1、TALKDESK RESEARCH REPORTThe promise (and pitfalls)of self-service automation in customer service5 ways to design virtual agents and chatbots to delight your customers.Foreword 03Methodology 04I.Companies recognize the potential of conversational AI to elevate CX at scale.05II.Conversational AI is fr

2、equently falling short of customer needs.09III.Five actions companies can take to realize the potential of conversational AI in their contact centers.16Wrap-up 21Table of contentsForewordIn light of recent economic headwinds,many companies are anticipating slower growth and responding to increased i

3、nvestor focus on sustainable cost structures by reducing their spend.Faced with the possibility of an extended downturn,customer retention has become even more critical.Meanwhile,customer expectations continue to rise.When it comes to customer service and support,artificial intelligence(AI)and autom

4、ation can be a win-win,reducing cost while enhancing the customer experience.As more companies invest in these technologies,they may be realizing the first win of lowering operational expenses,but far too often,its to the detriment of their customer experience.In this report,based on the results of

5、two surveys and three focus groups,we discuss companies adoption of AI and automation in customer service,consumer perceptions of virtual agents and chatbots,and provide actionable advice for deploying conversational AI to reduce cost and enhance customer experience.3MethodologySurvey 1,entitled 202

6、2 Talkdesk Consumer Insights Survey,was fielded in August 2022 to a sample of 1003 U.S.and U.K.-based consumers with at least one customer service interaction in the previous year.Survey 2,entitled 2022 Talkdesk CX Professional Survey,was fielded in July 2022 to a sample of 301 U.S.and U.K.-based pr

7、ofessionals.CX professionals included customer experience,customer service,CX operations,and IT and were employed by organizations across a range of industries with at least 100 employees.Insights from the two surveys were augmented by three focus groups with 29 U.S.consumers to understand what spec

8、ifically is frustrating about experiences with virtual agents and chatbots.All 29 responded to a screener survey in which they attested to having had an interaction with an AI or automated system in a customer service scenario within the previous seven days.Some of the following quotations from thes

9、e focus groups have been lightly edited for clarity.4I.Companies recognize the potential of conversational AI to elevate CX at scale.After a two year economic boom when many companies were in hyper-growth mode and customer acquisition was the priority,in the last several months,leaders have had to v

10、ery quickly tighten the purse strings and once again focus on customer retention.In fact,in the 2022 Talkdesk CX Professional Survey,54%of respondents indicated that their top desired outcome from customer service software investments was to increase customer retention.This was higher than any other

11、 category including revenue growth(40%)and operational expense reduction(30%).Top desired outcome from customer service software investmentsIncrease customer retentionNew customer acquisitionGrow revenue through the contact centerImprove agent capabilitiesIncrease first contact resolutionCreate digi

12、tal experiences at scaleReduce operational expenses54%33%30%30%39%40%48%5Companies are feeling the pressure to do more with less.With increased market competition and pressure on pricing,companies are leaning into their customer experience as a key differentiator for retention and loyalty.However,co

13、mpanies are also feeling the pressure to do more with less.When asked to indicate the barriers to improving CX,over 60%of CX professionals selected high costs or expenses for CX operations and 57%selected insufficient frontline staff capacity.Top barriers to improving CXHigh costs or expenses for CX

14、 operationsInsufficient frontline staff capacityInadequate frontline staff skillsPoor integration between customer-facing systems or touchpointsCurrent CX software restricts ability to make improvementsPoor integration between back-end platforms or processes that support CXLimited data collection an

15、d aggregation across CX platformsLack of alignment or support from leadership61%50%49%48%43%53%57%57%6How do leaders improve customer retention by way of enhanced customer experiences when there is limited(or no)budget to do so,whether as a result of restricted budget for systems/processes or insuff

16、icient frontline staff?The answer may lie with AI and automation.Theres no shortage of back-end use cases for AI in contact centers,from intelligent forecasting to predictive routing,and everything in-between.But its the front-end,external-facing AI and automation that may prove to be a major win-wi

17、n for companies by helping to reduce costs(e.g.,offloading of routine inquiries)while providing better experiences for customers(e.g.,enhanced self-service that expedites resolutions).Most of us have interacted with chatbots in the past,and are therefore familiar with the term.Technically speaking,t

18、hey are rules-based and do not utilize AI.As rules-based systems,they use a set of predefined rules or decision trees to determine how to respond to user input.These rules are based on keywords or patterns in the users message and the chatbots response is determined by the specific rule that is trig

19、gered.A virtual agent is an application that uses AI to simulate conversations with humans in customer service scenarios.These solutions fall under a broader branch of artificial intelligence called conversational AI that focuses on the ability to understand,interpret,and produce natural,human-quali

20、ty language.While virtual agents can serve a similar purpose as chatbotssuch as improve customer service efficiency,reduce wait times,provide 24/7 availabilitythey arent limited by a set of rules,and thus typically can handle a broader range of customer inquiries.Also,while a virtual agent can be bu

21、ilt into a web chat interface,they can also be used in other channels like voice and business messaging.7Many companies are currently using conversational AI or plan to soon.75%of CX professionals polled said they are investing in both voice-based AI virtual agents and natural language-based chatbot

22、s or already have these in place.However,the effectiveness of these investments is contingent on how well they are implementedand evidence from consumers suggests they are often not working as intended.Customer engagement capabilities in place and planned for investment in 6 monthsNatural language-b

23、ased chatbotSpeech and text analytics(e.g.intent prediction,sentiment analysis)Intelligent knowledge baseVoice-based AI virtual agentInvestingHave in place40%32%45%Simple(keyword-based)chatbot33%37%23%29%35%35%40%8II.Conversational AI is frequently falling short of customer needsDespite ongoing inve

24、stment in conversational AI,this technology is currently handling a relatively small share of consumer interactions.Less than 30%of consumers report that their most recent customer service episode involved interacting with a voice-based AI or a chatbot.Fewer still report that they would have preferr

25、ed to have interacted with either,with most consumers satisfied to interact via live chat or over the phone.Live phoneChatbotVoice-based AILive chatMost recent channel usedChannel preferred31%26%49%47%18%9%11%5%9Nonetheless,64%of consumers report being open to interacting with an AI system if it mea

26、ns solving their issue faster.Moreover,over half of consumers say that they wished more companies would enable them to solve their problems on their own.Together,these points indicate that if AI could enable consumers to quickly and fully resolve their needs on their own,most would be open to using

27、it.If AI could enable consumers to quickly and fully resolve their needs on their own,most would be open to using it.Percentage of consumers that agree with the following statementsI am open to interacting with an AI system if it means I can resolve an issue faster.64%I am comfortable interacting wi

28、th an AI system to resolve a customer service issue.48%Id like more companies to enable me to solve my problems on my own withtout needing to call customer service.52%An AI system can resolve a customer service issue just as well as a person in most cases.30%I trust an AI system more than interactin

29、g with a person to make sure my personal information is secure.27%10Unfortunately,automated systems today arent nearly as effective at resolving issues on first contact.In fact,depending on what theyre trying to accomplish(e.g.,seeking a refund or trying to learn about a product),human agents are be

30、tween 1.4 and 2.7 times more likely to resolve an issue on first contact than an automated system.It almost seems as though companies are saying we want to use AI,consumers are saying bring it on,yet AI systems are falling short of consumer expectations.Companies are saying we want to use AI,consume

31、rs are saying bring it on,yet AI systems are falling short of consumer expectations.First contact resolution by service inquiry typeTrying to learn more or receive a recommendation about a product or service already ownedTrying to learn more or receive a recommendation about a product or service und

32、er considerationSeeking information or an update about the delivery of an orderSeeking a refundRestoring a mismatch between a product or service that was ordered and receivedTrying to return or exchange an orderReporting an issue with a company employeeAUTOMATED SYSTEMSSERVICE AGENT51%48%42%32%32%27

33、%23%79%76%75%60%58%68%59%111.When chatbots dont understand the reason for a customers outreach.The majority of our focus group participants indicated that they felt like companies dont have an accurate understanding of their common pain points.Consumers are very aware of how much companies invest to

34、 understand the wants and needs of their customers,but are disappointed with how little is done with that information to their benefit during customer service interactions.When a customer contacts the customer service department,they likely need something very specific.They expect that the company(w

35、hether being represented by a chatbot or a human)is able to understand this need and respond accordingly.When that doesnt happen,the frustration that customers feel isnt just temporaryit can actually leave them with a long-term impression that the company doesnt value them enough to get these experi

36、ences right.From the perspective of one of our focus group participants:Consumer A:“Its almost like the bot sending me a link to a forum which didnt resolve my issue.that to me makes it seem like the company is just totally asleep at the wheel.”To develop a deeper understanding of how and why,we con

37、ducted three virtual focus groups with 29 consumers.Our conversations focused on three common causes of frustration when it came to their experiences with virtual agents and chatbots:When bots dont understand the reason for a customers outreach.When theres an inability to escalate from a bot to huma

38、n support.When bots cant access or pass along important information.Lets examine each of the three common causes of frustration from the perspective of our focus group participants.Note:All names within the following quotations have been anonymized.Focus group participants typically used the term ch

39、atbot or bot,rather than virtual agent,but were not intentionally distinguishing between the two.12312Another participant described having a love-hate relationship with bots.He loves them when theyre able to save time for him,but in his experience,theyve only worked well in basic scenarios.He provid

40、ed an example of a more complex question that a bot wasnt able to help him with:Consumer B:“But then I have some extreme experiences where I hate chatting with a bot.One example is shopping online for web hosting services.My question was about whether this company distributes my refund,meaning if I

41、use three-fourths of it do I get the partial back?And the chatbot couldnt seem to answer it.Instead,it was trying to refer me to some help guide online that didnt even answer my question.Then it tried sending me several more links and after reading the articles it still didnt help me.So I realized I

42、 just wasted 10 minutes of my life.”This participants example underscores a key takeaway.Consumers dont expect chatbots and virtual agents to be able to resolve every issue they may encounter.What consumers do expect is that chatbots will understand why they are reaching out,and provide accurate sol

43、utions for any inquiry that they attempt to solve directly,especially after repeated attempts.This was often the underlying reason why the overwhelming majority expressed a preference to speak to a human from the get go.In the words of another focus group participant:Consumer C:“I would prefer to sp

44、eak to a human so I dont have to repeat myself three or four times.I want to get things wrapped up and get off the phone instead of being bounced around for 10-15 minutes with a chatbot,trying to get them to understand what Im saying.”132.When theres an inability to escalate from a bot to human supp

45、ort.If chatbots cant proactively recognize their own limitations,consumers at a minimum expect an easy way to make that determination for themselves.Unfortunately,many of our focus group participants lamented about experiences where they felt trapped in a cyclical loop with a chatbot.One participant

46、 described her experience trying to refill a prescription:Consumer D:“The bot just offered me the same answers and I was trying to find the button to get to a human.It was awful.”Consumer D:“Finally,30 minutes after I called,I talked to a human agent and yes they helped me.But it was a nightmare bec

47、ause the 30 mins I was on the phone,I really needed to do something else.The bot kept asking:do you need to refill your prescription?But I just needed to ask a question about a different prescription.”Talkdesk Interviewer:“Did you get your request resolved?”Talkdesk Interviewer:“So the automated sys

48、tem didnt give you an option to talk to the pharmacist?”While many consumers appreciate having a choice to self-servethey often feel strongly that it should not be their only choice.In fact,the majority of focus group participants agreed that certain types of inquiries should always be handled by a

49、human,particularly those requiring empathy,sensitivity,or a personal touch:In these scenarios,few things are more frustrating to consumers than a protracted struggle to reach a human when that was their preference from the start.Consumer E:“When youre speaking with a human,you have that emotion and

50、that compassion.They can pick up on your mood if youre having issues with something.With an automated system?No.”Consumer D:“No.It gave me other options,but not to talk to the pharmacist.I got lost in the options and had to hang up and call back.I pressed the second option and that option got me to

51、another option.Eventually I was able to find the right path to talk to the pharmacist.”143.When bots cant access or pass along the right information.The third common cause of frustration for our focus group participants involved experiences with chatbots that either couldnt access basic information

52、about customers,couldnt pass along information that was already shared when escalating to a human agent,or both.As customer service technology continues to advance,a growing number of consumers view these capabilities as table stakes,and failing to meet their expectations can lead to the worst case

53、scenario of customer churn.In the words of one impassioned participant attempting to activate a new eSIM for her phone:Consumer F:“The chatbot kept asking me really stupid questions.It told me it didnt have my phone number.Ive been with this phone company for five yearsif they dont have my phone num

54、ber,I dont know what to say.I finally felt like I was going to have to take a sedative and I never did get an answer.So I decided to get a new phone company that had 5G.They made it easy to sign up online in a few minutes.”Consumer G:“If you have to go through all of the automated steps required to

55、finally reach the right person,please keep a record of everything that was said previously,so that you can go into that conversation and dont have to repeat,sometimes for the third time,what youve already told the chatbot.”The same sentiment applies when chatbots and virtual agents do escalate an in

56、teraction to a human agent,but fail to provide a record of what has already transpired:Consumers today expect that whoever they speak to,whether its a chatbot or a human,has access to the relevant information required to resolve their issue,and that when they are transferred,the history of their int

57、eraction will be passed along seamlessly.In each of the focus groups,the frustration that participants carried with them was palpable,even months after a negative experience transpired.With the advances in conversational AI over the last several years,companies no longer have an excuse for deliverin

58、g sub-par automated experiences,and a growing number of consumers are willing to explore their options if their expectations arent met.So what can companies do to ensure their AI investments are having a net positive impact on their customer experience?15Building a virtual agent can be challenging,b

59、ut it doesnt have to be.Regardless of whether youre new to conversational AI or a seasoned pro,the following steps will guide you in creating virtual agents that will reduce costs and enhance your customer experience.III.Five actions companies can take to realize the potential of conversational AI i

60、n their contact centers.161.Apply principles of good conversational design when implementing new virtual agents.Conversational design is important for creating a virtual agent that is engaging,helpful,and easy to use.The first fundamental,yet critical,step is to understand the most common reasons fo

61、r customers to contact your customer service department.One way to do this is to pair your virtual agent with an interaction analytics solution that can help analyze the underlying intents across all of your interactions and determine what the most critical points of friction are across your custome

62、r journey.Once confident in your understanding of the customer journey,its time to think about designing your virtual agent.Here are several best practices when it comes to good conversational design for your virtual agent:Use simple,straightforward language that is easy for users to understand.Avoi

63、d using jargon or technical terms unless they are necessary.Make the virtual agents tone and personality appropriate for the intended audience and the type of conversation.Use appropriate conversational cues,such as greetings,goodbyes,and prompts to help guide the conversation and make it feel more

64、natural.Make the virtual agents responses relevant and helpful to the users needs.This may involve using machine learning to personalize responses based on the users past interactions or other data.Design the conversation flow so that it is logical and easy for users to follow.This may involve using

65、 menus or other navigation elements to help users find the information or assistance they need.Test the virtual agent with a small group of users to get feedback and make improvements before launching it more broadly.This can help you identify any issues or areas for improvement and make sure the vi

66、rtual agent is delivering a good user experience.For more information,see our best practices guide on designing conversations.172.Implement a fallback mechanism in your virtual agent.A fallback mechanism is critical for handling novel input that your virtual agent doesnt understand.There are several

67、 ways to implement a fallback mechanism in a virtual agent to make sure it can handle unexpected questions:Use a default or catch-all intent.Create a default or catch-all intent that can handle any input that does not match any of the other intents.This can be used to provide a generic response such

68、 as“Im sorry,Im not sure I understand.Can you please rephrase your question?”or redirect the user to a human agent.Use a keyword-based fallback mechanism.Implement a keyword-based fallback mechanism that checks the input for specific keywords or patterns.If the input contains certain keywords,the vi

69、rtual agent can provide a specific response or redirect the user to a human agent.Use a confidence threshold.Some conversational AI platforms allow you to set a confidence threshold.If the confidence score of the intent is below the threshold,it can be considered a fallback scenario and handled acco

70、rdingly.Use a combination of the above.Use a combination of the above methods to create a robust fallback mechanism that can handle a wide range of unexpected inputs.Its important to note that fallback mechanisms should be designed to be as simple and user-friendly as possible.They should provide a

71、clear and concise response and guide the user in the right direction.Additionally,its also important to continuously test and monitor the fallback mechanism to ensure its working as intended and make any necessary adjustments.183.Provide a way for customers to opt-out of the conversation and be conn

72、ected to a human agent.More and more companies are realizing the importance of providing a way for users to get to a human agent,as it can improve the customer experience and increase customer satisfaction.In many cases,they are opting for hybrid models where a chatbot and/or virtual agent is used t

73、o handle simple and repetitive tasks and a human agent is used for complex or sensitive tasks.When integrating several channels into your customer experience strategy,it is best to use a CX platform built for omnichannel engagement that can seamlessly escalate between self-service and human-assisted

74、 interaction channels.This will not only remove a key point of friction for your customers,it will also provide your human agents with the context they need to resolve inquiries that couldnt be contained through self-service more quickly and effectively.Finally,if it is cost-prohibitive for you to a

75、llow customers to reach a human 24/7,consider a CX platform that supports callback capabilities.This way,when customers want to speak with a human but a human isnt available,they can schedule a callback for later.4.Continually improve your virtual agent through testing and training.Usability testing

76、 and user acceptance testing(UAT)are both methods of evaluating a product or system to ensure that it meets user needs.Usability testing is focused on identifying and addressing issues that make the product difficult to use,while user acceptance testing is focused on verifying that the product meets

77、 the required acceptance criteria and is ready for release.Doing usability testing or UAT once is good.Doing it often is better.Its clich,but truethe only thing certain in this world is change.Your product/service changes,customer expectations change,technology changes,and the list goes on.Your virt

78、ual agent could also use a tune-up every now and then.Incorporate regular(at least bi-annual)usability testing into your virtual agent maintenance plan.Finally,its important to continuously train and improve the machine learning models that are used to identify the intent of a users input and match

79、them with the right responses.This ongoing process helps ensure that your virtual agent is always equipped to handle changes to your business,customer inquiry trends,and the language or vernacular your customers use over time.Typically,this type of work requires complex,manual processes and the expe

80、rtise of data scientists or artificial intelligence experts.Fortunately,there are now human-in-the-loop tools available that streamline the process of training machine learning models without the need for specialized talent.These tools allow non-technical staff to easily apply their practical and bu

81、siness knowledge to continuously improve the effectiveness of their virtual agents.195.Reimagine virtual agents with GPT and large language models.Generative Pre-trained Transformers(GPT)is the large language model behind ChatGPT,which took the world by storm in late 2022 by demonstrating the immens

82、e power of generative AI.While broadly useful across a variety of industries and use cases,large language models like GPT are poised to quickly revolutionize the customer service industry.With little to no training,GPT can provide intent detection,entity extraction,sentiment analysis,redaction,summa

83、rization,and classification all of the core natural language processing(NLP)functions that are critical to customer service use cases.It will soon enable virtual agents to have increasingly complex,human-sounding conversations with customers,while also making them less expensive to implement,maintai

84、n,and improve over time.Although the vast potential of this technology is just beginning to be explored,the positive impact it will have on self-service automation in the contact center simply cant be overstated,and shouldnt be ignored.CX leaders who are using or planning to use virtual agents shoul

85、d ensure that their technology partners are incorporating GPT and other large language models into their solutions as soon as possible,and can provide a clear vision for how their customers can benefit from the next generation of AI in the contact center.20Wrap-upFollow these five steps to guide you

86、 through the process of building your first virtual agent or improving an existing one.Despite the myriad of poorly designed virtual agents out there now,I believe in a world where technology always improves human lives.Cheers to a bright,bot-friendly future!2121+1(888)743-3044 Talkdesk is a global

87、cloud contact center leader for customer-obsessed companies.Our automation-first customer experience solutions optimize our customers most critical customer service processes.Our speed of innovation,vertical expertise,and global footprint reflect our commitment to ensuring that businesses can delive

88、r better experiences across any industry and through any channel,resulting in higher customer satisfaction and accelerated business outcomes.2023 Talkdesk,Inc.All Rights Reserved Worldwide.Unauthorized use,duplication,or modification of this document in whole or in part without the written consent of Talkdesk,Inc.is strictly prohibited.Automation-first customer experience solutions