There’s a reason why people say “the first impression is the last impression.” Some 51% of customers never approach a business again after one bad experience. That puts pressure on every interaction—and every missed opportunity—with potential customers, recent purchasers, and long-time users.
Web chat is often the first impression for customer service interactions. While chat services initially connected consumers with real customer service staff, chatbots have become increasingly common—for obvious reasons and with obvious limitations.
Ultimately, no platform guarantees an improved customer experience. Consumers care most about solving their problems. Whether they do it with a person or a chatbot is secondary.
This post covers:
- What chat platforms can (and can’t) do.
- The questions you should ask to find out which platform is right for you.
Why you should care about chat
It’s what your consumers want. Preference for chat continues to grow: Some 70% of consumers prefer chat to voice if given the choice, and nearly two-thirds of customers who engage with a chat platform are more likely to return to a website and purchase again.
It can save you money. Collectively, companies spend more than $1 trillion annually fielding 265 billion calls. As IBM notes, “that’s a lot of money dedicated to sub-par experiences.” Further, call centers are notorious bastions of turnover, increasing employer costs. Estimates peg turnover at 30–45%, with each occurrence costing a company as much as $6,440.
Just as businesses have been happy to move away from costly call centers, they’ve been equally intrigued to experiment with chatbots. Not all experiments have been successful.
When chatbots work—and when they don’t
Several years ago, Gartner predicted that, by 2020, businesses would manage 85% of customer interactions without involving a person. AI-powered chatbots were central to that prediction.
A more recent prediction from Gartner hedged on humanless customer service but nonetheless anticipates that, in just over a year, 25% of all customer service operations will be virtual.
Currently, chatbots rely on one of two technologies:
- Command-based chatbots that answer a limited set of questions.
- AI-based chatbots that answer more ambiguous queries.
Both types of chatbots perform best for “Tier 1” questions, or those that are easily interpreted and whose answers can be extracted efficiently from a database (e.g. “What time does my flight leave tomorrow?”).
Both still struggle to answer complex questions and, in some cases, incredibly simple questions:
The relative power of AI-based chatbots depends, in large part, on their access to training data. Companies such as Acobot crawl client websites to generate a seed list of terms and knowledge in a matter of minutes.
As you can imagine, the richness of that source material—and the subsequent quality of chatbot responses—varies greatly.
Live chat vs. chatbots: 6 questions you should ask
More than anything else, Facebook catalyzed the rapid adoption of chatbots when it opened its messenger app to private developers in 2016. For companies, it was a new channel to reach their 1.5 billion consumers.
The rush to market generated spectacular successes and failures. These first movers, as well as more recent entrants, exemplify the situations when you should—or shouldn’t—deploy chatbots.
1. Are you struggling with customer service response times?
- Chatbots respond instantly, 24 hours a day, 7 days a week.
A human takes more than 15 seconds to respond to a query. A single 15-second delay may not impact the customer experience significantly, but the aggregate delay—15 seconds for each response in each interaction across every representative—can add up to longer wait times and higher staffing requirements.
Likewise, if consumers expect answers 24/7, chatbots are the most cost-effective (and, in some cases, only plausible) solution.
IBM cites Autodesk Virtual Agent (AVA) as a success story for chatbot response times: The company improved customer response time by 99% while also escalating complex questions to human agents (a system sometimes called “Hybrid AI” or “Agent-Assist”). The improved response time cut average resolution times down from 1.5 days to just 5 minutes.
AVA resolved many of its 1 million annual support requests without human support, including simple issues such as changes of address, activation codes, and contract questions. AVA estimates that the chatbot integration reduced total per-case costs from $15–200 to $1.
Even if user questions are ultimately escalated to a human representative, a chatbot’s instant response may keep someone from leaving your site. (Alternatively, an inane reply may chase users away—speed isn’t the only KPI for customer service.)
2. Do you waste time answering simple, fact-based questions?
- Chatbots can fetch basic information from company databases almost instantly.
Satisfi Labs provides AI-based chatbots to zoos, aquariums, museums, and similar organizations. Their chatbots collect visitor data while also helping consumers book tickets.
Chatbots solicit basic information with questions like “Which day would you prefer to attend the event?” or “How many people will be attending the event with you?”
This is one reason that the airline industry has been one of the quickest to adopt chatbots—many of their customer service interactions focus on simple questions about flight times, delays, or ticketing information whose answers are easily extracted from a database.
3. Do you want more data about customer service interactions?
- Chatbots can diagnose online trends and report back to your organization.
Pairing conversational data from chatbots with visitor location on a website helps organizations customize their experiences and improve conversion rates.
The same is not true in the case of live agents. Unless live chat agents have access to powerful integrations like Microsoft Dynamics, they may not be able to recall the customer history efficiently or understand past pain points. (See the above example from KLM.)
4. Are you trying to reduce customer service costs?
- Chatbots can reduce costs in high-volume customer service centers.
- Chatbots still require a backup, human-based option.
The time and money it takes to roll out a chatbot vary greatly, from a few minutes and a few dollars (with a service like ManyChat, which has tiered, audience-based pricing starting at $15 per month) to hundreds of hours of development costs to build a conversational bot from scratch.
How do those costs compare to live agents? A survey of hourly rates on Glassdoor estimates the hourly rate of a live chat agent at $7–15, and Upwork suggests that customer support representatives with field specialization often charge 20–50% more than other reps.
As one live chat agent manages only 4–6 chats at a time, a high-volume live chat service can quickly become a cost center, especially when layering training, management, and turnover costs on top of hourly expenses.
However, chatbots add another layer of technical risk to online experiences—a website, a chat platform, and its chatbots must remain operational. For most companies, that makes some kind of backup customer service option essential. (IBM predicts that phone-based customer service will exist solely as a backup option by 2020.)
5. Do you need to scale customer service operations?
- Chatbots can extend personalized customer service that would be prohibitively expensive to implement with real people.
Chatbots often disappoint when they’re intended to replace humans for core customer service needs. However, companies have found more opportunity for experimentation (and a higher user tolerance) when chatbots deliver a level of customer service beyond what consumers expect.
Examples of how companies have scaled customer service with chatbots:
- Adidas Studio LDN. Adidas created a female-focused workout studio to boost brand engagement. It uses chatbots on Facebook Messenger to create a personalized, interactive booking process. An effort to deploy that same level of personalization with human agents would’ve required the company to invest in “thousands of field team members,” well beyond its current staff of 30.
- The Edit. A vinyl record store uses text-to-buy chatbots to deliver personalized daily recommendations to consumers. The chatbot records responses and tailors recommendations based on previous answers. If consumers respond with a more detailed question, the request forwards to a human representative. The Edit has sold 50,000 records and grossed more than $1 million with the system.
- Kia. Kia rolled out a Facebook Messenger chatbot (“Kian”) in November 2017 to consolidate marketing information across 800 websites. The conversion rate via the messenger bot was 21% compared to just 7% on the main corporate website. Importantly, the messenger bot captures explicit, granular details of consumer needs—data quality vastly above that from website click paths or heatmaps. Kia uses those insights to improve retargeting.
- Emirates Vacations. Emirates Vacations deploys chatbot messaging within display ads. The company tailors display ads and chatbot interactions based on the context of the page on which the ad appears. The strategy allows consumers to engage with the brand without leaving their current page. Engagement rates rose 87% compared to its standard display ads.
- Madison Reed. A text-based chatbot helps guide beauty product consumers through a series of questions about their hair, such as “Do you ever use heat on your hair?” or “Is it chemically treated?” The service boosted brand engagement by 400%, though only 21% of users clicked through to the website, suggesting that the novelty of testing a chatbot’s capabilities may outpace its bottom-line impact.
6. Are your customer service conversations complex?
- Chatbots are nowhere close to passing the Turing test.
Users tend to prefer live chat because chatbots often fail to understand even basic requests:
Beyond such abject failures, the subtleties of human-to-human interactions are essential for on-brand communication.
- Everlane. The fashion retailer shuttered its chatbot, in part, because the unpredictable quality of chatbot interactions risked diluting the brand voice. Rather than trying to extend a pseudo-personalized reach, it refocused on email—a medium in which it controlled all messaging and could better meet consumer expectations of tailored but not hyper-personalized communication.
For complex sales funnels, human agents also excel at helping consumers navigate the decision process, addressing doubts and answering specific questions.
- Intercom. When CEO Eoghan McCabe announced a new chatbot earlier this month, he argued that the key to improving chatbots was to move away from the “text-based phone tree” model (which inevitably ends with a recommended email address or phone number—i.e. no solution) and instead to focus on answering Tier 1 questions while elevating the remainder to trained sales staff. Intercom’s Answer Bot mines past interaction data (all human-to-human responses) to build AI-based responses.
Chatbots face other well-known conversation limitations, especially when it comes to human emotion. For example, sarcasm is often confused with sincere communication.
Even for a successful chatbot, advances may dig a deeper hole: Facebook’s now-shuttered personal assistant M drew more difficult questions from users each time it had success. Everyone, sooner or later, was disappointed.
Live or automated, chat is growing. Much of the debate between live chat and chatbots boils down to a single type of customer experience, as told by Asonele Kotu to Marketing Technology Insights:
The customer might start out thinking they’re chatting to a person, but after a while, the truth is going to dawn. Then they’re really going to regret the five minutes spent describing their recent holiday to Jamaica.
Part of a strong customer success strategy is access. Customers want to reach your company without the frustrating journey to get to you. And by “reach your company” we mean they want to talk to a real person who breathes and everything.
There is a central truth here: Chatbots cannot fake human-based customer service (yet). If your chatbot strategy hinges on sustaining that deceit, you’ll fail. However, a chatbot doesn’t need to simulate a human to be successful.
The companies that have succeeded with chatbots recognize how to deploy them now (for Tier 1 questions), when to shift to live chat (for complex issues), and how to start planning for the future of customer service that, as of now, remains uncertain.