Are you looking to automate simple and repetitive calls with a callbot? While chatbots tend to be more generic, virtual switchboards meet specific use cases, such as signing up to a contract, reporting an incident or tracking an order. Depending on your pre-defined strategy, a callbot can handle end-to-end calls or pre-qualify a request and transfer the customer, and their conversation, to an operator. This article takes a look at the different stages involved in implementing a callbot with dydu.
Callbot Objectives and ROI
The first step before launching a callbot is to precise your objectives, your end-users’ profiles and their needs. This will enable you to define the bot’s scope and use case(s). Callbots, can be used to assist or partially replace call centers. The processes and scripts are therefore often largely pre-existing, which makes it fairly easy to define the bot’s scenario(s) and potential savings.
To ensure that the project is profitable, it is essential to establish certain KPIs from the outset. You can assess your ROI based on hypotheses of time management and savings rates. Let’s take for example an insurance claim scenario. If the callbot is in charge of checking the user ID before the conversation begins, then you can look at the percentage of correct authentications in wich timing… You can also compare and extrapolate with exclusive use of the callbot outside of business hours.
Customer Flow and Callbot Technical Requirements
The next step is to define where the callbot fits into your customer flow and any technical requirements. Ask yourself the following questions:
- Which number will the callbot be reached on? At what times?
- Will the callbot receive calls directly or will there be an intermediary IVR (Interactive Voice Response)?
- Will the caller be redirected to a human operator?
- Should any information from the conversation be passed onto this operator?
- How many simultaneous calls should you expect?
These questions are usually analysed during the commercial phase. The next stages focus on creating and integrating the bot.
User Journey and Data
Before building your bot’s knowledge base, it is important to define your user journey and the data you want to collect , thru two workshops.
Use Case Workshop
Callbots meet very specific use cases, such as reporting or checking the status of a claim. We therefore recommend creating a flowchart for each use case to establish the customer journey on the phone. This flowchart should include:
- The exact text spoken by the bot : this will enable you to check that the conversation flows smoothly. For example, asking “what day did you make your request?” rather than “request date” allows to ensure that the user knows what to do and what kind of answer to give.
- The answers expected by the bot : such as “yes” or “no”, as well as the structure of certain data required for calls to web services. For voice recognition, this often involves numeric data (phone number, licence plate, customer ID, file number, etc.) or alpha data (family name, type of problem, etc.).
- Decision Trees : this allows to identify what answers to expect from users. For example, if the callbot asks if the user knows their file number, you should provide a “yes” and a “no” branch. And maybe also an “I don’t know” or “where can I find my file number” branch.
- Interconnections with a webservice with data to be transmitted and values to be retrieved : this allows to see if the use case can retrieve all the necessary information and the structure of the expected values. Similarly, you can prepare use case answers with the list of possible responses.
- Escalation to human operator if applicable : in which situations and with which information.
At dydu, we support our clients with a project manager during the entire project. They also prepare the flowchart with you.
Analytics Workshop
Callbot conversations have a defined beginning and end. You can obtain analytics on completed dialogues, conversations where the customer hung up before the end, which use cases were involved, why the calls were made, etc.
This workshop helps to define the variables to add and the data to use as variables as well as the structure of the analytical report to be put in place for an optimum monitoring.
Building the Callbot Knowledge Base
Your use cases, spoken words and variables are now fully defined. You can start building your callbot’s knowledge base!
Integration of Knowledge Articles
At dydu, we’ve decided to let our conversational AI experts integrate the knowledge and not our clients. We configure the decision trees, knowledge articles and spoken phrases in our Bot Management System according to the flowchart approved during the previous phase. In most cases, clients can change their answers autonomously.
We also add sentences and matching sentences (groups of words and expressions that mean the same thing). The aim is to maximise the bot’s understanding.
Voice Design
The next step is to choose the voice of your callbotfrom a catalogue. Depending on the TTS (Text to Speech) provider, the client can also record a voice of their choice.
In some cases, we add SSML (Speech Synthesis Markup Language) tags to the knowledge articles. They allow to specify punctuation and pronunciation in the Text to Speech engine, the speech transcription brick. It is then possible to add pauses, read numbers or letters separately, read dates and times, and provide pronunciation instructions (for example, to read more slowly). It is also possible to manage blanks in the conversation and to automatically prompt the user.
The vocal aspect is important to make conversations as smooth and natural as possible for end-users.
Test & Learn and Release of Callbot
This step involves testing the callbot’s voice on the phone with a limited panel. If possible, the members on this test panel should be quite different (age, gender, accent, etc.) and use a wide range of wording to test the bot’s understanding.
We provide an acceptance test plan with all the items to test. The conversations generated during these tests help adjust the knowledge articles, spoken phrases, variables and tags if necessary. Once the acceptance phase has been completed, the client validates the bot’s release and redirects customer calls to the bot’s number.
Callbot Management and Development
Once the bot has been configured, we will train you to use our software. The aim is to ensure that you can manage your bot (dashboard and analytics), improve its understanding (adding phrases and matching groups) and update its spoken phrases autonomously. You can listen back to user conversations and read them since they are also transcribed into text.
Our teams will provide support for another month after your callbot’s go-live and help you analyse your analytics and past dialogues via weekly meetings. You can also benefit from extended support with a Customer Success Manager if you wish.
For a successful callbot and ROI that meets your expectations, make sure to choose a publisher with real expertise, which provides support tailored to your organisation and training for your staff, to ensure that they fully adopt the solution.
Feel free to contact us if you would a callbot demo.
