Conversing without these AI policies spells trouble
Before you jump in the deep end of the pool that is known as conversational AI, take the opportunity to consider the role of digital policies in keeping your enterprise protected while also reaping the rewards that the new channel provides.
A little more than a year ago, I shared how we used a chatbot to support marketing governance ( see Use a Chatbot for Digital Governance? Why Not?). This year, I found myself in a digital policy project focused on the rollout of enterprise chatbot capability. The organization is introducing the chatbot functionality as part of a broader digital transformation supporting prospects and customers, but the staff that interacts most often with the — customer service.
If bare decision trees and a chat with multiple links as choices are the concepts that just popped into your brain, you are sorely mistaken in what chatbot capability looks like in today’s business world. Enterprise chatbots are more sophisticated, more nuanced, and present more opportunity (but also risk!) than ever before. And the numbers the narrative:
- 85% of customer interaction will be handled without human agents by 2021.
- 50% of businesses plan to spend more on chatbots than on mobile apps.
- 64% of internet users say 24-hour service is the best feature of chatbots.
- 37% of people use a customer service bot to get a quick answer in an emergency.
Oh, and then there is this: chatbots can cut operational costs by up to 30%! That’s because, according to Gartner, by next year, 85% of customer interactions will be managed without humans, with customers getting answers instantly, and agents can spend their time solving more critical issues.
In the past two years alone, there has been an abrupt change in chatbot learning, and artificial intelligence (AI) has transformed not just capabilities but also how users interact with chatbots. The digital power has matured to where chatbots appear human and deal with more human conversations thanks to machine learning that sits behind the chatbot. Advances allow brands and services to create an interface that feels human and interacts in a way that people expect.
But before you jump in the deep end of the pool that is known as conversational AI, take the opportunity to consider the role of digital policies in keeping your enterprise protected while also reaping the rewards that the new channel provides. Here are a few of the lessons learned with my clients this year, including the policies that helped get operational alignment where it matters most.
- Usage analytics and reporting. The benefit of conversational AI systems comes from the enterprise improving service and freeing humans to focus on higher-value tasks, answering more complicated queries, and improving service to customers. We had to baseline analytics at the start of the initiative, pulse checks along the way, and report the resulting analytics in an aggregated manner. Headquarters took the lead on this effort, but we faced resistance in capturing and reporting analytics at the local market level due to already overburdened resources. We created a policy compliance scorecard, which, if completed successfully, translates into a 5% budget boost for local markets to use in marketing or sales efforts. The scorecard includes insights about customer activities, challenges, preferences, and frequency of interaction. We now get transparency into the effectiveness of the conversational AI capabilities and can detect a problem in a specific area of operations. The client has been able to update their website information and upgrade internal processes.
- Data ownership and privacy. It is great to have insight and visibility into a user’s behavior and use associated reporting to benefit. The other side of the coin is that the user data is collected, stored, and managed according to data policy requirements. Our policy focus was on the ownership of the user data inside of the enterprise and who has the authority to update, manage, and use it for marketing and sales purposes. Along with data ownership, the policy addressed responsibilities for the user data, including updating and deleting data when requested by a user. This might seem very straightforward, but toss in issues like data localization and the inability to centralize all data in a single repository, under harmonized rules, made this an area of particular interest and focus for the company. The biggest lesson learned is that you need to have this conversation well ahead of deploying conversational AI. You must understand your machine learning algorithms, including access to user data.
- User handoffs. Sometimes, your conversational AI bit doesn’t stack up or needs a friendly bot or human to intervene. When the existing tool cannot deliver, you need to have a plan to handoff. That might mean having a customer service bot handoff to a sales bot to get insights into quantity discounts, or creating a handoff between a recruitment bot and a human HR specialist. Our project’s policy focused on the handoffs: at what point is a handoff warranted, how many times do we want the user to attempt to ask the question differently, and what type of tone should be adopted by the bot depending on the user sentiment.
- Language, terminology, spelling, and bot personality. This is the most basic sounding of all policies I have worked on in my career, and yet it was by far the most complex one of the client engagements. A bot needs to have a personality, just like a human, and humans are very complicated. There was a lot of chatbot language customization for markets like Belgium, where users not only speak Dutch, French, and German proper but many dialects of the languages. Add slang to the mix (who knew klick had so many regional connotations?), variations of spelling (it’s more than tomato vs. tomatoe), and bot personality (how many emojis is too many?), and you can see there is nothing simple about the policies that can guide the conversational bot development.
- Accessibility, and inclusivity. For users without a hearing or sight impairment, finding a chatbot on their screen is easy. For the disabled, navigational issues can become debilitating when using a chatbot. To make the chatbot accessible and inclusive, the policy focused on making the conversation identifiable (every message in the chat has to be marked “from the user” and “from the bot” in code), avoiding jargon and acronyms to prevent user confusion when voice readers come across them, and simplifying replies to make understanding the chatbot easier for those with cognitive disabilities.
My list is a good representation of the critical policy areas around conversational AI. Not surprisingly, the more capabilities and experience the enterprise develops, the more policies can be expected. I am excited to continue supporting this initiative, as it has many implications for cost savings and increasing individual worker productivity.
Are you thinking about using conversational AI inside your enterprise for either marketing, sales, or in-house productivity? Drop a line and let me know what challenges and lessons learned you have to share. I’d love to hear what policies you are developing to help keep your risks and opportunities appropriately balanced.
Originally published at https://www.kpodnar.com.