Lesson 5. Mini-Spec and Next Steps#
Why This Matters#
Before starting work, it's important to document: what we're doing, why, and what the result will be.
Key Idea#
Mini-spec = goal + inputs + outputs + success metric
You don't need 20 pages. One page with key points is enough.
Mini-Spec Template#
# Mini-Spec: [project name]
## 1. Goal
What do we want to automate and why?
Example:
"Automate answers to standard customer questions to reduce manager workload by 60%."
## 2. Current Situation
How does it work now?
Example:
"Managers answer 30–40 questions/day manually, 5 minutes per answer. Total: 2.5–3 hours/day."
## 3. Agent Task
What should the agent do?
Example:
"Answer questions about business hours, services, prices, delivery. If it doesn't know — hand off to manager."
## 4. Input Data (context)
What data does the agent need?
Example:
- FAQ (20 Q&A pairs)
- Price list
- Business hours and contacts
## 5. Outputs (result)
What do we get at the end?
Example:
- Reply to customer in chat
- Log question in spreadsheet (for analytics)
- Notify manager if it couldn't handle it
## 6. Channels
Where does the agent work?
Example:
- Telegram bot
- Web widget on site (optional)
## 7. Success Metrics
How do we know it's working?
Example:
- Agent handles 60–70% of questions
- Average response time: under 1 minute
- Time saved: 10+ hours/week
## 8. Risks and Fallback
What could go wrong?
Example:
- Agent doesn't understand question → hands off to manager
- Agent makes mistake → customer can contact manager directly
## 9. Timeline
When is launch?
Example:
- Development: 5 days
- Testing: 2 days
- Launch: 1 week from start
## 10. Budget (optional)
How much does it cost?
Example:
- Development: 15,000 RUB
- Subscriptions (Zapier, integrations): 0 RUB/month (free tiers)
Example of a Completed Mini-Spec#
# Mini-Spec: FAQ bot for online clothing store
## 1. Goal
Automate answers to standard customer questions to reduce support workload by 60%.
## 2. Current Situation
Managers answer 40 questions/day, 5 minutes per answer. Total: 3.5 hours/day.
## 3. Agent Task
Answer questions about delivery, payment, returns, sizing. If it doesn't know — hand off to manager.
## 4. Input Data
- FAQ (25 Q&A pairs)
- Delivery and payment terms
- Return policy
## 5. Outputs
- Reply to customer in chat
- Log in spreadsheet for analytics
## 6. Channels
- Telegram bot
- Web widget on site
## 7. Success Metrics
- Agent handles 70% of questions
- Response in 30 seconds
- Savings: 12+ hours/week
## 8. Risks and Fallback
- Agent didn't understand → hands off to manager
- Agent made mistake → customer can call
## 9. Timeline
- Development: 3 days
- Testing: 2 days
- Launch: 5 days from start
## 10. Budget
- Development: 12,000 RUB
- Subscriptions: 0 RUB/month
Client Approval#
- Show the mini-spec to the client
- Clarify:
- is everything correct?
- did we miss anything?
- any changes needed?
- Get confirmation
- Start work
Module Practice#
Assignment 1: Mini-Interview#
Conduct a mini-interview with a client (real or practice):
- Ask 5–7 questions about processes
- Find 3–5 candidate tasks
- Record: what's done, how often, how much time
Assignment 2: Process Map#
Pick one task and create a process map:
- input (trigger)
- steps (sequence)
- output (result)
- bottlenecks
- where to automate
Assignment 3: Task Scoring#
Score 5–7 tasks using the scoring matrix:
- impact (1–3)
- complexity (1–3)
- risk (1–3)
- data (1–3)
- priority
Choose a task with high priority.
Assignment 4: Mini-Spec#
Write a mini-spec for the chosen task:
- goal
- current situation
- agent task
- input data
- outputs
- channels
- success metrics
- risks
- timeline
- budget (optional)
Templates and Artifacts#
Client Brief Questionnaire#
1. Which task takes the most time for the team?
2. How many times per day/week is this task performed?
3. How long does one iteration take?
4. Who performs this task?
5. What data is used?
6. Are there ready documents/knowledge bases?
7. What result do you want?
8. How will you know the task is solved?
Process Map Template#
Process: _______________________
Input (trigger): _______________________
Steps:
1. _______________________
2. _______________________
3. _______________________
Output (result): _______________________
Bottlenecks: _______________________
Where to automate: _______________________
Task Scoring Table#
| Task | Impact | Complexity | Risk | Data | Total | Priority |
|---|---|---|---|---|---|---|
| [task 1] | ||||||
| [task 2] |
Impact Estimation Template#
Task: _______________________
Frequency: _____ times per day/week/month
Time per task: _____ minutes
Hourly rate: _____ RUB
Total time per month: _____ hours
Cost of time: _____ RUB
Automation percentage: _____%
Savings: _____ RUB/month
Review Questions#
- Which processes are considered "suitable" for automation?
- What is task scoring and why is it useful?
- Why are data more important than the idea?
- What questions to ask the client at the start?
- How to estimate impact without exact numbers?
- When is it better to turn down a case?
- What risks are typical for an agent?
- What must be in a mini-spec?
Correct answers:
- Repeatable, standard, with clear rules, performed 10+ times per week
- Evaluating tasks by impact, complexity, risk, and data for prioritization
- Without data the agent can't work, and the idea stays just an idea
- What takes time? How often is it done? Is data available? What result is expected?
- Rough estimate: frequency × time × hourly rate × automation percentage
- When impact is unclear, risks are high, no data, complexity is too high
- Data errors, misunderstanding the request, information leakage
- Goal, current situation, agent task, inputs, outputs, metrics, risks, timeline
Module Summary#
You learned:
- how to find business tasks through observation and interviews
- how to create a process map
- how to evaluate tasks (scoring) and set priorities
- how to choose the best case to start with
- how to write a mini-spec
You received:
- client brief questionnaire
- process map template
- scoring table
- mini-spec template
- impact estimation template
Summary of Modules 1–5:
You've gone from basic understanding of AI agents to building working solutions:
- understood the market and opportunities (Module 1)
- learned to work with AI (Module 2)
- explored tools (Module 3)
- built automations and chatbots in Zapier (Module 4)
- learned to find and evaluate business tasks (Module 5)
Next steps:
- Modules 6–14 deepen your knowledge: IT concepts, prompting, data, API, n8n, security, monetization
- final project: a full agent for a real task
You're ready to earn with AI agents! 🚀
Materials for the Site#
Flashcards#
Flashcards: Finding and Evaluating Tasks1 / 8
Known: 0 (0%)