Lesson 6. Testing and Improvement#
Why This Matters#
The first version of an automation or chatbot is almost never perfect. It's important to test, gather feedback, and improve.
Key Idea#
Iterations: launch → test → fix → repeat
Better to launch a simple version quickly than wait for the «perfect» automation.
Test Scenarios for the Chatbot#
Create 10–15 typical requests:
Category 1: Simple questions
- «How much does a haircut cost?»
- «What are your hours?»
- «Where are you located?»
Category 2: Questions not in the knowledge base
- «When do you have discounts?» (if there are no promotions)
- «How much does a massage cost?» (if you don't offer it)
Category 3: Emotional requests
- «I'm unhappy with the service quality!»
- «You're scammers, refund my money!»
Category 4: Vague requests
- «I want to come» (didn't say for what)
- «Book me for tomorrow» (didn't say time)
Category 5: Actions
- «I'd like to book a manicure»
- «Cancel my appointment»
Testing Zaps#
- Use test data (not real!)
- Check each step: is data passed correctly?
- Check edge cases: what if a field is empty? what if the text is long?
- Check Task History — log of all completed tasks
Quality Criteria#
✅ Good:
- answer is correct and complete
- tone is polite
- doesn't make up data
- asks for clarification when needed
- action completed correctly
❌ Bad:
- answer is wrong or incomplete
- rude tone
- hallucination (made things up)
- didn't ask for clarification
- action failed or completed with errors
Common Problems and Solutions#
Problem 1: Chatbot can't find the answer in the knowledge base
Solution:
- Rephrase the Q&A in the knowledge base
- Add synonyms
- Increase the amount of text in the base
Problem 2: Chatbot makes up data
Solution:
- Strengthen the limitation in the prompt: «Use ONLY the knowledge base. If the answer isn't there, say: "I couldn't find that information, I'll check with the administrator"»
Problem 3: Zap doesn't trigger
Solution:
- Check that the Zap is on (Published)
- Check account connections
- Look at Task History for errors
Problem 4: Data is passed incorrectly
Solution:
- Double-check field mapping
- Make sure table columns are labeled
- Test the trigger again
Gathering Feedback#
First week: test yourself and with the team
Second week: launch for customers
Every week:
- review Task History (Zap logs)
- review chatbot conversations
- look for errors and patterns
- improve the prompt and knowledge base
Metrics:
- how many tasks completed
- how many errors in Task History
- common unanswered questions → add to knowledge base
Module Practice: Creating a Chatbot and Automations#
Assignment 1: FAQ Chatbot for a Service#
Task: create an AI chatbot that answers common questions
Steps:
- Choose a niche (salon, store, agency, school, etc.)
- Gather FAQ: 10–20 common questions and answers
- Create a chatbot in Zapier Chatbots
- Write the prompt (role + task + rules)
- Upload the knowledge base (FAQ text)
- Test with 10 requests
- Embed on a website (get the widget code)
- Share the link (for your portfolio)
Example FAQ:
Question: What services do you offer?
Answer: We do website development, Yandex and Google advertising, SMM, and design.
Question: How much does a website cost?
Answer: Business card site — from 30,000 rubles, online store — from 80,000 rubles. Exact price depends on requirements.
Question: How do I contact you?
Answer: Message us on Telegram @agency or email hello@agency.ru.
... (7–17 more questions)
Assignment 2: Lead Qualification Automation#
Task: create a Zap that processes submissions and qualifies leads
Steps:
- Create a form in Zapier Interfaces (fields: name, email, task, budget, timeline)
- Create a table in Zapier Tables (columns: name, email, task, budget, timeline, score)
- Create a Zap:
- Trigger: new form response (Interfaces)
- Action 1: write to table (Tables)
- Action 2: send notification (email or Telegram)
- Test with 5–7 test submissions
- Publish the form (share the link)
Bonus: connect a Zapier Agent for automatic lead scoring:
You are a sales assistant for a digital agency.
Score the lead based on the client's answers:
- Hot: budget 200k+, urgent
- Warm: budget 50–200k, within a month
- Cold: everything else
Record the score in the table.
Useful Resources#
Zapier Video Tutorials#
- COMPLETE ZAPIER GUIDE: Tutorials, Tips and TRICKS for PROFESSIONALS
- Automate work with GPTs and Zapier: ChatGPT finds news for you
Official Resources#
- Getting Started with Zapier
- Zapier Agents Guide
- Zapier Chatbots Guide
- Zapier Tables Guide
- Zapier Interfaces Guide
- Zapier Copilot Guide
- Zapier App Directory (7,000+)
Templates and Artifacts#
Chatbot Prompt Template#
You are [role, e.g.: "beauty salon assistant"].
Your task: [what to do, e.g.: "answer questions and help with booking"].
Rules:
- [rule 1]
- [rule 2]
- [rule 3]
Response format:
- [how to respond]
Limitations:
- [what not to do]
FAQ Base Template#
Question: [question 1]
Answer: [answer 1]
Question: [question 2]
Answer: [answer 2]
... (10–20 pairs)
Testing Checklist#
- Simple questions from the base — does it answer correctly?
- Questions NOT in the base — does it say «I don't know»?
- Similar questions (different wording) — does it find the answer?
- Vague requests — does it ask for clarification?
- Emotional requests — is it polite?
- Actions (booking, sending) — does it perform them?
- Doesn't make up data?
- Appropriate tone?
- Do Zaps trigger correctly?
- Is data passed correctly?
Review Questions#
- What is a Zap and what does it consist of?
- How does a Zapier Agent differ from a Zapier Chatbot?
- What is field mapping and why is it important?
- How do you add a knowledge base to a chatbot?
- Which actions are considered safe for a first automation?
- How do you test a Zap without real data?
- How do you know if an automation is effective?
- What is Task History and what is it for?
Answers:
- A Zap is an automated workflow. It consists of a trigger (start event) and one or more actions
- A Chatbot communicates with customers through a chat widget. An Agent works in the background, performing tasks on a schedule or event
- Mapping is specifying which data from the trigger to pass to the action. Without mapping, Zapier doesn't know what to write where
- In chatbot settings, upload documents (PDF, TXT) or paste FAQ text
- Writing data to a table, sending notifications, preparing reply drafts
- Use test data (fictional names and emails), check Task History
- By metrics: time saved, fewer errors, conversion growth, number of tasks processed
- Task History is the log of all completed Zap tasks. It shows successful and failed runs, errors, and data passed
Module Summary#
You learned:
- what Zapier is and its main products (Zaps, Tables, Interfaces, Agents, Chatbots)
- how to create Zaps (trigger → action)
- how to create AI chatbots with prompts and knowledge bases
- how to use Zapier Agents for background tasks
- how to work with Tables and Interfaces
- how to test and improve automations
You created:
- an FAQ chatbot
- a lead processing automation
Next step: Module 5 — how to find business tasks and choose the best use cases.
Check Your Knowledge#
Quiz: Zapier and Creating Automations#
Flashcards#
Materials for the Site#
Knowledge Check: Zapier PlatformQuestion 1 of 5
What is Zapier?
Key Zapier Terms1 / 10
Known: 0 (0%)