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 back. In the previous episode we built an automation map and chose one main project. The next question is: how do you prove that the project really helps? In this episode we measure three things: work hours saved, costs against the value of work hours, and customer response. Then we sum them up in a simple dashboard in Google Sheets, without coding.
Step 1: Three measures of impact
Many businesses feel that AI helps them, but they can't answer when asked: how many hours did it save? What does it cost? Are customers happier? Without numbers, it's hard to decide whether an automation is worth keeping, fixing, or stopping. That's why we measure before and after, in a way simple enough to do while you keep selling. Numbers also help when you talk to other people. Dimas can show the results to Laras, and Dewi can report them to the directors of Maju Bersama Co. to get support for the next project. There are three measures. First, work hours saved: how long the task took before the automation, and how long it takes after. Second, cost versus value. How much you pay for subscriptions and API usage, compared with the value of the work hours you saved. Third, customer response: how quickly chats get answered, how many after-hours chats get a reply, and how many orders are logged correctly. These three measures were chosen on purpose, because anyone can calculate them with Google Sheets. You don't need a complicated analytics app. The pattern is always the same. First, log one week the old way, before the automation runs. If your automation is already running, switch it off for a day, or use old records you still have. Second, run the automation for exactly the same tasks. Third, log another week the new way. Try to make the weeks comparable, for example both ordinary weeks, not a long holiday week. Fourth, compare the results on the dashboard. One important note: all the Sunset Coffee numbers shown in this episode are illustrative examples, not promised results. Your numbers may be bigger or smaller.
Step 2: Work hours: log one week before and after
The easiest way to log time is a short form. Dimas creates a Google Form called Sunset Coffee Work Time Log. The form records the date and time of each entry automatically, so nobody has to fill those in. The first field, done by: who did the work. The second field, the task. It's best as a fixed list of choices that matches the automation map, so there are no differently spelled entries. The third field, time in minutes. For example, this morning Dimas spent ninety-five minutes answering chats about prices and stock. The fourth field, method: manual or bot-assisted. After each block of work, Dimas fills in this form. It takes ten seconds, and every answer flows into Google Sheets automatically. For an office like Dewi's, the admin team can use the same form. Just change the list of tasks, for example summarizing emails or putting together the weekly report. Here's the result in Sheets. In the first week, everything is done by hand. On Monday, Dimas spends ninety-five minutes answering price and stock questions, then seventy minutes logging orders. Tuesday looks similar. Records like these keep coming in for a full week. The following week, the WhatsApp agent starts working. Notice how the tasks change. Dimas now spends twenty-five minutes monitoring and checking the bot, fifteen minutes answering escalated chats, and ten minutes checking the logged orders. This is a key point: time spent monitoring and fixing the bot must be logged too. If it isn't, your savings will look bigger than they really are. Before we look at the summary, here are four tips to keep logging from becoming a burden. First, log per work block, not per chat. Logging every chat would eat up the very time you're trying to save. Second, the task choices in the form are a fixed list, exactly matching the names in the automation map. That way the formulas in Sheets can group them correctly. Third, put the form link on your phone's home screen. One tap and you're filling it in, even between serving customers. Fourth, measure only the tasks in your main project. For Sunset Coffee, that's just answering price and stock questions and logging orders. Other tasks can be measured later. In the Summary tab, add up the minutes per task and convert them to hours. Answering price and stock questions: eight and a half hours before, two hours after. The saved column holds the formula equals B2 minus C2. That means hours before minus hours after. The result is six and a half hours a week. Logging orders: six hours down to an hour and a half, saving four and a half hours. In total, from fourteen and a half hours down to three and a half. That saves eleven hours a week, a sample number, and it already includes the time spent monitoring the bot.
Step 3: Costs: subscriptions and API vs the value of work hours
The second measure is cost. Take the numbers from real bills, not from memory. The first component is the AI app subscription. Note that you may already pay for it for other uses, so only count a fair share of it. The second component is the automation platform. Look at the n8n or Make billing page. Both have free plans, but watch the limits on the number of executions. The third component is AI model API usage, meaning the cost each time the agent thinks. Check the API provider's usage dashboard. The number goes up and down with the number of chats, as we discussed in Session 9. The fourth component is a third-party WhatsApp service, only if you use one. Plan prices change often. That's why this class doesn't quote specific prices. Always check the latest pricing pages and your own bills. Now the value of the work hours. First, calculate the hourly rate: monthly salary divided by the number of working hours per month. For this example, we round Dimas's rate to Rp25,000 an hour. Second, the monthly value equals the hours saved per week, times four weeks, times the rate. Eleven times four times Rp25,000 gives Rp1,100,000 a month. Third, and this often gets forgotten: note what the saved hours are used for. The value only becomes real if Dimas uses the time for production, sales, or serving customers in person. Let's line them up. Say the total cost of all of Sunset Coffee's tools is Rp400,000 a month. This is a sample number, not the price of any particular plan. The value of the forty-four work hours saved, as we just calculated, is Rp1,100,000. The difference is Rp700,000 a month. If the difference is negative, don't panic right away. Maybe the automation has only just started, or a tool could be moved to a cheaper plan. This illustration is just an example, so use your own bills and rates. There are also benefits that are hard to put in rupiah, such as Dimas no longer replying to chats late into the night. Write benefits like this as a note below the dashboard, but don't mix them in with the rupiah numbers.
Step 4: Customer response: reply time and logged orders
The third measure is customer response. Look at this sample chat. The customer asks at 21:48, but Sunset Coffee closes at 21:00. The agent replies within the same minute: the one-liter palm sugar iced coffee is still available, Rp85,000 per bottle, so two bottles come to Rp170,000 plus Rp10,000 for delivery. The customer replies, asking for delivery at nine in the morning. The agent logs the order with a total of Rp180,000, and says the admin will still confirm payment tomorrow morning. These are the three things we measure: reply time, after-hours chats that get answered, and orders that are logged correctly. Where does data like this come from? The good news is, you don't have to log it by hand. In the WhatsApp agent workflow, each incoming chat already carries its arrival time. The agent answers and logs orders as usual. The reply is sent, and the reply time is recorded too. Just add one node at the end: append a row to a sheet called Chat Log. Every conversation automatically leaves a trail you can measure. If you use Make or a third-party WhatsApp service, it works the same way: add a save-to-Sheets step at the end of the flow. All of this is recorded in the Chat Log tab. The columns are: arrival time, reply time, the difference in minutes, whether it was after hours, whether it was answered, and whether the order was logged correctly. The 21:48 chat from before is logged as after hours, answered, with a correct order. There's also a morning chat at 07:40 that took twenty-five minutes, because it was escalated to Dimas. That's normal, and it's still logged as it happened. And most important, also log when the bot gets it wrong. In the last row, the order quantity was logged incorrectly. Records of mistakes like this are material for improvement, not something to be ashamed of. From that log we calculate three numbers. Average reply time: the average of the minutes column. In this example, it drops from thirty-eight minutes to two minutes. After-hours chats answered: the number of night chats answered divided by all night chats. From three out of twelve, to twelve out of twelve. Correctly logged orders: every week, take twenty random orders and match what's in Sheets against the chat. From seventeen correct to nineteen correct. There's still one wrong, and that's the one to fix.
Step 5: A simple dashboard in Google Sheets
Now let's bring it all together in the Dashboard tab. The columns are: the measure, the before number, the after number, and how it's calculated. Remember, every number here is an example. The first row is work hours per week. Click the after cell, and the formula bar shows SUM of the hours column in the Summary tab. SUM means add up all the numbers in that range. The second row is the average reply time. The formula is AVERAGE of the minutes column in the Log tab. AVERAGE means all the numbers are added up, then divided by the number of rows. The third row is after-hours chats answered. The formula is COUNTIFS, which means count the rows that meet every condition: the after hours column says Yes, and the answered column also says Yes. The fourth row, correctly logged orders, works the same way: count the samples marked Yes. The result is nineteen out of twenty. The fifth row, tool costs per month, is simply typed in from the bills. And the sixth row, the value of hours saved: hours saved times four weeks times the hourly rate in another cell. You don't need to memorize these formulas. Ask the AI to write them by naming the tab and the columns. Then check: count a few rows by hand and make sure the results match. The separator inside formulas can be a comma or a semicolon, depending on your Sheets language settings. If you're not used to formulas yet, use this prompt. The context block names the tab and its list of columns, plus the rows where the data sits. The task block says what you want to calculate, for example the number of after-hours chats that were answered. The format block asks for the formula on one line, plus a one-sentence explanation in everyday language, so you understand what's being calculated. And the limits block: use the argument separator your Sheets uses, comma or semicolon, and no scripts or macros. This class is code-free, and regular formulas are more than enough for a dashboard. At the top of the Dashboard tab, show four big numbers like these so it can be read in five seconds. First, work hours down by eleven hours a week. Second, an average reply time of two minutes. Third, twelve out of twelve night chats answered. Fourth, nineteen out of twenty orders logged correctly. Update this dashboard every Monday from the previous week's records. Again, these are illustrative numbers. One tip: put these four numbers in the very top row of the Dashboard tab, with the detailed table below. Laras only has to open one tab to know how this week went.
Step 6: Reading the results and making decisions
Once the numbers are tidy, AI can turn them into a report. Dimas opens Claude and pastes in only the dashboard numbers, with no customer chat content, then asks for a five-sentence report for Laras, plus two recommendations. The result is a short report the owner can read easily: hours saved, reply time, night chats, logged orders, and the value compared with the costs. The recommendations make sense too: trace the one order with the wrong quantity, and make sure the stock in Sheets is updated every morning before opening. Still check the numbers in the AI's report against the real dashboard before sending it to Laras. Dashboard numbers are there to help you decide. There are three possibilities. First, keep: the hours saved are clear, customers are served, and the costs are worth it. Carry on, and measure again every month. Second, fix: it saves time, but some mistakes keep repeating, like wrong order quantities. Fix the agent's instructions or its data, then measure for one more week. Sunset Coffee is in this position. Third, stop or replace: the costs are higher than the value, or customer complaints actually go up. It's fine to go back to the old way, or to pick a simpler tool from the automation map. Decisions like this are actually a sign that you're managing AI maturely. Finally, four traps that happen a lot. The first trap is measuring only after, with no before record. Without a comparison, the after number means nothing. The second trap is comparing an ordinary week with a long holiday week. The number of chats and orders will clearly differ, so the result is misleading. The third trap is forgetting to count the time spent monitoring and fixing the bot. That makes the savings look bigger than they really are. The fourth trap is treating the sample numbers in this video as promised results. Every business is different. The right way: log things as they are, including when the bot gets it wrong. Recorded mistakes are material for improvement in your thirty-day plan. And measure for at least one week, ideally two, so that a day that happens to be busy or quiet doesn't skew the results.
Done!
Let's recap. We measure work hours with a one-week log before and after, compare costs with the value of work hours, measure customer response from the chat log, then sum it all up in a Sheets dashboard with simple formulas. In the next episode, we put together a thirty-day implementation plan: an internal trial, some customers, full rollout, then evaluation, complete with team roles and a backup plan. See you in episode forty-eight.
Practice exercises
Try these on your own after watching to lock in what you learned.
Log the work time for your main project's tasks for one week before the automation runs, using a form or a Sheets tab.
Calculate your monthly tool costs from real bills, then compare them with the value of the work hours saved using the hours × hourly rate formula.
Create a Dashboard tab with before and after measures, complete with sum, average, and conditional count formulas.
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 classes1,279 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.
