Launching an MVP with a chatbot.
The Context
This document is not about how we improved the conversion rate of a complex transactional process using no-code solutions. Rather, it’s a practical guide to identifying the right problems and designing the appropriate processes.
The User’s Problem:
In Spain, many users wished to review their past tax submissions. Around 30% of these users, who had never used a tax advisor, might be eligible for refunds. Their main goal was to reclaim their money.
Our Problem:
We faced a challenge where users were hesitant to pay upfront, especially if they were uncertain about receiving a refund. Our process, based on a “No win-No fee” model, was too manual to scale efficiently. We needed a method to filter users to balance manual work with return on investment.
Our Solution:
We implemented a conversational chatbot equipped with a tailored knowledge base. It asked users specific questions in a predefined order. Payment was processed via Stripe, and registration occurred on Typeform.
For a quick summary, here is our 30-second elevator pitch:
- Start with a specific problem and avoid overcomplicating solutions.
- List all potential winning hypotheses.
- Establish metrics for deciding whether to proceed with or abandon the experiment.
- We recommend using Chatbase for the chatbot, which is enhanced with integrations and legal compliance.
- Regularly review user conversations for insights and evolution.
- Keep an eye on Zapier, as it’s crucial for scalability.
- Develop a coherent narrative for the user, encompassing everything from advertising to emails.
For those interested in understanding the underlying rationale:
Start with Why: (Be Practical)
Everything begins with the user’s needs, which might range from eligibility queries to support requests. Being practical involves focusing on problem-solution frameworks like Jobs to Be Done, which helps to identify users based on their problems and fosters a more specific focus.
For more on this, check out Christian Christiansen’s Jobs to Be Done Milkshake story here.
How to Choose:
Create a list of customer problems, possibly using a Miro flow. Concentrate on the most common or financially significant issues. Consult other teams for different perspectives on the problem.
Experiment Planning
After pinpointing the problem, devise a flow from raising user awareness to gathering post-experience reviews. The aim isn’t to build something complex, but to define an easily scalable process. Establish clear criteria for continuing or discontinuing the experiment and involve all relevant team members in a shared vision.
Our hypotheses included:
- User engagement with the chatbot.
- The accuracy and relevance of chatbot responses.
- The users’ willingness to pay.
- User compliance with a clearly defined Typeform process.
- The marketing team’s ability to segment users based on this information.
Understanding these hypotheses helps shape your process design.
Building Blocks
Delivering a conversational experience is relatively straightforward with tools like Open AI, but understanding user interactions and GDPR compliance requires a more nuanced approach.
Top 3 Solutions and the Future Winner
ChatGPT:
Rapid for testing conversational flows but limits learning due to its closed platform structure.
SiteGPT: Link
Features a simple interface and effectively engages users. However, it lacks certain advanced features and may not sufficiently engage all users.
ChatBase: Link
Our chosen solution for its simplicity, effective integration capabilities, and legal compliance, especially with third-party tools like WhatsApp and Messenger.
Zapier:
Though not currently meeting all our needs, Zapier’s potential for connecting various tools makes it a promising candidate for future scalability.
Marketing and Communication
Let’s start with an example: Phrases like “Hey, I want you to pay for a service delivered by a chatbot” or “Hey, I want you to understand that engaging with the chatbot could lead to human interaction” are not very appealing and don’t align with our brand voice. However, we don’t need to build our narrative this way.
Our approach embraces the concept of exponentiality under the 4Ds: Digitalize a service, democratize access, demonetize the product, and disrupt the market. In this context, what and when we communicate is crucial.
We differentiated between two components of the experience:
- AI interaction driver: Eligibility, accessibility, expertise, and rapid response.
- Payment wall: Tangible service delivered by a human.
The challenge was engaging users with the experience. We wanted users to actively seek and interact with it. A tip here is to use tools like Hotjar to track page scroll and user attention.
Engaging users with the chatbot involves:
- Communicating through banners, visuals, and storytelling. Messages like “Discover if you are eligible for XXXX for free with our chatbot” or “24/7 free tax support” are vital for setting the right expectations. It’s important to attract users who are genuinely interested and will provide valuable feedback.
- Explaining the rationale: It’s crucial to clarify how users can participate in the solution, detailing the process, outcome, and the paywall.
Channeling the conversation is like starting a conversation with a stranger; giving hints and quick prompts can be very useful.
Value Capture & Payment Flow:
Including payment from day one might not be ideal, but it becomes essential around day seven of the experiment. Moving the payment and registration step to Typeform simplifies the process and allows for integration and analytics tracking. However, it also means users may need to re-enter information, which can degrade the experience.
We addressed this by integrating with Stripe:
- Inform eligible users to click on a URL.
- The URL includes a tracking link and redirects to a Stripe payment page, then returns with the payment ID.
- After payment, users are asked to add the ID to the conversation.
- Once the ID is entered, the conversation prompts for human interaction.
- The tax operations team then has all the necessary information for follow-up.
Discoveries Along the Way:
- User demographics vary significantly.
- The imagery used in chatbot conversations impacts conversion rates.
- Continuously analyzing conversations is essential for evolution.
- The process is more straightforward than anticipated.
- This is just the beginning of our journey.