How to set up a chatbot?

Do you have a Chatbot project? Here is the guide from A to Z to create your first Chatbot!

For some time, chatbots have been at the heart of new web marketing strategies (especially with the use of Facebook Messenger).

Related topic : LG's new smartphone is not flexible, but has two screens

But now they allow us to go further in terms of automation, especially in corporate customer services.

Plan de l'article

What are the different types of Chatbots?

Others readers like : How to make your Instagram Stories successful?

Chatbots can complement traditional devices (chat advisors, phone…), whether you are in B2B or B2C, or even help the sales department improve sales cycle conversion.

However, you should know that there are several types of Chatbots:

  • Those that are very simple, with a weak ability to understand, but allow manage simple interactions (e.g. ManyChat, Chatfuel…).

These Chatbots work on the principle of keywords, and highly scripted conversation scenarios (the goal is to guide users).

For example, if you use ChatFuel.com and a customer has several ways to ask the same question, you will need to teach the bot most (fortunately not 100%!) ways to apply.

  • Those that allow richer interactions , via a better semantic analysis engine (understanding customer requests via context and synonyms, not keyword analysis), and the ability to launch actions (e.g. placing an order, managing a return voucher…).

For beginners, the design of chatbots seems complex, yet when we understand the basic concept (a system of question/answer between the chatbot and the user), one can realize that a simple and basic bot can be built simply thanks to simple and visual Bot Design tools.

The principle is to define a series of triggers (keywords, conversations, buttons…), and the actions to be carried out afterwards (e.g. answer with a message, ask another question…).

The only essential trap to avoid is letting the customer type their request on their own.

For example, the deadly error is to start the Chatbot with a question like “I am the XXX Chatbot, what can I do for you?”. Doing this will cause the user to ask all sorts of questions (and in very different ways) while the Chatbot will not have been trained for it!

It is therefore absolutely necessary to guide the user very the departure, declaring the mission of the Sculpin (which he knows how to do!) , and from the outset guide the person to the actions to be taken (buttons…).

In this special file Succeeding your Chabot Project, I will precisely show you what a chatbot is, how to build it… in 10 steps, focusing on chatbots in Customer Service.

The principle is the same for a Chatbot Marketing, the difference being that you have to start the conversation with the user, and bring it quickly value (e.g. help with choice…) and then retrieve his contact information (or have it ordered online).

Why use a Chatbot in a company’s Customer Service?

Optimizing customer journeys has become one of the keys to delivering a good customer experience.

The problem is that customers do not follow a single path, and they have needs, expectations… different.

There are a multitude of ways to make a purchase online or in-store, customer journeys and communication channels are multiple and non-linear:

  • visit the site on smartphone,
  • email reception,
  • in-store testing,
  • search for opinions on social media

As many points of contact as possible, with as many key moments to start the conversation with customers and prospects .

But it is also not necessary to stop at a shopping journey… Indeed the Bot can also bring value during other moments of truth:

  • First use
  • Call to Customer Service
  • Invoice renewal
  • Upgrade or Upgrade

With chatbots you enter the “conversational trade”, where businesses and buyers get connect and exchange together.

This can be done via messaging tools (Facebook Messenger, WhatsApp…) or an online chat (chat on the web, in a customer area…), or even on the phone or via connected speakers like Google Home & Alexa thanks to VoiceBots .

Today’s businesses can use chatbots to instantly communicate with their customers and solve their problems across multiple channels, such as Facebook, their online store, their website or the customer area…

The value of chatbots is that they are available 24 hours a day and use artificial intelligence to analyze customer and prospect requests to create a personalized and responsive shopping experience, or serve as an alternative to emailings or phone calls via sending Custom Message on Facebook Messenger, by SMS…

Learning how to implement conversational marketing tools is not always easy as bot technologies are still in development and continue to evolve from month to month, with an exponential learning curve.

What is true now, probably will no longer be in 2 years…

Nevertheless, here are the different family of “Bots”:

  • Chatbots are the bots that appear on the website
  • Messenger chatbots, which are used on Facebook and in Facebook Messenger, often via a smartphone. You can also use Messenger to send messages to Fans, animate & inform live event attendees…
  • Voicebots, ie bots using voice via smart speakers, for example on Google Home, Alexa…
  • Callbots , which interpret what people say on the phone through voice recognition (Speech to text, then semantic analysis of the request) and then give a response via text to speech.

To find out which type of bot is best for you, start by defining the use case (s) (i.e. problems to be solved or solutions to the customer) to find out what a chatbot will need to do to improve the customer experience and bring you a good ROI (ROI).

When you have done this analysis of the most frequent requests (while looking for a minimum of customization of responses not to redo a Bot FAQ that will often be less fast and less effective than an FAQ Classic), and you’ll be ready to start your first chatbot project.

Below is a comparison between Chat /chatbot/FAQ/Tel/Emails with the pros and cons of these different solutions:

Generally this phase is done through the analysis of the most common customer service requests, asking advisors and managers what are the recurring requests from customers (if possible via an estimate of the daily /monthly volume of requests).

If you have a CRM, you can get statistics, otherwise ask for a simple count in an Excel file.

Attention: this analysis must be done with a fairly fine granularity and not in an overly global way.

  • For example, a company may note that order tracking requests represent 15% of calls and emails, with about 500 requests/days.
  • However if there is in order tracking requests for both delivery tracking and parcel management not received, this represents in fact 2 use cases (order tracking & parcel declaration not received).

You don’t have to start realizing your Bot without defining its framework, otherwise you will fall into the deadly trap of the “Bot FAQ” that tries to answer all the questions, but that will never happen…

Indeed, a “FAQ bot” will cause several problems:

  • If you do not restrict his officially presented field of expertise, users will be tempted to ask him everything (and therefore you will have to train him to respond better and better to a request, with the X way to say it)
  • If you add too many questions, the bot will end up mixing the answers, and thus bring bad solutions.
  • If you leave the user write or say anything he wants, he will do it by 1000 different. .. which will require you to program a huge part of it, because the bot will not be able to understand on its own at first.

It is therefore essential to guide clients in the choices to make, and put them back on the “right path”.

Here is an example of a very simple chatbot programmed to animate a customer event, in the beginning message the bot says “here’s what I can do for you”, with a carousel of the top 5 capabilities:

Similarly, the capabilities of the Bot are recalled after each misunderstood question to put the client back in the right path:

With chatbots, a company can immediately respond to customers’ requests to create a personalized and useful customer experience.

Chatbots are more than just computer programs on the web, on a site on a Facebook page: they are a way to create useful and enjoyable shopping experiences for buyers, and to rethink the customer journey (e.g. on boarding process, coaching new customers…).

Today, companies are beginning to recognize the usefulness of this technology and they are beginning to be ready to integrate bots into their customer service.

What is a chatbot?

First of all, it is important to understand what a chatbot is and how it works.

A chatbot is defined as a software to understand a text (or voice conversation), to interpret it in order to deduce one or more actions to take.

These actions can be:

  • respond directly to a request
  • ask a supplementary question,
  • launch a action,
  • escalate the demand to a human (live chat, callback, email…).

This is not a program stupidly scripting or responding solely on the basis of keywords, but rather the beginnings of artificial intelligence, with a program that begins to understand what people say (what some call a CUI or “Conversational User Interface”).

You can also define the chatbot in this way: chatbots are virtual assistants, who help you find information, remember things or buy things.

Think of Apple’s Siri, Amazon’s Alexa, Microsoft’s Cortana and Google Assistant who are the first pieces of these virtual assistants that will be available to you soon.

Here is an example of what bots will be able to do in a few years/months (the demonstration of Google below) being very specific):

If you return to classic chatbots that you can use in the company, you should know that they were initially programmed, and then they are powered by machine learning (autonomous machine learning or with the help of a human).

These bots rely on artificial intelligence & semantic analysis to learn and understand new things.

This learning can be autonomous (the bot itself deduces what to do via a trial/error system), or it is guided by “bot trainers” that train them to improve their understanding or learn new use cases through a manual selection of what to do.

Most bots use the guided training system : once the Bot has been initially programmed, more and more questions are put to it, and the bot trainer tells the bot what to respond to requests it failed to process.

By setting up a Chatbot, businesses and brands can be available online 24 hours a day, 7 days a week, offering near-immediate customer support (just a second to a bot to analyze and answer a question).

For example, it is possible to offer instant responses to common problems, to process simple and recurring requests faster, to implement simple actions like handling complaints…

The bot can “talk” to customers via web chat, Facebook Messenger, Kik, WeChat, Slack…

The important thing is to be present on the channel that is used by customers, and not force the customer to use the channel you find most convenient (e.g. email, call…).

The bot is a communication channel additional, which made almost 11% of customer service interactions in France in 2019.

The chatbot is now an additional channel of communication:

At the base of the bot, there is a technology and a training to understand and respond to requests, which gives it all its intelligence, complementing the link with the information system.

Connecting to CRM, ERP… allows you to output generic responses and offer personalized responses.

Chatbots use learning algorithms, often developed by Google, Amazon, Microsoft, IBM… but also developed by many challengers (recast…).

Thanks to their NLP (Semantic Analysis — Natural Language Processing) engines, chatbots can understand requests and provide appropriate responses according to queries user specific.

As bots allow continuous learning (via a “sandbox” containing unresolved answers), over time they learn to handle more and more cases (e.g. X variant of “I want to know where my order is”/“where is my parcel”/…).

But the real impact of this technology can only be seen if the initial programming /learning of the bot has been done well (the bot being able to understand only what was taught to him, we must teach it all).

Indeed, if initially the company already has difficulty obtaining information (e.g. a specific discount according to the products…), then the Bot will not be able to do better… on the contrary, it will do worse because it responds in a “stupid and nasty” way.

In this, a BOT is NOT a “good genius” who understands by itself, but rather an animal to learn tricks.

To make it simple, a bot

  • is not smart, he only knows what he was taught
  • is not omnitasks (1 chatbot = a selection of use cases)
  • do not have to do a FAQ bot, it is necessary to orient the requests
  • requires time to train it (project in 1 to 2 months)
  • must be put on a subject with volume and minimal customization
  • must be used on a subject where I am already good, otherwise the situation is amplified

Hence the importance of limiting the “tricks” that must be taught to the bot, so that he can succeed them automatically and in a hyper fluid way…

Thus, well-designed chatbots are always focused on a few specific use cases, with a specific personality to excuse misunderstanding errors.

Why is a personality so important to a chatbot?

To give a little context and take a step back, let’s look at the example of the Microsoft paperclip “Clippy”:

This ancestor of the chatbots has been present for many years in the Office Suite, knowing that it was removed afterwards because the AI it used was too limited (the Trombone rarely giving the right answers expected… there is the problem of Chatbots too “FAQ”).

At the origin of the paper clip, there is a person who launched the topic “How to learn the annoying techniques of MS-Word, in a fun way? ”.

And probably someone thought about the paper clip, because it was an object of the everyday life of Word users.

Then they thought about the custom of the Bot: it had to be “nice” but not too invasive, he had to have a little humor…

Because it is known that the personality of the Bot plays its important role in the engagement of users and their degree of “forgiveness” on the imperfections of the answers.

In addition, it is usually out of curiosity that users test the bot to see what it says next.

This makes the product all the more attractive and leaves the user with a pleasant experience.

Is the chatbots now?

User studies show that buyers are more willing than ever to use and buy online with Bots.

Here are a few more reasons why your business should use an email app to host a bot and increase sales:

  • People use more in addition to instant messaging apps.
    • According to a Business Insider report, consumers use the four leading Facebook Messenger, WhatsApp, WeChat and Viber messaging apps more frequently than the four major social networks Facebook, Twitter, Instagram and LinkedIn.
  • People want an instant response, which allows the chat:
    • People don’t just use messaging apps to chat with their friends.
    • They are also ready to ask after-service questions, or even make purchases, with the robots of the brands.
    • Here are some statistics to show the willingness of buyers to buy with email apps:
      • 47% of users are open to purchase items through a chatbot.
      • 71% of users are ready to receive customer service through an app of messaging.

These figures show that many buyers trust chatbots as a way to interact with businesses.

And usage should only increase as early users adopt and bots implemented by leading brands, so now adopt a Chatbot to meet the growing demand for robots.

For example, you can test the Bot of Yes SNCF whose use is significantly increasing, especially among young people:

This is partly linked to the change in channel usage by generation, with new generations who are big users of messenging apps, chat and social networks:

The basics of designing a chatbot!

Here are the 10 steps to make a good chatbot:

STEP 1 — List the use cases that the bot might deal with

The first and most important step is to list the use cases where the chatbot would be the most suitable for solving a user request.

Start thinking about your product around these use cases: service request, order tracking request, duplicate invoice request, price request…

Then estimate the volume of requests /day, in order to classify those with the most demand volume (to make the bot workout profitable, each use case requires a Bot training, therefore a significant cost).

The goal is to list a Top 20 requests, possibly by theme.

This list will then be used to study the feasibility of processing these requests (e.g. connection to CRM…).

To be noted

:

  • you can also list moments of truth when the customer experience is not optimal, to act proactively.
  • for your ChatBot to be economically profitable, it is interesting to also find opportunities to cross-sell, additional selling… Often the chatbot will be there to engage the conversation on a site and guide the prospect to the products, and make an escalation to advisors to transfer the qualified file.
  • Use cases are usually not “static” questions (e.g. the list of distributors, contact details of the service…), but personalized answers for a request (either via a qualification system via question/answer or via a connection to the information system).
  • If you have the chatbot done by a provider, you can stop at this level and fill out a specification that you will deliver to it (see this Specification Template for a Chatbot Project ).

STEP 2 — Select 1 to 4 uses very precise boxes to be processed by the bot.

It is necessary to make it simple at the start, and therefore from your main list of use cases, define 1 to 4 maximum use cases that could be handled by the Chatbot.

This can even be 1 use case, provided that it has enough volume (mini 50 to 100 requests/days), it is customized according to the user (otherwise a classic FAQ would suffice), that it saves employees time, and that we can respond correctly to the request.

For example, for a telephone operator, a bot is particularly suitable for unlocking SIM cards.

On the other hand, getting the code RIO should not be given by a bot (this request has a greater interest in being processed by a retention cell to try to prevent the client from competing).

With this selection you have the ability to determine a Macro conversational design with the Bot’s mission (s) with the great options:

Note: do not forget to include simple answers, the answer to forms… that will complement the cases of major uses of the bot. These static questions are not included in the 1-4 use cases.

STEP 3 — Validate that the selected use cases can be answered by the bot.

The goal is to choose the most relevant cases, both in terms of service rendered to customers, in terms of profitability or image for the company, but especially in terms of feasibility technical.

Indeed, sometimes it will require a connection to the information system, which can be more or less complex depending on CRM or ERP.

To do this, ask your IT department if it is able to “feed” the bot to process these requests.

Other times, use cases, scenarios… are so complex, that it will have to restrict the perimeter (e.g. a population, a link to an ERP…) rather to set up a chatbot “Swiss knife”.

STEP 4 — Model the conversation flow

Convert use cases into a flow of conversations between the client and the bot.

For this write a tree with the script of the conversation, which will allow you to build a prototype of the conversational design.

Do the flow of exchanges on paper/post-it with the script of the conversation.

The principle is this: the customer asks a question, the bot answers…

Again, you need to guide the user from the start so as not to make the conversation disapept:

It is essential to detail in order to list not only the use cases (the requests to be resolved), but also the intentions (i.e. all the questions that the bot must understand for a use case).

Thus a use case “I want to cancel my order”, breaks down into several intentions (questions answers):

  • What is your order number or customer number
  • Do you want to cancel the whole order or just a part

Here are some frequently used prototyping tools:

  • The post it. .. simple but very effective at the beginning to model his conversations !
  • Whatsapp : Just use a colleague’s phone and start typing on your phone and hers. Take screenshots for every conversation.
  • Sketch Marvel : You can easily get templates for online chat elements. Start making art signs with messages and circulate them on a “marvelapp”. This enables prototypes that are very specific to your messaging platform.
  • Justinmind.com to make prototypes with interfaces
  • Miro.com: a mindmapping tool
  • Powerpoint

Until now, it was an easy but crucial part, the realization of the conversational design… But it’s only about 10% of the work, now it remains to train the bot!

Note:

  • If you know well computing, you can start modeling a typical scenario with a tool like Botnation, Manychat, Google’s dialogflow.com … in order to quickly create a conversational design even if everything is scripted (answers and questions are manual and pre-filled).
  • Always start at the end and start, and do not build the decision tree as you go, otherwise you will get lost in complexity… Start with a paper & pencil, and put on both ends of the sheet “Start” and on the other side “Result”, and then add all the steps.
  • From that moment on, cases of use of the chatbot must no longer evolve… otherwise you will constantly evolve the concept and never finish the dossier… It is essential to move forward in these cases of use and finalize the project, and only then expand the scope.

STEP 5 — The Bot Modeling

This is where the real work really begins, because it will have to move from theory (a simple and linear scenario) to practice…

All the content needed to build the chatbot, such as questions, discussion flow, start discussions, tasks to launch, CRM connections, answers to errors… must be collected in a common Excel sheet or in Trello.com (or project management tools like Asana, Wrike…).

The goal is to give the designer/developer of the bot all the elements, so that he starts building the chatbot.

Once the bot is developed on the bot build platform, you should immediately start interacting with the chatbot being built and testing conversations.

It is essential to start your tests right from the start so as not to act once all the coded conversational design… the backward will be very expensive (e.g. redoing the decision tree).

STEP 6 — Optimize Questions and Answers

Based on the set goal, work on the personality and tone of the bot, typical responses, generic messages…

The chatbot tone is important because it defines how the bot should respond to the end user.

For example you can learn jokes at the bot:

STEP 7 — Testing with Operational

Ask people in contact with customers (customer service advisors…) to do tests by having conversations “as customers do” with the chatbot.

You will notice exactly where the chat feed is interrupted, the new ways of posing questions, options you haven’t thought about…

List all these lifts to make adjustments.

Keep in mind that quality control is of paramount importance, otherwise the customer experience will be disaptive!

Indeed, once in production, some creative users will also start asking random questions to the bot, asking twisted questions… and if you have not anticipated the minimum you will have negative lifts.

Note: specify well before the operationals what knows or does not know how to do the sculpin… Indeed there are many fears and fantasies about Chatbots, while finally it is only a tool like the Others.

STEP 8 — Make continuous improvement until you reach the sufficient level of quality

Every time you find out that the bot is wrong, you need to continue training your bot to correctly respond to the most common requests.

If you have too many related requests, appendices… redirect customers with buttons, questions funnels, carousels, buttons…

It is essential not to let the customer type what they want, otherwise he would quickly have made ridicule your bot…

Take note, in the first place, of all the experiences and use cases that have been missed and work around them.

Attention: between the questions asked during the test phases and real life, there is often a big difference… Customers who tend to ask fewer questions than your customer service employees who will always look for the most complex cases…

STEP 9 — Permanently validate the relevance of the bot

It is not necessary officially release an imperfect bot, except to experiment openly with a small number of customers (beta test phase over a limited time) indicating that it is a test.

It is therefore crucial to pass an acceptance test of bot responses to experts in contact with customers.

Again, do not try to answer everything…

The bot needs to clarify the specific tasks it can help accomplish… and what it does not. This is the mission that is declared at the launch of the Chatbot.

Train your robot to say “no I don’t have the answer” in a subtle and polite way, and relaunch on what he knows.

STEP 10 — Fine-tune the communication on the bot and correct the shot with the first returns

Pay special attention to the communication on the bot so that it is used ( newsletter, visibility on the site…).

For the chatbot to be used, it must be naturally integrated into the customer journey.

This means you need to analyze current customer contact points, and see where you can talk about Chatbot as a new medium of communication:

  • In the IVR when waiting phone, with a message indicating the URL of the chatbot and what it can do
  • By making a challenge with the people of the customer service/contact center agent for those who train customers at the chatbot (it’s important to show, see doing with the customer).
  • Send an email to your customer base with a presentation of the Chatbot
  • By adding it on the key pages of the site (e.g. contact page…) and in the client extranet
  • On invoices, quotes, order acknowledgement
  • In the parcel with a flyer
  • During visits by sales reps in the field

All these actions will help develop the use of the chatbot, especially if it is not accessible in a classic customer journey (e.g. it is only in the customer area on a specific page, and not on the entire site).

Also validate from the start that the bot meets expectations, as well as small phrases /chatting… (what is called smalltalk in English).

Simple words like “Super”, “Hi! goodbye, you’re a bot, you don’t understand anything… should have well-designed answers, based on the personality of the bot.

Important: the chatbot should not be just an icon at the bottom right of the screen… it must start and start the conversation via a pop-up.

Tips and tricks for build an engaging personality for your chatbot

TIP 1: Orientate User Requests

A chatbot should direct users to ask relevant questions (not to make unmanaged requests).

These suggestions can be simple message buttons such as “Order Tracking” /“ Invoice Duplicate”…

This helps users choose the right features of the chatbot and understand what the bot does or does not.

And if the user asks a question to which the bot does not know what to say, he must propose tasks that he can perform for the user or pass his hand to a human advisor.

Here are two examples, one or the user is stuck in a “conversation tunnel”, and the other where the bot puts the conversation back on rails:

And what it does should not do:

TIP 2: Indicate what you can do from the start

An effective chatbot should probably start saying hello, and directly indicate how it can help the user with a list/buttons.

Then the bot can be more intimate, and take up the name of the user, greet him, congratulate him, etc. These are practices that make conversations more personalized.

Similarly, from the start show what the Chatbot can do, either via an example (e.g. a practical case…), or by providing value-added information in seconds (e.g. a tip…).

If you do not do this, the person will get tired very quickly (maximum 5 exchanges) of the Chatbot.

It has 2 risks to encourage the client to make free text entry systematically in the mission:

  • Let the chatbot do not understand, and so that the answer is disaptive
    • Because the request is drowned, in a phrase and too broad “I have a problem with my doll I received yesterday, the hair brush is missing in the box”/“my parcel arrived all dirty and the delivery man was rude. I want to complain.”
    • Because the request is too close to another intention: “I want an invoice” (request an invoice) and “there is an error in my invoice” (a claim)
  • That the customer ask questions not foreseen
    • In this case, it is an endless run to handle all specific cases for example “do you have in stock dolls lol?
    • ” Therefore, it is better then to recall the missions and say what they know how to do, and display the message of misunderstanding (with customer service contact information if necessary)

So we have to start the customer’s place, and display the most numerous use cases (plus calls to marketing action), and accept that low-volume requests be handled by advisors.

TIP 3: Sometimes it’s a matter of timing

In some use cases, it takes some time to process the user’s request (e.g. connection to the CRM…).

In such cases, add information messages to maintain user engagement.

Never make the user wait without doing anything in front of a frozen screen, this is a basic rule.

Similarly, avoid having more than 5 questions to answer a request, otherwise you will make the customer impatient…

TIP 4: Add emoticons to your conversation

Much of the conversations include emotion.. hence the use of emoticons to give volume and body to interactions.

For example, you can copy and paste emoticons on https://getemoji.com/

The emotion behind your conversation must be consistent with the purpose of the robot.

  • If the user has completed a task, give a joyful answer.
  • If the user was unable to complete a task, feel the downside and pass it in the right way.

TIP 5: Add a “Home” button to go back to the home level

Sometimes customers navigate the Chatbot (maximum 3 levels), and so they may want something else or go up to the next level.

To avoid these manipulations, plan navigation buttons, such as a return to the home:

TIP 6 — Do not try to solve everything automatically, and in this case give preference to climbing to an advisor

By For example, if you want to set up a password recovery system, do not try to process everything automatically by a chatbot.

1 — If an application makes 80% volumes, then make a web service (5000€)

2 — For other applications, putting links to the password reset, it is up to the customer to do the process 3- If the customer does not get there, then climb to the advises

If you want to set up a Callbot, the principle is the same, we can propose to choose the application, then request a login (or so identification of the phone number) and then send an SMS with the reset link of the password.

These are simply designed chatbots that work best!

It is interesting to note that a chatbot has the ability to provide solutions to many uses cases… but that more you will give him more tasks to do, the more complex it will be to create and make it relevant.

Sometimes it is better to create X specific bots, possibly driven by a “master bot”, than to create a “universal” bot.

With a specific bot, there is almost no need to handle different switches, similar request types… because the bots will not be visible or launched in the same contexts.

For example, one can do a chatbot only on the service, another for quotation requests…

Chatbots will change the world… are you ready?

According to a report relayed on the internet, 63% of people would consider using an online chatbot to communicate with a company or brand.

A survey conducted by specialized agencies also revealed that they would use chatbots to obtain “quick response in case of emergency”.

In addition, a majority of consumers say they are willing to make a purchase via a chatbot.

A study showed that 37% of consumers are willing to receive recommendations or advice from chatbots, in particular:

  • products from retail stores (22%),
  • hotels and accommodation (20%),
  • travel (18%),
  • Pharmacy products (12%)
  • fashion (9%).

However 73% of people said they would not use a company’s chatbot after a bad experience … hence the importance of testing your bot well before launching it

Finally, an overwhelming majority of consumers (75%) want to know if they are chatting with a chatbot or a human.

But it is quite possible that within a few months it will difficult to differentiate between Bot and Human.

According to Ray Kurzweil, inventor, futurist and engineer at Google, chatbots will have a human-level linguistic ability by 2029 “If you think you can have a meaningful conversation with a human, you can have a meaningful conversation with an AI in 2029. But you can have interesting conversations long before that.”

Macro schedule to set up a Chatbot

Below is an example of a macro schedule for a very simple Chatbot in “FAQ” mode with generic questions and answers.

This type of project is simple to implement and can be deployed within 1 to 2 months maximum.

Important: as part of a Chatbot connected to the information system, the workload will be much higher, with at least 4 meetings additional (1 2h framing workshop 1 2h specification workshop 2 1 hour technical test workshops). This kind of Chatbot project takes place over 2 to 3 months.

  • Before the start of the project (2 hours): 1 marketing person & 1 customer service person
    • Set the goal of your chatbot (0.5 h):
      • Is it to relieve customer service, help marketing…
      • What is your chatbot user persona?
      • What is the expected timing?
    • List the most frequently asked questions to Customer Service or by email or phone according to your intuition (0. 5 h):
      • Just list the questions you think are most frequently asked. Check this list for example by analyzing the most viewed web pages (FAQ…) on your website or simply your intuition.
      • Take the opportunity to put these questions in an Excel /Google Sheets file these questions, and if possible the URLs with the answers (or if you need to write a text or a web page)
        • If you have a lot of questions, try to really limit yourself to the essentials (if necessary put a rating of 0 to 10 depending on the priority)
    • Ask Customer Service Operatives for the Top 10 of the most frequently asked questions, in COVID context and in normal context, and if they have standard answers by email/tele/ Web (1h):
      • Gather customer emails on these questions, and typical answers to send them back to us to feed our knowledge.
      • If customer service is not able to give you the answer immediately, just ask 1 or 2 customer service agents to “beat” on Excel the one-day requests and quantity.
      • Then differentiate between “dynamic” questions (customized according to the customer, for example tracking an order, a balance of points…) and static questions (e.g. where to find my points balance, how to make a product return, is what stores are open, what is the phone number of a store…).
      • It is important to note that a personalized question can be diverted by pointing it to the self care (ex: I want to track my order è return to the customer area button to request his password again)
      • Prioritize the questions/answers according to the volume and the ability of the chatbot to answer that answer.

==> At the end of this step you must have a list of 15 to 20 questions, and an idea of the answers on Excel (whether they already exist or not).

==> It may be interesting to already group the questions into themes

==> Ideally you can give us this Excel file in upstream of the launch meeting so that we can make a pre-mockup.

  • Launch Meeting (2 hours) — Project Leader 1 Customer Service Person
    • Introduce us your persona (the goal is to help us know who the person is)
    • List FAQs and Categories to Organize Chatbot Architecture
      • Welcome Message (the main categories)
      • Subcategories with FAQs
    • Exchange on personas, FAQs… if possible around a model

=> At the end we have a Powerpoint document with a macro tree of categories of questions. But if possible we are already working on a model of Chatbot

  • Conversational Macro Design Validation Meeting (2h) — Project Manager 1 Customer Service Person
    • Introduce the model following Launch Meeting with Architecture
    • Have your feedback on the organization, the macro scrolling, the first built-in FAQs, the tone of the bot…

==> At the end we have a validated macro conversational design (categories, tree).

==> We will deliver a model to test a few days later.

  • Work on the model to test selected FAQs (1h): Project Manager and Customer Service
    • Check that for the selected FAQs you have the questions and answers, and if necessary create web pages.
  • Recipe Workshop 1 (2h) — Project Manager 1 Customer Service Person
    • Introduce the model following the macro conversational design validation meeting
    • Have your feedback on the organization, the macro scrolling, the first built-in FAQs, the tone of the bot…

==> At the end we have evolved the model (categories, tree).

==> We will deliver a model to test a few days later.

  • Correction check and improvement (1h) — You test the corrections made over the water.
  • Recipe Workshop 2 (2h) — Project Manager 1 Customer Service Person
    • Introduce the model following the recipe meeting 1 but this time on the Production environment
    • Have your feedback on the organization, the macro scrolling, the first built-in FAQs, the tone of the bot…

==> At the end we have evolved the model (categories, tree).

==> We will deliver a model to test a few days later.

  • Correction verification and improvement (1h) — You test the corrections made over the water.
  • 2 Session of 0.75 h by 2 or 3 people of customer service, and possibly 1 or 2 customers beta tester.
  • Recipe Workshop 2 (2h) — Project Manager 1 Customer Service Person
    • Introduce the mock-up following the recipe meeting 2 in the production environment
    • Have your feedback on the organization, the macro scrolling, the first built-in FAQs, the tone of the bot…
  • Correction check and improvement (1h) — You test the corrections made over the water.
  • Validation Workshop (2h) — Project Manager 1 Customer Service Person
    • Introduce the mock-up following the recipe meeting 2 in the production environment
    • Have your feedback on the organization, the macro scrolling, the first built-in FAQs, the tone of the bot…

==> At the end, we validate the conversational design and give the GO /No Go for production

==> We deliver the script to install the recipe PV to sign.

  • Post-launch project follow-up: 3 x 1h — Project Manager 1 Customer Service person over 3 weeks
    • Debrief on the first days after launch

==> We make the necessary adjustments (e.g. terms, misunderstandings…)

  • Team training (1.5 days, in 2 sessions) — Project Manager 1 Customer Service Person
    • Chatbot Level 1 Administrator Training

==> Make yourself autonomous on the solution (adding use cases, correcting misunderstandings, adding buttons…)

Mistakes not to be made in a conversational design

Writing messages in a chatbot requires a little habit, because there are limitations related to the size of the window, reading the mobile…

Don’t say…

  • in introduction of the chatbot “How can I help you” but list via buttons the main missions of the chatbot (80% of the volume)
  • click on this link, but instead propose buttons with emoji
  • “Call us at 0 825 xxx xxx Monday to Friday from 9am to 6pm” but “Call us ☎️ 0 825 xxx xxx Monday to Friday 📅 9am to 6pm”
  • “if you have other questions type your request” but add a “Home ↩️” button at the end of each sentence
  • Large message pads, but divide your texts into small blocks, or subconversations for ease of understanding, or offer a PDF to download.
  • “Our in-store teams are at your disposal! because in writing punctuation could make it clear that you “shout”
  • In the lists of bullets like — or*, but more graphic emoji ➡️
  • Do not use negative words but results-oriented texts (e.g. “search”: Search è Find)

Conclusion… the keys to a successful Chatbot project

If there is one thing you need to remember is that if you want to set up a chatbot project, you must first define the goals, especially the tasks that the bot needs to accomplish.

It is essential to define the right use cases (those that interest your customers and you know how to do well).

If try to automate something you don’t already know how to do manually (or that is too complex), the Chatbot will not do it better (on the contrary!).

For this to work, you need to offer a feature that has volume, brings value to the customer and brings added value to your customers.

For example, this can be a configurator in order to find out which product to buy (but it must bring a real value not a list too simple), so that the person can make the right choice and buy.

To have this list of “key” functions, you can ask your customer service for which volume and automable requests are (simply list the types of requests for 2 or 3 days and see in this list the volume concerned), or ask your reps how a chatbot could help them optimize the sales cycle (e.g.: a chatbot that opens on the site automatically to start the discussion and qualify a prospect).

Here is an example of the importance of having the right use case: the first graph presents the average satisfaction of a chatbot (62%), and the right one is the case of a chatbot with a use case and an optimized conversational design.

The clearer your specifications, the more time and efficiency you will gain in setting up the Bot.

Note that climbing must do it differently depending on the request:

  • For some key actions the contact by phone/live chat must be highlighted because it is essential or it brings added value (ex: finalizing an order, pb of delivery…)
  • For the form embedded in the chatbot, use it in case of dissatisfaction after the satisfaction measurement so that you are kept informed to correct the dissatisfaction.
  • On the other hand, other less important requests are the form that must be preferred in order to this is part of the current processes

Download the Checklist to succeed your Chatbot project!

Do you want to set up a chatbot?

Send me an email to fred@conseilsmarketing.com to validate together where and how to set up a chatbot in your company.

I also offer you my Checklist to set up a chatbot in your company.

4.2 / 5 ( 34 votes )

RELATED POSTS