起来很像几年前的云问题）。

尽管我们仍处于采访服务中心主管和研究市场上的各种产品和供应商的阶段，但我们发现的一些共同点促使我们将这些供应商归入某种类别。我们目前的研究范围涵盖聊天机器人、虚拟代理和智能问答——稍后介绍语音助手。

我们发现可以用两个因素区分这三种技术：（1）它可以执行多少任务；（2）它可以在多少渠道上运作。任务可以很简单，比如获取新闻；也可以比较复杂，比如询问未完成的订单。渠道是指发生对话的方式，例如电子邮件、移动应用程序、聊天平台（如Facebook或Slack），甚至是独立设备（如Amazon Echo或Google Home）。

大多数情况下，聊天机器人在少数几个渠道上处理单个任务。这些任务不需要大量的上下文信息，个性化或共鸣。聊天机器人擅长诸如企业品牌推广和产品销售（例如，Burberry在Facebook Messenger上的聊天机器人）等任务，或者说它们擅长做单一的事情（例如使用x.ai作为你的个人会议助理）。

另一方面，虚拟代理正在用于回答更复杂的问题和支持现有客户，它们需要了解问题的上下文，并且至少拥有一些对受挫顾客表达共鸣的能力。智能问答技术更像是一种猜测型（under-the-covers）解决方案。它与你的知识库集成，并为多个渠道的客户提供最相关的信息。例如，当你在搜索栏中输入内容时，会实时弹出最相关的信息——这就是智能问答。

我们比较产品的另一种方式是它们的“对话”能力，即它们以更像人类的方式与人类交互的能力。在我们的框架中，泡沫的大小表明技术通过语音或文本完成这项工作的程度。我们不仅比较对话式AI技术理解人类意图和模仿人类感情的能力，还比较这些技术根据位置、设备和已经搜索过的内容了解问题上下文的程度。这还包括响应的个性化程度。例如，知道你是否喜欢过道座位、脱脂牛奶或爱情喜剧。

为了创造这种情境化和个性化的体验，智能助手技术需要整合到诸如客户关系管理（CRM）、忠诚度系统和内部知识库等更多的后端系统中。当然，虚拟代理需要更长的时间来实施，并且比聊天机器人更加昂贵。这并不意味着谁比谁更好，它们只是解决不同的问题而已。

我们的研究不涵盖语音助理（VA）。如今，相对于企业领域，诸如Amazon Alexa和Google Assistant等语音助理更关注“个人”领域。随着企业开始使用语音助理进行品牌推广、销售和客户支持，这种情况将会迅速发生变化，但我们认为目前还未实现这一点（例如，最近亚马逊宣布将推出企业版Alexa）。此外，语音助手本身也是虚拟代理和聊天机器人部署的渠道，情况可能会因此有一点混乱。

这是一个激动人心且快速发展的话题，目睹AI应用程序如何继续改变客户与品牌之间的关系将会非常有趣。

《Conversational AI is Ready for Prime Time》

https://www.informationweek.com/big-data/ai-machine-learning/conversational-ai-is-ready-for-prime-time/a/d-id/1331196?

3/6/2018
02:00 PM

Stanton Jones, director & principal analyst, Information Services Group

Don't lump the various types of AI-based technologies into one basket. There are differences among them, as a look at call center tech highlights.

We’re seeing some dramatic changes in the call center outsourcing market, driven by big changes in customer expectations and exponential improvements in technology. Until fairly recently, call center executives were focused primarily on cost reduction and productivity improvements. This kind of thinking has changing quickly, for a few reasons: 

�         Customer demographics are changing. Millennials now have significant buying power, and they expect a very different experience when interacting with a brand. They want a seamless, omni-channel experience that can take place across a physical store, a mobile app and a Facebook messenger chat.

�         Switching costs are also very low, given how easy it is to swap one brand for another via a mobile app. Companies are trying to find ways to improve customer stickiness by creating a personalized experience that surfaces the right products and services to customers at the right time and at the right place.

�         Conversational AI is ready. Technologies that can understand human intent and can hold a human-like conversation have improved dramatically over the past 12 months. Contact center executives are telling us that conversational AI is rapidly moving beyond prototypes and into production-ready products. Recent ISG research reinforces this: Adoption of virtual agents and chatbots will more than double over the next two years.

We’re seeing a marked increase in interest from enterprise buyers in chatbots, virtual agents and intelligent Q&A, a basket of technologies we’re collectively calling “intelligent assistance”. At the same time, we’re seeing a lot of confusion on the differences between them. Because technology vendors and service providers use these terms in different ways, it’s hard to determine which vendor or provider can solve the specific problem at hand (which sounds a lot like cloud a few years ago).

While we’re still in the process of interviewing call center executives and researching the various products and vendors in the market, there are a few commonalities emerging, which is enabling us to start to put these vendors into some categories.  The scope of our current research covers chatbots, virtual agents and intelligent Q&A – more about voice assistants in a minute.

We see two main axes that differentiate these three technologies: 1) the number of tasks it can perform, and 2) the number of channels it operates on. Tasks can be simple, like fetching news, or complex like asking a question about an open order. Channels are where conversations take place, like email, a mobile app, a messenger platform, like Facebook or Slack, or even a standalone device, like Amazon Echo, or Google Home.

For the most part, we’re seeing chatbots handle single tasks across a small number of channels. These tasks don’t require a lot of context, personalization or empathy. Chatbots are good at things like corporate branding and product sales (see Burberry’s chatbot on Facebook Messenger, for example) or at doing one thing really well (using x.ai as your personal meeting scheduler, for example).

On the other hand, virtual agents are being used to answer more complex questions and support existing customers, areas that require understanding the context of a question and at least some ability to mimic empathy for a frustrated customer. Intelligent Q&A technology is more of an under-the-covers solution. It integrates with your knowledgebase(s), and surfaces the most relevant information for customers across multiple channels. Think about when you are typing something into a search bar and the most relevant information pops up in real-time – that’s intelligent Q&A.

The other way we are comparing products is by their “conversational” ability. This means their ability to interact with a human in a more human-like way. In our framework, the size of the bubble indicates how well the technology does this, via voice or text. We’re comparing this not just on the conversational AI technology (interpreting intent, mimicking empathy, etc.), but also the degree to which the technology understands the context of what is being asked based on things like location, device and what’s already been searched for. It also includes the degree to which the response is personalized. For example, knowing if you prefer an aisle seat, skim milk or romantic comedies.

To create this kind of contextualized and personalized experience, it means that the intelligent assistance technology needs to be integrated into a lot more backend systems like CRM, loyalty systems, and internal knowledge bases. Of course, virtual agents are going to take longer to implement and are going to be more expensive than chatbots, for example. It does not mean that one is better than the other, they just solve different problems.

Our research purposefully didn’t cover voice assistants (VAs). VAs like Amazon Alexa and Google Assistant are more focused on the “personal” realm than the enterprise realm today. That is going to change quickly as enterprises start to use VAs for branding, selling, and support, but we don’t think they are just there yet (for example, see Amazon’s recent announcement on Alexa for Business). Also, the VAs themselves are channels upon which virtual agents and chatbots will eventually be deployed, so the picture can get a little muddy with them included.

This is an exciting and fast-moving topic, and it will be fascinating to watch how AI applications continue to transform the relationship between customers and brands.

 附件：

《Conversational AI is Ready for Prime Time》--原文.pdf

《Conversational AI is Ready for Prime Time》--译文.pdf