Building a customer service chatbot, from a product owner’s perspective (2022)

After 1.5 years working on a chatbot project at a large financial institution, I have learned a thing or two about building a virtual assistant. If you are thinking about starting up a similar project or struggle with an ongoing implementation, I hope this article can help you in some way. If you want to deep-dive into the matter, don’t hesitate to leave a comment or reach out directly.

First a little bit of inevitable jargon, to make sure we are all on the same page:

  • Intents: the task that a user wants to do (get information, do a (trans)action, report a problem, …).
  • Expressions: how a user phrases his intent (sentences, questions, commands, …). Also referred to as utterances.
  • DialogStates: nodes in a conversational tree structure, that correspond with a answer that the chatbot can provide. Intents are basically ‘shortcuts’ that initiate a chatbot session at a certain DialogState.
  • Bot replies: the actual answer a user will receive from the chatbot.
  • NLP: Natural Language Processing, a discipline within AI that focuses on processing human language by using machine learning techniques.

Before the start of any project, it’s important to clearly state the objectives. This chatbot is positioned within a Customer Care context and the project had 2 goals:

  1. Learning about new technologies: NLP, AI and conversational UI
  2. Explore if a chatbot has potential to replace website FAQs as the #1 digital self-service support channel

In todays world of structured and unstructured big data, online analytics, metrics, dashboards, etc. it’s important to focus on what’s useful for the task at hand. When building a customer service chatbot, you basically need one thing: CONVERSATIONS.

When we started our project, there was little to no textual conversational data available. The large majority of all customer interactions were phone calls, .wav recordings stored away on servers within the bank. So we had to choose our initial scope based on FAQ article views and keyword search analytics. In retrospect, not an ideal starting point for setting up a chatbot. Why? Because FAQs are relatively large articles that contain a lot of information. People with rather specific questions can all find an answer within the same FAQ article. So you can consider this a game of “Jeopardy”, we know the answer and we need to guess the question(s) that go with it.

(Video) How to build a customer service chatbot | Freshdesk Webinar

Building a customer service chatbot, from a product owner’s perspective (1)

In an ideal world, you have access to a large set of chat sessions between contact center agents and customers. This conversational dataset allows you to identify the real needs of a customer, formulated in his/her own words. Then you can start clustering these expressions and identify intents. Once you have a set of intents, you can proceed with building a conversation tree to structure DialogStates and link intents accordingly.

If you don’t have the necessary conversational data, I would suggest to use the “Mechanical Turk” principle. This is basically putting a human workforce behind the chatbot to provide answers during a POC/start-up phase. You need a conversational UI that simulates a chatbot user experience and a couple of customer service agents that handle the chatbot conversations in real-time. Users will get a top notch user experience and you will be able to gather the necessary conversational data.

So you now have access to the necessary conversational data and you managed to cluster a number of expressions into intents. One of the hardest things to do now is coming up with an intent naming convention that is unambiguous, logical & hierarchical. I can’t stress enough how important this is, because your whole chatbot team needs to be able to work with it and its the core of your NLP. It’s one of the biggest lessons learned for my team so far.


For any given user expression, it should be 100% clear to your NLP trainers under which intent it must be categorised. If there is any doubt, you risk that multiple AI trainers will categorise similar expressions under different intents => this will lead to intent confusion, which degrades your overall NLP performance.

(Video) How To Build A Customer Service Chatbot In 10 Minutes (Manychat Tutorial 2019)


As your chatbot grows, the number of supported intents will increase massively. Our chatbot currently has +/- 150 intents, but we are preparing for an upscaling towards 1000–1500 intents. In order to do that, you need a logical structuring of intents. When an AI trainer analyses an expression for intent matching, (s)he needs to deconstruct the sentence and identify subject, verb, details, … this logic should be reflected in the intent naming, so that matching can happen effectively.


Building on the logical aspect of the intent naming structure, it is a sound practice to include a hierarchical approach. The idea here is to build on so-called “base intents”, which hold very basic expressions (very short sentences). The more detail a user provides in his expression, the more detailed a NLP trainer can assign the appropriate intent. As soon as there is information missing or room for interpretation, the NLP trainer should be able to default to a higher intent and let the DialogState pick up the conversation from there.

Building a customer service chatbot, from a product owner’s perspective (2)

(Video) Let's Talk Chatbots! w/ Radhika Naikankatte (Product Owner at SAP Conversational AI)

Building a chatbot needs the right team to make it work. The following roles have proven to be indispensable:

  • NLP trainer: maintains the conversation tree, DialogStates, bot replies; provides expressions for existing intents; suggests creation of new intents.
  • NLP gatekeeper: maintains the intent structure (CRUD) and tests for intent confusion
  • Developer: integration chatbot on your digital channels (website, app, …)
  • Product owner: sets out overall strategy of the chatbot, responsible for stakeholder management and sets task priority within the team.

Limit expression length

This is particularly useful for customer service chatbots. People tend to include a lot of information in their initial expression. This makes it very difficult for a chatbot to understand and the resulting bot reply is often the dreaded “Sorry I did not understand your question, please rephrase”. In order to anticipate this, you can limit the number of characters a user can type. Ideally a user expression should be limited to a concise question. Aim for max. 100 characters (but make sure to check your conversational data set & tweak the number to your user base).

Chat with a human: ok, but not so fast

Your chatbot will often fail and leave the customer frustrated (especially in the early stages of your bot development). It’s a best practice to give people the option to chat with a human when the chatbot gets lost in translation. However, make sure this “exit strategy” is not too convenient. If someone can call in the help of a human with a single command, they will learn quickly and it will become the norm for returning users. Make sure users try at least 2x to express their need, otherwise the bot becomes obsolete.

Give your chatbot a personality, but don’t try to be a person

Make it clear upfront that people are chatting with a chatbot. Otherwise they will get confused and irritated when things go wrong. But don’t let it discourage you to create a bot personality. This allows you (among other things like name, avatar, …) to choose a “Tone of Voice”, which is an effective tool to explain the scope and connect with your target audience.

Building a customer service chatbot, from a product owner’s perspective (3)

(Video) Experience Low-Code Chatbot Building with SAP Conversational AI + DEMO | SAP TechEd in 2021

Use NLP, but use buttons and media whenever possible

Try to capture the user expression with an NLP engine, but embrace the use of buttons and media (images, videos) for your bot replies. It gives users confidence and the feeling they are getting closer to the answer to their question.

Building a customer service chatbot, from a product owner’s perspective (4)

The use of chatbots is still in a experimental phase at the moment (2018–2019). There are a lot of NLP providers out there, so make sure you don’t tie yourself into 1 system too much. Keep ownership of your intent+expressions catalog and bot replies. Test out competing NLP providers and compare the results.

Building a customer service chatbot, from a product owner’s perspective (5)


How do I create a chatbot customer service? ›

Build a simple support bot
  1. Step 1: Create a basic knowledge base. The goal for any customer support bot is to provide answers to simple, straightforward questions. ...
  2. Step 2: Diagram your path to transaction. ...
  3. Step 3: Write a dialogue flow script. ...
  4. Step 4: Choose a chatbot platform. ...
  5. Step 5: Build and launch your bot.
Dec 16, 2020

How can chatbots be used for customer service? ›

AI chatbots use your existing resources, such as FAQs or knowledge base articles, to help answer and resolve your customers' questions. They can recognize and answer multiple forms of the same question and can be trained to give instant responses using your preferred voice and tone.

What are the 7 steps to create a chatbot strategy? ›

Let's walk through them in order.
  1. Audience. The first key to a successful strategy is to profile your ideal customers. ...
  2. Goal. To define the purpose or goal for your chatbot strategy, begin with the end in mind. ...
  3. Performance. ...
  4. Key Intents. ...
  5. Storytelling. ...
  6. Platform Strengths: ...
  7. Feedback.

What is chatbot example? ›

A chatbot helps in collecting contact information, providing available listings, and book viewings. Structurely's chatbot, Asia Holmes, is a great AI chatbot example to handle customer queries in real-time and make conversations effective.

Why AI chatbots are transforming the customer experience? ›

Chatbots help brands to fulfil customer desire by giving prompt response and farfetched resolutions. This, as a positive consequence, leads to uplifted brand loyalty. Owing to artificial intelligence, chatbots are also able to give human touch during service interactions. This again, leads to high CX levels.

How chatbot works step by step? ›

A chatbot communicates similarly to instant messaging. A chatbot is software that simulates human conversations. It enables the communication between a human and a machine, which can take the form of messages or voice commands. A chatbot is designed to work without the assistance of a human operator.

Which terms are important to know before you build your bot? ›

Before you take the leap into the exciting world of chatbots, keep in mind 1) what it needs to do, 2) where it will live, 3) what information it needs, 4) when it will escalate, and 5) who will build it. Answering these five questions will set you—and your chatbot—on the path to success.

When determining the goals for your chatbot solution What should you consider? ›

10 Steps to Define Your Chatbot Strategy
  1. Define Your Goals. Before you develop a chatbot, you should outline your goals. ...
  2. Understand Your Users. ...
  3. Learn from Competitors. ...
  4. Pick a Platform. ...
  5. Capture Requirements. ...
  6. Prioritize Your Desires. ...
  7. Consider Brand and Build Your Bot's Personality. ...
  8. Design a Conversation Flow.
Jul 13, 2017

Which platform is best for chatbot? ›

14 Most Powerful Chatbot Development Platforms To Build A Chatbot For Your Business
  • WotNot. ...
  • Intercom. ...
  • Drift Chatbot. ...
  • ...
  • LivePerson. ...
  • Bold360. ...
  • Octane AI. ...
  • Flow XO.

Which framework is best for chatbot? ›

Best Chatbot Frameworks in 2022 and Beyond
  • Watson Assistant by IBM.
  • Chatfuel.
  • RASA.
  • CX DialogueFlow.
  • Microsoft Bot Framework.
  • Amazon LEX.
  • Final words.
Jun 27, 2022

How difficult is it to build a chatbot? ›

Because building a chatbot with code is immensely difficult for people with no development background and limited exposure to coding languages, it's good to research sample chatbot code from expert developers as a jumping-off point for those determined to learn how to build their own bot without help.

What should be included in a chatbot? ›

What Makes a Great Chatbot?
  • Understanding Context. Another way of putting this is conversational maturity. ...
  • Seamless. Chatbots should be omni-capable, utilizing data from a range of sources. ...
  • Integrated. ...
  • Emotionally Intelligent. ...
  • Flexible. ...
  • Be Available Where Your Users Are. ...
  • It Should be A Conversationalist! ...
  • Well Integrated.

What is the purpose of a chatbot? ›

At the most basic level, a chatbot is a computer program that simulates and processes human conversation (either written or spoken), allowing humans to interact with digital devices as if they were communicating with a real person.

What is the main use of chatbot? ›

A chatbot is a software or computer program that simulates human conversation or "chatter" through text or voice interactions. Users in both business-to-consumer (B2C) and business-to-business (B2B) environments increasingly use chatbot virtual assistants to handle simple tasks.

How can AI help customer service? ›

One of the most common uses of AI in customer service is chatbots. Businesses already use chatbots of varying complexity to handle routine questions such as delivery dates, balance owed, order status or anything else derived from internal systems.

What is chatbot service? ›

At the most basic level, a chatbot is a computer program that simulates and processes human conversation (either written or spoken), allowing humans to interact with digital devices as if they were communicating with a real person.

What is customer service automation? ›

Customer service automation is the process of automatically resolving customer queries through the means of self-service resources, messaging, and web chat functions.

Why is AI good for customer service? ›

Personalized user experiences

AI can play a huge role in helping customers find the right information more efficiently. Artificial Intelligence helps analyze customers' data and key metrics, and recommend products or services to customers based on their browsing/buying preferences.

How does AI improve customer experience? ›

Why use AI to improve Customer Experience? Artificial Intelligence, with its power to gather and analyze customer data in real-time, is helping in getting a better understanding of customer behavior and needs, and eventually creating a personalized customer experience strategy.

Is AI the future of customer service? ›

It's predicted that the use of artificial intelligence (AI) in customer service will increase by 143% by late 2020. That means that AI can't be ignored, for both consumers and businesses. The reality is, many people are still suspicious or nervous about AI and its implication for their business.


1. A Fireside Chat with Kyle Cogger, Senior Product Owner for Expansion of Revolut
(NYU Stern)
2. New year, New Bot
(Conversational Collective)
3. The four-letter code to selling anything | Derek Thompson | TEDxBinghamtonUniversity
(TEDx Talks)
4. Agile User Stories : How to write a Good User story
5. How to Develop a Successful Chatbot Concept
6. GTC 2022 Keynote with NVIDIA CEO Jensen Huang

Top Articles

Latest Posts

Article information

Author: Cheryll Lueilwitz

Last Updated: 01/26/2023

Views: 6213

Rating: 4.3 / 5 (54 voted)

Reviews: 93% of readers found this page helpful

Author information

Name: Cheryll Lueilwitz

Birthday: 1997-12-23

Address: 4653 O'Kon Hill, Lake Juanstad, AR 65469

Phone: +494124489301

Job: Marketing Representative

Hobby: Reading, Ice skating, Foraging, BASE jumping, Hiking, Skateboarding, Kayaking

Introduction: My name is Cheryll Lueilwitz, I am a sparkling, clean, super, lucky, joyous, outstanding, lucky person who loves writing and wants to share my knowledge and understanding with you.