10 Questions to Ask Before Starting an Enterprise Chatbot Project
You might be thinking about starting an Enterprise chatbot project or evaluating a chatbot solution.
Great! That means you already know the importance of this new channel for the customer experience and how it will bring conversations with your brand to the next level.
But where do you start?
For a successful enterprise-level chatbot project there are many requirements:
- Defining the scope of the chatbot solution
- Understanding the meaning behind training data
- Dedicated marketing operations to maintain the chatbot conversation flow
- And, of course, a skilled team!
We have compiled some of the most important questions and answers to help you get ready for your enterprise chatbot project.
What kind of chatbot is the best for enterprise?
Chatbots leveraging Artificial intelligence, Machine Learning and Natural Language Processing (NLP) are the best fit for enterprises.
The advantages of AI chatbots are many: they get smarter over time and learn by interacting with their audience.
AI chatbots are more flexible and allow integration with core systems and other applications. They enable better interaction and personalization.
Plus, the look and feel can be adapted to branding to offer a coherent, beautiful user experience.
AI chatbots are the only feasible enterprise-grade solutions and offer way more capabilities compared to rule-based chatbots.
How to define the scope of a chatbot project?
Start by brainstorming with your team. An ideation workshop is a good first step.
This will help you identify how your chatbot could be used within your organization, and what purpose it will serve.
Keep in mind that you are opening a brand-new channel for your clients, similar to a website or mobile app. Scoping the project deserves proper attention.
Finally, once you have collected a lot of ideas, group them together by main topics. You will find that some topics will be related to marketing, others to sales or customer service.
We don’t recommend building the first version of your chatbot with a broad purpose in mind.
Pick one primary topic and start a first concept for the chatbot, also called a Minimal Valuable Product (MVP).
Define the MVP by analyzing the requirements from a business perspective:
- Who is your target audience?
- What problems are they facing?
- What benefits/solutions does the chatbot offer?
Knowing your audience is key to identify their needs and how they could benefit from a chatbot.
Once you have all information and input from various stakeholders, the MVP can take shape and you can begin discussing first key features.
The MVP represents your initial scope. All the remaining ideas are your vision and roadmap for the future.
What skills and team members are required for a chatbot project?
A chatbot project is a very interdisciplinary project. Skills from all areas are needed.
- Leadership: they know the processes, concepts and business goals. All non-technical decisions will be driven by them.
- Project managers: as many people and areas are involved, somebody needs to keep an overview. Independent of the chosen process, a project manager needs to ensure that all required coordination between the different skills take place.
- Chatbot experts: The person who knows chatbot concepts and AI. This expert is able to implement the chatbot but also knows how to maintain a chatbot and what the technology’s limitations are.
- Content editor: somebody who writes the bot’s responses and creates content.
- Software architects: The bigger your system and environment is, the more important. In enterprise environments a proper architecture will be required, especially to integrate business and third-party APIs.
- Developers: Frontend and backend implementation must be done to create an enterprise chatbot solution. Depending on the degree of customization and special use cases (business logics) the amount of implementation can vary.
- Testers / QA: as in every project, someone needs to drive a strategy that ensures quality.
How to start with training data?
Training data is often misunderstood.
It’s not automatically generated by the bot, rather the opposite. You will have to create a set of training data to make the bot smarter.
One key advice: use data as realistically as possible.
Ensure this by using the following data sources:
- Existing conversational channels: Get inspired by your customer support email communication, social media communications, phone, recordings/statistics or even the comments on your blog.
- Ask the experts: employees in direct contact with end customers know the common questions and the way customers communicate.
- Leadership: they’re not in direct contact with customers, but it’s their job to know the customer. Convince them to invest the time required to gather data. If they’re not willing, the project is likely not important enough and lacks the required support from business/marketing teams.
Depending on the data source you will need to adapt it to a conversation style.
To get started, you need a minimal set of data. Once your chatbot is live, you can monitor conversations with the chatbot to fine-tune dialogs and retrain your system.
Your generated phrases typically will not cover all phrases entered by real users. Therefore, rather invest in a beta phase instead of writing a huge training data set by hand.
Is development needed for an enterprise chatbot?
Entreprises expect solutions tailored to their exact needs. This means that yes, custom development is needed.
However, not everything must be created from scratch when implementing a chatbot solution.
The core engine of the bot, the AI, can leverage state-of-the art frameworks such as DialogFlow to enable natural language processing.
The development effort will be focused on the look and feel of the bot interface within your website or mobile app.
Plus, integration with your core system or marketing tools (such as analytics or marketing automation) will require a dedicated effort.
How to connect a chatbot with a marketing automation or any other third-party systems?
Chatbot frameworks offer webhooks to trigger actions or modify responses given parameters and intent.
Within the actions, processes can be started or third-party systems used.
On the other hand, we are able to post-process responses from the chatbot framework with our Adobe Experience Manager module for chatbot management.
Given the detected intent and parameters, we can start business processes, validate parameters, but also use business or third-party APIs to gather further information.
Additionally, marketing automation and targeting/analytics information can be updated. Thanks to the availability of additional information, very specific processing becomes feasible.
What effort is needed to maintain a chatbot?
An important thing to realize is that a chatbot must be maintained.
A chatbot needs regular maintenance even though it uses AI. The following maintenance will be required:
Training Data of AI
Your chatbot will never understand all phrases.
Improve the system by adding training phrases provided by your visitors. Map them to the correct intention. Additionally, add unforeseen intents that your visitors asked for.
Take care of responses. Content can change over time, e.g. office locations.
It’s important that you update information delivered by the chatbot as soon you update something related on the website. An integration of the chatbot management into the CMS, e.g. with our Adobe Experience Manager (AEM) Module for chatbot, enables you to re-use content in multiple channels.
You will notice that rephrasing improves the user experience. New intent responses must be created over time.
After the go-live
Directly after go-live the effort to maintain the chatbot is high.
Optimize the system as quickly as possible – customers are impatient and if their first experience with the chatbot is not pleasant, they will hesitate to chat with the bot again.
After a couple of weeks, the system will understand typical requests from your visitors. Only occasionally, unpredictable phrases will appear.
Check the system frequently and provide the best possible user experience by optimizing all responses.
How to personalize the answers of a chatbot?
Typically, chatbot frameworks do not offer response personalization.
However, if responses are maintained and provided by a dedicated system – like the content management system – personalization is feasible.
With our AI chatbot module for AEM, you benefit from personalization features for chatbot responses.
Because the responses are AEM components, we can target and personalize them. We rely on the functionality for personalization provided by AEM. We prevent specific implementation for chatbot personalization.
Why not just adjust the on-site search instead of developing a chatbot?
A counter question: why have on-site search? You already have a navigation on the website which ideally guides the user to all content.
One of the benefits of a chatbot is that it covers preferences of many different users. The chatbot is an additional channel which can guide the user towards content they’re looking for.
How does the chatbot differentiate from an on-site search? Both enable the user to search for content.
An on-site search uses the content available on the website. It is centered around the website terminology and is a one-step approach: it maps queries to a result list.
A chatbot involves three different concepts:
- Natural language understanding: it enables the user to use full sentences.
- Stories: it guides the user by stories and uses a multi-step conversational flow to find out which information the visitor needs. Instead of a list of results, the chatbot can provide a specific answer for each user.
- Real-world content: it goes beyond mapping website content to use real phrases and terminology of users. It matches the terminology users actually use.
A chatbot is a complement to an on-site search and not an opponent.
What about multi-language sites and chatbots?
All major chatbot frameworks handle multiple languages.
However, this means you also need to maintain training data and responses in multiple languages. Don’t mix all languages in one chatbot (AI model). Otherwise, you’ll complicate learning.
In our AEM chatbot module, multiple languages are handled. We follow the same multi-language approach as for multi-language websites. This enables authors to easily maintain multi-language chatbots.
Starting a chatbot project might feel complicated at first, especially since some tasks such as conversation design and training data are not standard with other digital solution projects.
But the project doesn’t have to be unnecessarily complicated.
With smart planning and the right requirements you can make it happen for your enterprise.
The key to success is to understand the steps, the effort, and the resources required. To continue learning about chatbots for enterprises, we have created a complete guide to chatbots for enterprises.
Senior Software Architect