Getting quick answers to common questions or needs, from back-office tech or systems support to consumer-facing customer service applications, sounds great. Recently companies have turned to conversational Artificial Intelligence (AI) technology for just this goal. They have hoped that the technology will reduce demands on human intervention, save time and money, move a process forward efficiently and quickly, and deliver better satisfaction to the end-user.
Despite lofty promises about ease of deployment for business services, ease of use for end-users, and ROI for the organization, too often voice and chat interface implementations fall short of the mark. The result can be failure to perform intended functions or worse — a negative brand experience for users. Add to this a proliferation of disparate conversational AI platforms, systems and applications within an organization, and you end up with a mess for both IT and the business side of the house to sort out. Reusability? Interconnectivity? Forget about it. You are left with bot bedlam.
How to pull your company out of a failed initial foray into conversational AI? There are a multitude of issues that hinder the success of bot implementations. Fixing them starts with identifying the problems that need to be solved. It might be that your bots — AI-based chat and voice automations — are languishing not being used. Perhaps they are delivering stale or incorrect data, causing users to think of them as no longer useful. Maybe they are failing at conversational language, leaving users to think they couldn’t get anything done in the first place. Many of these issues may lead to a bad customer service experience. Happily, they can be fixed.
Getting it right requires orchestration and a mixed set of skills and knowledge.
Making a bot effective requires it to be built around a well-defined (and successful) business process. A good start is to identify which parts of interactions are best suited to automation and which require human intervention. Next, it’s helpful to map the attendant process flow. Just as any business processes which evolve over time to meet emerging demands change, both human and technical resources must keep pace. There must be a set of supporting processes in place to surface need, disseminate changes, re-train bots, and refresh datasets. Bottom line: if your business processes and human agents are delivering an inconsistent experience due to lack of understanding, knowledge, training or supervision, then there is little hope your bot can do any better.
Back-end system readiness will facilitate successful bot implementation. Technical management of bots can be a challenge, particularly when they have been deployed on an ad hoc basis by various lines of business. Likely, there are multiple platforms, systems, APIs and UIs in operation — some duplicative in task or function. Ideally, you can identify opportunities to share or reuse components to streamline automation applications across your organization. Job one is to get the big picture, examining where you have bots in place, their intended functions, how they are (or are not) performing and when they are triggered to interact with users. Evaluate their conformance to your organization’s technical infrastructure, standards and guidelines. Explore what data pipelines are in place to monitor usage, training data to improve customer experience. Consider how your bots are maintained and what technical support is needed.
Your goal is conversational content readiness. Bots should use language the way people actually speak, not how they read or write. While your databases may include content relevant to visual information delivery, it may not be ready as conversational format. This is critically important. Your traditional branding guidelines can also sabotage success. Perhaps you historically insist on full product names; these may be cumbersome in casual conversation. Your bot must understand what is being asked; it needs varied ways to ask a common question. It must understand synonyms, slang, nicknames, and industry jargon, as well as handle accents, common mispronunciations or misspellings. A well-trained and well-designed bot should detect consumer frustration and “sense” when to pull in a live agent. Finally, your bot’s answers should be succinct, on point and actionable.
A well-trained and well-designed bot should detect consumer frustration and “sense” when to pull in a live agent. |
How will your bot learn over time? Having metrics in place to track and assess performance is essential to long-term implementation success. You need systems in place to collect user feedback to better understand and improve the customer experience. You need easily used tools to teach the bot to understand more user language. Your business users can be great coaches. Listening to calls, they can provide insight into new questions and best answers. As subject matter experts, they can own jargon and the knowledge base of commonly encountered questions or requests, along with up-to-date, approved and effective responses. Their engagement is the fastest path to add new capabilities to your automated system. Bot developers can help too. Let them test and deploy new versions in a way that’s integrated with your IT processes.
The latest thinking from our technology and subject matter professionals. See more
If you’d like to discuss fixing the bot bedlam in your organization or simply how to design a sound conversational AI strategy from the jump, talk to us. At KPMG, our talented and experienced AI technology professionals are armed with advanced tools. But just as important, we have professionals with business and sector experience who can help you understand the right path for your conversational AI initiatives. We’ll assemble a team of IT pros, data engineers, data scientists, linguists, and business process specialists to make your conversational AI strategy deliver on the promise of the technology.
Arthur Franke
Data Scientist, KPMG Tech Enablement
arthurfranke@kpmg.com
Na Li
Data Scientist, KPMG Tech Enablement
nali7@kpmg.com