AI Agents
Module 07Lesson 6

Lesson 6. Example Ready-Made TS Documents

Hands-on: Zapier

Lesson 6. Example Ready-Made TS Documents#

Example 1: FAQ Bot for a Beauty Salon#

TECHNICAL SPECIFICATION

Project name: AI agent FAQ for beauty salon "Beauty Spot"
Client: M.A. Ivanova, sole proprietor (salon "Beauty Spot")
Date: 02/01/2026
Author: Petrov I., AI consultant

---

GOAL AND TASKS

Goal:
Automate answers to common customer questions, reduce administrator workload.

Tasks:
- Answer questions about services, prices, hours, address
- Help customers book services
- Collect booking data (name, phone, service, date, time)
- Record data in Google Sheets

Expected result:
- 70% of common questions handled automatically
- Administrator freed from routine questions
- All requests collected in one table

---

TARGET AUDIENCE

- Salon clients (women 25–50)
- Level: from beginners to regular clients
- Expectations: quick answers, convenient booking

---

SCENARIO

1. Customer messages the bot on Telegram ("Hello, how much does a haircut cost?")
2. Agent reads the knowledge base and answers ("Women's haircut — from 1500 rub., men's — from 800 rub. Would you like to book?")
3. If the customer wants to book, the agent asks:
   - What's your name?
   - Your phone number?
   - Which service?
   - Which date?
   - What time?
4. Agent confirms the data
5. Agent records in Google Sheets
6. Agent says: "Done! I've booked you. See you soon!"

---

PROMPT

You are the assistant at beauty salon "Beauty Spot".

Your task: answer common customer questions and help book services.

Rules:
- Use only information from the knowledge base
- If you don't know the answer, say: "I'll check with the administrator, message us @salon_telegram"
- Tone: friendly and polite
- Keep answers brief: 2–3 sentences
- If the customer wants to book, collect: name, phone, service, date, time
- After collecting all data, repeat for confirmation
- If the customer confirmed, record data via Google Sheets

Format:
- Use emojis for friendliness (1–2 per message)
- Use a list when there are multiple options

Limitations:
- Don't make up prices, services, or promotions
- Don't promise what you can't guarantee

---

KNOWLEDGE BASE

FAQ (text base):

Question: What services do you offer?
Answer: Haircut, coloring, manicure, pedicure, facial massage.

Question: How much does a women's haircut cost?
Answer: Women's haircut — from 1500 rub., shoulder length. If longer — from 2000 rub.

Question: What are your hours?
Answer: Mon–Fri: 10:00–20:00, Sat–Sun: 11:00–19:00

Question: Where are you located?
Answer: 10 Lenin St., 2nd floor. Entrance from the courtyard.

(10–15 more questions)

---

INTEGRATIONS

1. Telegram Bot API
2. Google Sheets (spreadsheet "Beauty Spot Bookings")
   - Columns: Date, Name, Phone, Service, Booking Date, Booking Time

---

METRICS

- Number of requests processed per day
- Share of answers without administrator involvement
- Number of bookings through the bot

Target KPIs:
- 70% of questions handled automatically
- 10+ bookings per week through the bot

---

TIMELINE

1. Prompt and knowledge base development (1 day)
2. Create bot and configure in Zapier (1 day)
3. Testing (1 day)
4. Launch in test mode (3 days)
5. Full launch (day 7)

Module Practice: Writing a TS#

Assignment: Write a TS for One of the Scenarios#

Choose one scenario:

  1. FAQ bot for an online store

    • Answers questions about products, delivery, payment
    • Helps find products
    • Escalates complex questions to support

  2. Lead qualification agent for a real estate agency

    • Asks questions: what are you looking for (apartment/house), budget, area, timeline
    • Evaluates the lead
    • Hands off to the agent

  3. Course enrollment agent

    • Collects: name, phone, which course, when you want to start
    • Records in Google Sheets
    • Sends notification to the manager

What to do:

  1. Fill in all TS sections (using the template from Lesson 5)
  2. Write the prompt
  3. Describe the agent workflow scenario (step by step)
  4. Specify integrations

Bonus Assignment: Zapier Practice — Improve the Agent's Prompt#

Task:
Take one of the agents created in Module 4 and improve its prompt using what you learned in this module.

What to do:

  1. Open the chatbot or agent in Zapier (FAQ bot or automation from Module 4)

  2. Analyze the current prompt:

    • Are all elements present? (role, task, context, rules, format)
    • What's missing?

  3. Improve the prompt using the checklist:

    • Role is clearly defined
    • Task is clear
    • Context is sufficient
    • Rules are complete (information source, tone, clarifications, actions)
    • Response format is described
    • Limitations are set (what not to do)
    • Examples added (optional)

  4. Add rules from Lesson 3:

    • Source rule: "Use only the knowledge base"
    • Tone rule: "Tone: friendly but not overly familiar"
    • Format rule: "Keep answers brief: 2–3 sentences"
    • Clarification rule: "If the customer didn't specify [parameter], be sure to ask"
    • Limitation: "Don't make up prices, services, or promotions"

  5. Create a test set of queries (10–15 queries from Lesson 4 categories):

    • Simple questions
    • Questions without answers in the base
    • Vague requests
    • Emotional requests
    • Actions
    • Off-topic questions

  6. Test the improved prompt:

    • Run all test queries
    • Compare response quality (before vs after)
    • Record the results

  7. Iterate:

    • If quality improved → save the new prompt
    • If there are issues → improve again and retest

Checklist:

  • Prompt analyzed (missing elements identified)
  • Prompt improved (rules and limitations added)
  • Test set created (10–15 queries)
  • Testing completed (all queries run)
  • Quality compared (before vs after)
  • Iterations done (if needed)

What you'll get:
An improved agent with a professional prompt. Hands-on experience with iterative prompt improvement.

Time: 30–40 minutes

Templates and Artifacts#

Universal Prompt Template#

Role:
You are [role].

Task:
Your task: [what to do].

Context:
[Information about the business, services, terms.]

Rules:
- [rule 1]
- [rule 2]
- [rule 3]

Response Format:
- [how to respond]

Examples:
[Example 1]
[Example 2]

Limitations:
- [what not to do]

TS Template for AI Agent#

TECHNICAL SPECIFICATION

Project name:
Client:
Date:
Author:

---

GOAL AND TASKS

Goal:
Tasks:
Expected result:

---

TARGET AUDIENCE

---

SCENARIO

---

PROMPT

---

KNOWLEDGE BASE (if needed)

---

INTEGRATIONS AND TOOLS

---

METRICS AND KPIs

---

TIMELINE AND STAGES

---

RISKS AND LIMITATIONS

Good Prompt Checklist#

  • Role defined (who is the agent)
  • Task defined (what to do)
  • Context defined (what to know)
  • Rules defined (how to work)
  • Response format defined (how to respond)
  • Limitations defined (what not to do)
  • Examples (optional but recommended)
  • Prompt tested on 10+ queries
  • Agent doesn't make up information
  • Tone is appropriate

Review Questions#

  1. What elements should be in a good prompt?
  2. Why is the role needed in a prompt?
  3. How does an informational agent differ from a qualification agent?
  4. What are rules and limitations in a prompt?
  5. How do you test a prompt?
  6. What should be in a technical specification for an AI agent?
  7. Why is the agent workflow scenario needed in a TS?
  8. How do you improve a prompt if the agent makes up information?

Correct answers:

  1. Role, task, context, rules, response format, examples (optional), limitations.
  2. The role sets the context and style of the agent (e.g., friendly assistant or professional consultant).
  3. An informational agent answers questions (FAQ); a qualification agent asks questions and evaluates the lead.
  4. Rules — what is allowed and required. Limitations — what is not allowed.
  5. Create a set of test queries (10–20), run them through the agent, evaluate responses, improve the prompt, repeat.
  6. Goal, tasks, target audience, scenario, prompt, knowledge base (if needed), integrations, metrics, timeline, risks.
  7. The scenario shows step by step what happens when the agent works. Helps understand the logic and test it.
  8. Strengthen the limitation: "Use ONLY the knowledge base. If the answer isn't there, say: 'I didn't find that information.'"

Module Summary#

You've learned:

  • the structure of an ideal prompt (role, task, rules, format, examples)
  • prompt types for different tasks (FAQ, qualification, consultation, data collection, support)
  • how to set rules and limitations
  • how to test and improve prompts
  • how to write a technical specification for an AI agent

You can now:

  • write a clear prompt for any task
  • test a prompt and improve it
  • create a complete TS for a project
  • hand off a TS to a client or developer

Next step: Module 8 — how to work with data and storage (Google Sheets, Airtable, Notion).

Check Your Knowledge#

Quiz: Prompting and TS#

Flashcards for Memorization#

Materials for the Site#

Knowledge Check: Prompts and Technical SpecificationQuestion 1 of 5

What structure is considered ideal for a prompt?

Key Prompting Terms1 / 8
Known: 0 (0%)
Question

Prompt

👆 Click to flip

Answer

Instruction for an AI model. Technical specification: who you are, what you do, how you respond. Structure: Role → Task → Data → Rules.

👆 Click to flip back