Video version: about 14 min
Introduction and Concepts
About 5 minutesLin Yue's Dilemma
She has completed the first four lessons. She knows how to use AI and write scripts, but she doesn't know which task to tackle first.
She wants to delegate everything to AI: meeting scheduling, notifications, purchase requests, and meeting minutes.
Department Manager Zheng passed by and asked, 'Is what you're doing actually worth it?'
Lin Yue couldn't answer.
Principles for Selecting Tasks: Specific, Profound, Real
Review Your Weekly Work
About 8 minutesAfter reviewing, categorize tasks into three types
Repetitive
Done weekly, with similar procedures
Time-consuming
Takes more than 30 minutes per instance
Clear rules
You can clearly define what 'done correctly' looks like
Example: Lin Yue's weekly review (illustrative)
Let the AI turn your log into a candidate list
Suggested timeHere is my work log for one week. It only includes task names and approximate time spent, with no sensitive information. Please act as a senior administrative manager with twenty years of experience and help me with three things: 1. Merge similar tasks and create a table showing the 'active name, weekly frequency, and approximate duration per occurrence.' 2. Identify which tasks are 'repetitive, time-consuming, and have clear rules,' and explain your reasoning. 3. List any small tasks you think I might have missed or overlooked (just a reminder; do not invent details). Do not make choices for me. Finally, list your questions for me to answer. [Paste your weekly log here]
Practice A-1: Circle candidate tasks
Five-Point Scoring
About 15 minutesFive-point scoring: The higher the score, the more suitable the task is to tackle first
Frequency
Five points for daily or multiple times per week; three points for once per week; one point for once per quarter
Time consumption
Five points for over one hour; three points for about half an hour; one point for under ten minutes
Clarity of rules
Five points for clear standards with examples; three points for roughly defined standards; one point for purely subjective judgment
Cost of error
The lighter the consequences, the higher the score. Tasks involving money, contracts, clients, personal safety, or irrecoverable outcomes receive one point.
Availability of materials
Five points for readily available and easily anonymized data; three points for requiring a search; one point for scattered data or data that cannot be shared with AI
Scoring guideline: Tasks with a high cost of error should be set aside first
Example: Lin Yue's 'Department Weekly Meeting Minutes' Score (Illustrative)
Score Reviewer: It only asks follow-up questions, it does not assign scores for you
Suggested timeI am scoring a task to determine whether it is worth assigning to AI for assistance. The five criteria are: frequency, time required, rule clarity, cost of error (higher score for lower consequences, lower score for higher consequences), and data availability. Each criterion is scored from 1 to 5. Here is my score and reasoning. Please act as a critical internal auditor and point out for each criterion: whether my score might be too high or too low, and whether there are any ambiguities in my reasoning. Ask at most two follow-up questions per criterion. Do not change my scores, do not provide your own scores, and output only a list of follow-up questions at the end. Task Name: [Enter] My Score and Reasoning: [Enter scores and one-sentence reasoning for each of the five criteria]
Practice A-2: He Lei's four tasks, with reference scores
Important, but put aside for now
C8 Duchuan Legal Compliance: Review of open-source components and client confidentiality clauses
- Total 11 points
- Conclusions can only come from the legal head; original contract texts are red-line materials
How to handle it
Not that it's never done. Wait until the first project with lighter consequences runs smoothly, then do a version that 'only organizes, does not conclude.'
- Only group and excerpt clauses, listing what to ask legal about
- Do not write 'usable' or 'not usable'
Interview Colleagues
About 6 minutesInterview colleagues: Ask about the current state, not whether to use AI
Suggested timeOpening: I'm organizing my current tasks and want to understand how you actually do them. This isn't an inspection, and names won't be recorded. Can I take ten minutes of your time? Question 1: When was the last time you did [this task]? Please walk me through it from start to finish, step by step. Question 2: Which step takes the most time or frustrates you the most? Roughly how much? Question 3: What errors have occurred with this task? How were they discovered, and how were they remedied? Question 4: Who receives your final output? How do they judge that 'the task is done'? Question 5: If this task could take half the time each time, what would you do with the saved time?
Colleague A's interview
After the meeting, I wait for the minutes, usually receiving them by the next afternoon.
I'm most afraid when action items don't list my name or have no deadlines, forcing me to ask people again.
Last month, I assigned a colleague B's action item to colleague C, and it wasn't discovered until a week later.
The department manager reviews the minutes mainly to see what was decided, who is responsible, and when it will be completed.
If I have the minutes for the day, I can schedule next week's work on the same day.
Value Proposition, Baseline, and Two Lines
About 10 minutesOne-sentence value proposition: Has an object, has numbers, has a supplement
Suggested timeI help [who], on [what task], shorten [how long] to [how long], and [a quality or time supplement].
A good value proposition has three characteristics.
Identifies the audience.
Who is being helped? Department colleagues, not everyone in general.
Includes specific numbers.
From how long to how long. These numbers are targets now; once you have real data in Lesson 9, you can revise them.
Includes additional benefits.
Beyond time savings, what else improves? For example, same-day distribution, fewer omissions, or consistent formatting.
Baseline: Use one standard for everything.
Lin Yue's Baseline (Illustrative)
Success and Stop-Loss Lines: Written before the trial run, and not changed afterward.
Success Line (all three criteria must be met for success)
- Average time not exceeding 45 minutes (half of the 90-minute baseline)
- Average missed items not exceeding 1.0, and lower than the baseline of 2.1
- Zero red-line incidents
Stop-Loss Line (stop immediately if any one criterion is triggered)
- One red-line incident: stop immediately, report first, then review
- Two consecutive trial runs taking 90 minutes or more
- After four cumulative trials, the average time still exceeds 72 minutes (80% of the baseline)
Pre-Work Communication Package
About 6 minutesPre-Work: Project initiation message to your direct supervisor
Suggested time[Salutation], I plan to run a two-week pilot on [this task], using AI to draft the initial version, which I will personally verify line by line before sending. The materials used are [type of materials], which will be desensitized according to company policy before being sent to the AI; the pilot is expected to take [two weeks], and I will record the time and error rates before and after. I need your confirmation on two points: first, whether this task can be piloted this way; second, are there any contents you require me not to process with AI?
Release Record: If there is no reply that can be retained, do not start the work
Quiz and Assignments
About 5 minutesThree quick questions to check yourself
Question 1: Quiz Question 1: In the five-point scoring, what does a score of 1 for 'Cost of Error' mean?
Question 2: Quiz Question 2: When interviewing a colleague, which of the following questions is appropriate?
Question 3: Quiz Question 3: When should you write down the success criteria and the stop-loss threshold?
Assignment: Project Card, sections 1, 2, 3, 13, and 14
Buffer and Wrap-up
About 5 minutesNext lesson preview
Next lesson: Break down the process by dividing the selected task into inputs, processing, outputs, and checkpoints.
- Create a three-color responsibility matrix and write an acceptance checklist.
- Bring the 'stuck points' column from the baseline sheet, as well as seven previous completed works.
- The project card will fill in columns 4 and 5.
Show the full transcript (video narration)
Identify Pain Points
Welcome to Lesson 5 of the AI Implementation Course. The first four lessons laid the foundation: using AI, giving instructions, writing scripts, and verifying results. Starting today, we take the first step into the main track: identifying pain points. This lesson addresses only one thing: selecting the first task from your weekly workload to delegate to AI. Once completed, you will have a Five-Point Scoring Sheet and a Success/Stop-Loss Line Card.
Lin Yue's Dilemma
Today's question is introduced through the scenario of Lin Yue, an administrative specialist. She knows how to use AI and write scripts, but she doesn't know which task to tackle first. She wants to delegate everything to AI: meeting scheduling, notifications, purchase requests, and meeting minutes. Department Manager Zheng passed by and asked, 'Is what you're doing actually worth it?' Lin Yue couldn't answer. If you choose correctly, the remaining nine lessons will yield twice the result with half the effort. If you choose incorrectly, no one will care how beautifully you execute it.
Principles for Selecting Tasks: Specific, Profound, Real
The three principles for choosing tasks: Specific, Profound, Real. Specific means a concrete task. Profound means it's the work you know best; since you are the gatekeeper, handing off work you don't understand to the AI makes it impossible to verify the results. Real means it's work you are currently doing. Two more reasons: Why time it first? There is no 'after' without a 'before'. Lesson 9 calculates savings based on today's baseline. Why define success and stop-loss criteria first? If you set standards after finishing, you can always claim success.
After reviewing, categorize tasks into three types
There are two ways to review: a one-week time log, recording one line per task completed; or a one-day quick recall, spending ten minutes listing all tasks done in the past week. Aim for completeness, not precision. After reviewing, categorize tasks into three types: repetitive, time-consuming, and clear rules. Tasks fitting all three go into the candidate list.
Example: Lin Yue's weekly review (illustrative)
Look at Lin Yue's review. Weekly meeting minutes, prepared after the meeting, approx. 90 minutes. Purchase request verification, approx. 60 minutes. Event notification, approx. 30 minutes. Other miscellaneous tasks include internal consultation replies, meeting room registration, visitor appointments, and task archiving. The numbers are illustrative; please replace them with your own data. Lin Yue identified these three as fitting all three categories.
Let the AI turn your log into a candidate list
If you want to use AI to help organize, you can use this prompt. It does only three things: merge similar tasks, identify repetitive, time-consuming, and rule-clear tasks, and remind you of small tasks you might have missed. It does not make choices for you. Note: only write the task names and time spent, not client names, amounts, or staff names.
Practice A-1: Circle candidate tasks
Now, do Practice A-1. Lay out your logs or notes, and circle at least three tasks that are simultaneously repetitive, time-consuming, and rule-clear. For each, define what the input is and what the output is. If a candidate task is too broad, such as 'do administrative work well,' narrow it down to a specific, single task.
Five-point scoring: The higher the score, the more suitable the task is to tackle first
If there are multiple candidate tasks, measure them using the same yardstick. The yardstick has five items, each scored out of 5. Frequency, time consumption, clarity of rules, cost of error, and availability of materials. Note the direction: the higher the score, the more suitable the task is to tackle first. The direction for 'cost of error' is the easiest to get wrong. Remember the mantra: the lighter the consequences, the higher the score.
Scoring guideline: Tasks with a high cost of error should be set aside first
There are five scoring guidelines. First, each of the five items is scored from 1 to 5, with a maximum total of 25. Second, tasks involving money, contracts, clients, personal safety, external commitments, or online systems receive a cost-of-error score of only 1 or 2. Third, regardless of the total score, such tasks should be set aside and not be the first project. They are not impossible to do, but should be done only after the first successful experience and all safeguard measures are in place. Fourth, when scores are close, choose the one with the higher cost-of-error score; if still close, choose the one you understand better. Fifth, you are the one who scores.
Example: Lin Yue's 'Department Weekly Meeting Minutes' Score (Illustrative)
Reviewing Lin Yue's score. Frequency is 4, once a week. Time required is 4, about 90 minutes per session. Rule clarity is 4, with a fixed format and past examples. Cost of error is 3, as missing action items may trigger follow-ups but can be corrected later. Data availability is 5, with records on hand. Total is 20 out of 25. Numbers are illustrative.
Score Reviewer: It only asks follow-up questions, it does not assign scores for you
This prompt assigns the AI the role of a score reviewer. Its function is to question you, not to score for you. The instructor runs the exercise once on site, and we observe four actions. First, timing: do not cut pauses or errors. Second, pausing midway: stop if a reason is missing and fill it in first. Third, review its questions: pick one key question, such as 'What is the basis for scoring cost of error as 3?', answer it yourself, and articulate the premise you had not previously considered. Fourth, change only one condition and retest to see if the ranking changes.
Practice A-2: He Lei's four tasks, with reference scores
Practice A-2: If you don't have your own example, use Project Manager He Lei's week. Four tasks: weekly meeting minutes, schedule updates, client risk list, and handling miscellaneous messages. Reference scores: weekly meeting minutes 19, schedule updates 17, handling miscellaneous messages 14, client risk list 12. Why is the most time-consuming task, the client risk list, not suitable as the first project? Because it involves contracts and breaches of contract, making the cost of errors the heaviest. Your reasoning is more important than the score.
Important, but put aside for now
Some tasks are important and time-consuming, yet yield a low total score. For example, the review of open-source components and client confidentiality clauses for Duchuan Telecommunications Legal Compliance totals 11 points, because conclusions can only come from the legal head, and original contract texts are red-line materials. How to handle it? Not that it's never done. Wait until the first project with lighter consequences runs smoothly, then do a version that only organizes, does not conclude, such as only grouping and excerpting clauses, listing what to ask legal about.
Interview colleagues: Ask about the current state, not whether to use AI
The principle for interviewing colleagues is: Ask about the current state, not whether to use AI. If you ask if they want to use AI, they will likely politely say 'yes,' which provides no useful information. Ask about facts: How was it done last time? Where are the bottlenecks? What errors have occurred? The five questions take under ten minutes each, one colleague at a time. After asking a question, wait three seconds; the other person often adds a candid remark.
Colleague A's interview
Here are the key points from Lin Yue's interview with a colleague. Minutes are usually received by the next afternoon. The biggest fear is action items missing names or deadlines. Last month, an action item was assigned to the wrong person, and it wasn't discovered until a week later. The manager reviews the minutes mainly for decisions, responsible persons, and completion times. What colleagues want most is to have the day's minutes so they can schedule next week's work on the same day. Lin Yue originally thought the biggest problem was slow organization, but what colleagues care about is whether they can see the minutes on the same day.
One-sentence value proposition: Has an object, has numbers, has a supplement
Write a one-sentence value proposition using this structure. Lin Yue wrote: 'I help my department colleagues reduce the time spent compiling weekly meeting minutes from 90 minutes to under 30 minutes, and ensure they are distributed on the same day.' A good value proposition has three characteristics: it identifies the audience (department colleagues, not everyone in general); it includes specific numbers (from how long to how long; these are targets, not promises); and it includes additional benefits (what else improves besides time savings, such as same-day distribution). After writing it, read it aloud three times, then ask yourself: 'What is the first step I will take tomorrow morning?'
A good value proposition has three characteristics.
A good value proposition has three characteristics: it identifies the audience (who is being helped); it includes specific numbers (from how long to how long; these are targets, not promises, and can be revised once you have real data in Lesson 9); and it includes additional benefits (what else improves, such as same-day distribution, fewer omissions, or consistent formatting). After writing, you can ask the AI to role-play as a department manager who only asks about value and does not rewrite. The prompt is in the post-class section.
Baseline: Use one standard for everything.
The baseline table should record time consumption, frequency, errors, and bottlenecks. Standardize the approach: prioritize past records or actual measurements for time; label estimates as 'estimated.' For omissions, wait until C06 produces an acceptance checklist to score seven previous manual outputs. Crucially, the baseline error count and subsequent test runs must use the same acceptance checklist. If the baseline uses one ruler and the test run uses another, the comparison is invalid.
Lin Yue's Baseline (Illustrative)
Looking at Lin Yue's baseline, the numbers are illustrative. The average time was 90 minutes; in practice, every task had a recorded duration. The average number of missed items was about 2.1 per session, totaling 15 across seven sessions, conducted once a week. The bottleneck: repeatedly listening to recordings to verify who was responsible for what, and meeting minutes were often not issued until the next day. Note: these 2.1 missed items are from manual minute-taking; this is the baseline, not the assistant's performance.
Success and Stop-Loss Lines: Written before the trial run, and not changed afterward.
Success and Stop-Loss Lines: Written before the trial run, including the date, reviewed by your direct supervisor, and not changed afterward. Using Lin Yue as an example, the Success Line: average time not exceeding 45 minutes, average missed items not exceeding 1.0, zero red-line incidents. The Stop-Loss Line: stop immediately upon one red-line incident; stop if two consecutive trial runs take 90 minutes or more; stop if after four cumulative trials the average time still exceeds 72 minutes (80% of the baseline). The numbers are illustrative. Lines written after completion do not count.
Pre-Work: Project initiation message to your direct supervisor
Before starting, send a three-sentence project initiation message to your direct supervisor: what you are doing, what materials you are using, and what you need them to confirm. Lin Yue's version sent to Mr. Zheng was a two-week pilot, using AI to draft the initial version, with final verification done by her personally, and materials desensitized first. Expected responses: approval, conditional approval, or rejection. All three are useful outcomes. For conditional approvals, copy the conditions exactly into the release record. For rejections, ask for the reason and choose a different task.
Release Record: If there is no reply that can be retained, do not start the work
Authorization is not just someone saying it's okay, but having a record. The authorization record form must clearly state: who authorized it, when, how, the scope of permission, the scope of prohibition, validity period, and attached conditions. One-sentence rule: Without written or retainable confirmation, do not enable automatic execution or paste real materials. If the other party only gave verbal approval, send a text message back on the same day; if they do not object, this message serves as the record. If you cannot get authorization, use fictional materials to practice the process while switching to a task with cleaner data.
Quiz Question 1
Quiz Question 1. In the five-point scoring system, what does it mean if the cost of making a mistake is one point? A, It doesn't matter if you make a mistake. B, The consequences of a mistake are serious, involving money, contracts, clients, or personal safety, and are difficult to reverse. C, The task is easy. The answer is B. Scoring is based on consequences: the lighter the consequences, the higher the score. A score of 1 means the consequences are the heaviest, so you should put it aside for now.
Quiz Question 2
Quiz Question 2. When interviewing a colleague, which question is appropriate? A, "Do you want to use AI to do this?" B, "Last time you did this task, please walk me through it from start to finish." C, "Do you think AI will replace you?" The answer is B. Asking about the current situation and specific past instances yields facts, not just polite pleasantries.
Quiz Question 3
Quiz Question 3. When should you write down the success criteria and the stop-loss threshold? A, After the trial run, review the data and decide. B, Write them down before the trial run, include the date, and do not change them afterward. C, No need to write them; just go by feel. The answer is B. Defining standards after completion allows anyone to claim success. Writing them down beforehand ensures your conclusions are credible two weeks later.
Assignment: Project Card, sections 1, 2, 3, 13, and 14
There are five parts to the assignment. First, a five-point scoring sheet for at least three candidate tasks. Second, a de-identified colleague interview record, including answers to the five questions and one unexpected finding. Third, the selected task's name, responsible person, frequency, and duration per instance. Fourth, a value proposition statement, a baseline record sheet, and a success/stop-loss threshold card. Fifth, the reply to the project initiation message and the completed release record form. AI can help you polish your text, but it cannot fabricate your time logs, your colleagues' exact words, or your manager's replies.
Next lesson preview
The next lesson is 'Deconstructing Processes.' We will break down the selected work into inputs, processing, outputs, and checkpoints, create a three-color division of labor table, and write an acceptance checklist. Please bring the column where the card is stuck in the baseline table, as well as the seven previous completed works. For the next lesson's assignment, you will use the acceptance checklist to review them. The people, companies, and numbers in this lesson are fictional examples. The numbers are illustrative; please replace them with your own data.