At most one email a week. Unfollow anytime with one click.
Transcript
The full narration of this video. The current sentence is highlighted while the video plays; click any sentence to jump there.
Show transcript
Introduction
Welcome to Session 10, the final session of Agentic AI Without Coding. We've been through nine sessions together, and now it's time to bring everything together into a real project for your business. In this episode we build an automation map of your business. We list your processes, choose the right tool for each task, then decide what to do first with an impact times ease score.
Step 1: An automation map: one page for the whole business
Over nine sessions, you've tried lots of tools: chatbots, custom assistants, connectors, workflows, agents, and even agent teams. The problem is, lots of tools without a direction can just be confusing. That's why we make an automation map. It answers three questions: which work gets automated, with which tool, and where to start. One Google Sheets page is enough, and that page becomes the foundation for your final project and your thirty-day plan. The map is useful for the team too. Dimas, the Sunset Coffee admin, can see at a glance which tasks the AI will help with, which ones he keeps doing himself, and when the changes begin. The good news is, you're not starting from zero. From Session 1, you already have a list of five repetitive tasks you want to hand over to AI. That list is today's main material. From Sessions 2 and 3, you have a library of prompt templates, plus a customer service assistant that knows your products and prices. From Sessions 4 and 5, you have a market research agent, and experience connecting AI to Gmail, Calendar, and Google Sheets through connectors. From Sessions 6 to 8, you've built workflows in n8n or Make, an admin agent that logs orders, and customer service on WhatsApp or Telegram. And from Session 9, you have a content pipeline with an agent team, plus a security checklist. So the automation map doesn't start from zero. It ties the output of every session into one plan. So nothing gets missed, we split the business into five process areas. The first area is sales: taking orders, checking stock, and confirming payments. The second area is customer service: answering product questions, handling complaints, and giving delivery status updates. The third area is content: finding ideas, writing captions, and scheduling social media posts. The fourth area is admin: email, meeting schedules, documents, and data from forms. Office professionals like Dewi feel this area the most. The fifth area is reports: sales recaps, stock, and weekly performance. These five areas apply to a coffee shop, an online store, a distributor's office, or a tutoring center like the one where Mr. Hendra teaches.
Step 2: Listing repetitive work per area
Let's practice with Laras's Sunset Coffee. Open a new Google Sheets file, name it Automation Map, then make six columns: area, task, frequency, hours per week, done by, and main problem. Start with the sales area. Dimas logs about twenty-five WhatsApp orders a day into a spreadsheet. That's roughly six hours a week, and the problem is frequent typos in the quantities. Dimas also checks payment transfers, about three hours a week. The problem is that the bank statement sometimes gets checked late, so orders are held up. In customer service, answering questions about prices and stock takes eight hours a week. It's the biggest row, and chats that come in at night often go unanswered. Handling complaints happens less often, about three times a week, but it always needs a decision from Laras as the owner. In the content area, Laras writes captions and schedules four Instagram posts a week, about four hours. The problem is that posts often go out late. Then admin: replying to supplier emails, two hours a week. And reports: the weekly sales recap that's still counted by hand, also two hours. Notice that each row starts with a clear verb: log, confirm, answer, handle. Avoid vague rows like customer stuff, because vague work is hard to automate. A few tips to make your list useful. First, each row holds just one task, and always starts with a verb. Second, write the frequency as it really is, per day or per week. Work that happens dozens of times a day is usually a strong candidate for automation. Third, hours per week can be an estimate for now. Don't stop just because you don't know the exact number. In Episode 47 we measure it with real records for one week. Fourth, note who does it. This matters for the plan later, because the person doing the work now usually becomes the person who monitors its automation. Fifth, the main problem column. This is the most important column, because automation is built to solve problems, not just because the tool is impressive.
Step 3: Choosing the right tool from Sessions 1–9
The next column on the map is the tool. Remember the principle from Episode 1: choose the simplest way that already solves the problem. The first rung on the ladder is a chatbot with a prompt template, for one-off writing where you pick the result yourself. The second rung is a custom assistant from Session 3. It fits when answers must be consistent and based on your business's SOPs, catalog, or FAQ. Third, connectors from Session 5. The AI reads Gmail, Calendar, or Sheets, but still on your command, in the chat window. Fourth, a workflow in n8n or Make from Session 6. It fits when the steps are always the same and set off by an event, such as a form submission or a schedule every Monday. Fifth, agents with tools, or agent teams from Sessions 7 to 9. These are for work that needs to make its own decisions and use several apps at once. One more: web research and browser agents from Session 4 suit recurring research tasks, like monitoring competitor prices every month. Now let's fill in the tool column for Sunset Coffee. Logging WhatsApp orders goes to an agent with tools, exactly the admin agent you built in Session 7. Confirming transfers is deliberately marked stays human. Work that involves money is safer when a person checks it. A good map also records what doesn't get automated. Answering price and stock questions uses the WhatsApp agent from Session 8, with escalation to Dimas. Complaints just need an assistant that drafts replies, while the decision stays with Laras. Instagram captions and scheduling use the multi-agent pipeline from Session 9, with Laras's approval before posting. Supplier emails just need the Gmail connector to prepare drafts, not to send on its own. And the weekly recap uses a workflow with an AI step that runs every Monday morning. Seven tasks, five kinds of tools, and one task deliberately left manual.
Step 4: Setting priorities: impact × ease
The list is complete, and the tools are chosen. The question is, where do you start? We use two scores from one to five. An impact score of five means it saves many hours, and customers feel the benefit right away. An ease score of five means the data is already tidy and you've built the tool before. A score of three means moderate: it saves a few hours, or it needs a bit of data cleanup and learning a new tool. An impact score of one means it saves little or rarely happens. An ease score of one means the data is messy, or the risk is high if the AI gets it wrong. Why count ease at all? Because the final project has a deadline. A project with big impact that needs three months of preparation won't get done within a thirty-day plan. Open a new tab called Scores. The columns are: task, impact, ease, score, and rank. Fill in impact and ease based on your own judgment. Look at the formula bar above the table: equals B2 times C2. It means the value in the impact column is multiplied by the value in the ease column. Drag that formula down, and every row is calculated automatically. Answering price and stock questions gets five times four, which is twenty. Logging orders also scores twenty. Both use the same Sheets data, so later they can be combined into one agent. The weekly recap scores fifteen. Its impact is moderate, but it's very easy, because the data is already in the orders sheet. Then sort from the highest score to the lowest using the sort range menu. Menu names may differ slightly because the app is updated often. Instagram captions come fourth, supplier emails fifth, complaints sixth. Confirming transfers comes last with a score of three, because its ease is one: the risk is too high if it goes wrong. This number matches our decision to leave it manual. Score results usually fall into four groups. First, do first: high impact and easy. These are your final project candidates. Second, big projects: high impact but hard. Don't throw them out. Break them into small stages and put them in next month's plan. Third, quick wins: moderate impact but very easy. Do them along the way, because small results that show up fast build the team's trust in automation. Fourth, put on hold: low impact and hard. Leave them for now. Focus is the key to a successful thirty-day project. For Sunset Coffee, answering price and stock questions and logging orders go into do first. The weekly recap is a quick win. Confirming transfers goes on hold, and it's intentionally left manual.
Step 5: The Sunset Coffee sample map
Before deciding, ask the AI for a second opinion. We use the prompt structure from Session 2. The role block: the AI is asked to be a practical, careful automation consultant for small businesses. The context block holds variables you fill in yourself: the business name, the number of people on the team, and the tools you've learned during this class. The data block holds your automation map table. Notice the note: no customer names, numbers, or addresses. Personal data isn't needed, and it shouldn't be pasted in. The task block: review the tool choices and the priority order, then mark the tasks that are better kept manual. The format block asks for short notes per task, ending with the three biggest risks. And the limits block: don't suggest tools I haven't mentioned, and say you're unsure when you are. This limit stops the AI from making up solutions that sound great but that you can't actually run. Laras uses that prompt in the Sunset Coffee project in Claude, then pastes in the list of seven tasks with their tools and scores. Once again, no customer data is pasted in. Claude agrees with the order and adds valuable notes. Answering prices and logging orders should be combined into one WhatsApp agent, because both use the same spreadsheet. Claude also points out the biggest risk: if the stock in Sheets isn't updated, the agent will give wrong answers. That's an important note for the thirty-day plan later. Then Laras asks the question that confuses participants most often: if I can only manage one project in thirty days, which one should I pick? The answer is clear: the WhatsApp agent for prices, stock, and order logging, about fourteen hours a week. The weekly recap follows as a quick win. Still remember, the AI gives advice, but the final decision stays with Laras. This is what the Sunset Coffee main project looks like when drawn as a draft in n8n. The trigger is a WhatsApp chat from a customer. The chat goes to an AI Agent node holding the Sunset Coffee work instructions, and the answer is sent back through WhatsApp. Below it, the agent uses an AI model as its brain, a stock sheet tool to read prices and inventory, and an orders sheet tool to log orders. And an escalation tool to notify Dimas when there's a complaint or a question it can't answer. You've already built every one of these parts in Sessions 7 and 8. The final project just ties them together and polishes them. Let's sum up the Sunset Coffee map. There are seven tasks listed across five process areas. Together, they add up to about twenty-six work hours a week. Remember, that's still Dimas and Laras's estimate. And the final result is one main project plus one quick win. We'll measure those weekly hours with real records in the next episode, so the impact can be proven, not just felt.
Step 6: Dewi's sample map and your own
The same approach works in an office. This is Dewi's map. She's the operations manager at Maju Bersama Co., an office supplies distributor. In first place: summarizing about sixty emails a day with the Gmail connector. Impact four, ease five. Second place, the weekly sales report, with an agent that reads Sheets. Its impact is the highest, five, because the directors wait for it every Monday. Its ease is three, because the sales data needs cleaning up first. Third place, scheduling team meetings with the Calendar connector, scoring twelve. Fourth place, order data from customer forms with a workflow and an AI step, also twelve. And last, competitor price research with a research agent. Useful, but only needed once a month, so it scores six. Interestingly, Dewi picks the weekly report as her main project, not the email summaries. The email summaries are enough as a quick win, because she can use them right away without building anything. The score helps you decide, but you still use common sense. Now it's your turn. First, create an Automation Map in Google Sheets with the six columns we just covered. Second, fill in at least seven tasks from the five process areas. Start with your list of five tasks from Session 1, then add any new ones that come to mind. Third, choose the simplest tool for each task. Don't hesitate to write stays human for work that involves money or sensitive decisions. Fourth, give impact and ease scores, calculate them with the multiplication formula, then sort from the highest. Fifth, mark one main project and one quick win. The main project is the one you'll polish and present at the end of the class. Keep this map safe. In episode forty-seven we add impact numbers to it, and in episode forty-eight this map turns into a thirty-day implementation plan.
Done!
Let's recap. An automation map is built in three moves: list the repetitive work across five areas, choose the simplest tool for each, then rank them with an impact times ease score. In the next episode we prove the impact with numbers: work hours saved, costs, and customer response, along with a simple dashboard in Google Sheets. See you in episode forty-seven.
Practice exercises
Try these on your own after watching to lock in what you learned.
Complete your Automation Map with at least 7 tasks from the five areas: sales, customer service, content, admin, and reports.
For each task, choose the simplest tool that solves the problem, and mark the tasks that are better left to humans.
Give impact and ease scores of 1–5, calculate the score with a multiplication formula, sort, then mark one main project and one quick win.
Ask the class material
Type a question or keyword; we find the part of this episode that explains it and jump the video there.
My notes
Notes are saved with the current video time and are only visible to you.
Questions & answers
Stuck on a step? Ask here; the instructor and other learners can answer.
No questions yet.
Your instructor
Cerdas Inovasi Digital Team
Software developers, BIM & 3D engineers, and AI practitioners
53 video classes1,278 episodes
We build software, BIM models, games and AI automations for clients every day, and turn that hands-on work into step-by-step classes: every step is shown on screen and comes with a PDF guide you can keep.
