← Back to AI Agent Scenarios

SCENARIO 20·HR·Responsibility Inventory

Ask Employees "What Do You Want AI to Do?" and Most Responses Are Blank or Just Say "Improve Efficiency"

The first step in AI implementation is often to send out a needs assessment form and ask each department what it wants AI to do. However, employees must first understand what AI can do. This is difficult for people who have not used AI.

Management wants to know where to deploy AI first and how much time it can save. However, HR and the implementation team only receive responses such as "purchasing management" and "quality management," which do not provide enough information to set priorities.

At a glance

How it was done

  1. Send a form asking "What should AI do?"
  2. Receive job function names or blanks
  3. Start with tasks that appear likely to deliver the fastest results
  4. Managers select the first tasks

Where it gets stuck

Most responses are job function names such as "purchasing management." The team cannot set priorities, so it starts with tasks that are easiest to measure. Responsibilities that consume the most work hours are not identified.

With an AI Agent

  1. Send a form asking "What should AI do?"
  2. Receive job function names or blanks
  3. Start with tasks that appear likely to deliver the fastest results
  4. Managers select the first tasksHuman approval

Ask employees what they do in their regular work, calculate monthly work hours, and use the TURBO criteria to flag responsibilities suitable for AI Agent. Department managers and the implementation team decide which tasks to start with and which should not be assigned to AI.

01

How it was done

  1. 01Send out a needs assessment form and ask each department what it wants AI to help with
  2. 02Most responses are job function names or general expectations
  3. 03The implementation team can only start with tasks that are easiest to measure and appear most likely to deliver quick results
  4. 04Responsibilities that consume the most work hours are not identified
AI 需求調查_回收彙整.xlsxExcel
A1Example Machinery: AI Implementation Needs Survey Response Summary
Example Machinery: AI Implementation Needs Survey Response Summary
Survey Question: Describe what you want AI to help you do
DepartmentWhat should AI help with?
PurchasingPurchasing management
PurchasingResponding to suppliers
Quality AssuranceQuality management
FinanceHelp me review invoices
SalesAutomatically respond to price inquiries
AdministrationCompile weekly reports
Production(blank)
Response Summary100%
Illustration only. When asked, "What do you want AI to help with?" employees respond with job function names such as purchasing management and quality management, or seemingly specific answers such as "help me review invoices" and "compile weekly reports." However, they do not identify which data to process, what to produce, or how often the task occurs. Some departments return blank forms. The implementation team cannot set priorities.
02

With an AI Agent

  1. 01
    What it doesCalculate monthly work hours
    Standard appliedMonthly frequency × time per task × number of users; count daily as 22 times, weekly as 4 times, monthly as 1 time, and quarterly as 1/3 time. Convert durations not stated in minutes and flag them
    OutputMonthly work hours for each responsibility
  2. 02
    What it doesFlag using the TURBO criteria
    Standard appliedEvaluate one item for each of the five criteria: time required meets the company's defined threshold (T), enough users perform the task (U), the task occurs at least once per week (R), someone can review it (B), and the steps are clear and easy to break down (O). List the letters for the criteria met
    OutputResponsibilities ranked by the number of criteria met
  3. 03
    What it doesList data that needs correction
    Standard appliedIf an entry only names a job function, does not show its inputs and outputs, or omits the frequency or time required, do not calculate its work hours or rank it. Add it to the correction list. If two people enter the same content, calculate both entries as usual and flag them as possible duplicates. Do not rewrite or merge them
    OutputCorrection list and department totals
  4. 04
    What it doesSend to department managers and the implementation team
    Standard appliedReview department totals only. Do not rank employees or compare individuals. Remind department managers and the implementation team to review implementation progress together. They decide which responsibilities to handle first
    OutputDepartment managers and the implementation team select the first tasks
Human approval
Department managers and the implementation team select the first tasks. They decide which responsibilities to handle first and which should not be assigned to AI. AI Agent organizes and initially flags the items but does not set priorities for people.
Platform modules used
Skills

We recommend describing each responsibility as an action with identifiable inputs and outputs, rather than listing only a job function. Initially, prioritize tasks with clear rules, consistent inputs and outputs, and visible errors. TURBO provides five criteria for selecting the first tasks: time required, number of users, high frequency, availability of a reviewer, and ease of breaking down the process.

Screenshots show the Traditional Chinese interface.

01Upload the Responsibility Inventory and Instructions

EgentWrX conversation screen: An employee uploads the Purchasing Department's responsibility inventory. The instructions define how to calculate monthly work hours and the thresholds for the five TURBO criteria. Data requiring correction is not ranked, rewritten, or merged.
  1. 1Upload the Purchasing Department's responsibility inventory.
  2. 2The instructions define how to calculate monthly work hours.
  3. 3Each of the five TURBO criteria has a specific threshold.
  4. 4Do not rank, rewrite, or merge data that requires correction.

02Execution Process

EgentWrX execution screen: AI Agent reads the responsibility inventory item by item. The fields include responsibility, frequency, time required, number of users, availability of a reviewer, and whether the steps are clear.
  1. 1Read the inventory in 3 steps.
  2. 2Read the entire inventory item by item.
  3. 3Fields include responsibility, frequency, time required, number of users, availability of a reviewer, and whether the steps are clear.

03Responsibility List

EgentWrX results screen: The responsibility list is ranked by the number of TURBO criteria met. It shows monthly work hours and the applicable letters, with notes identifying conversions and possible duplicates.
  1. 1The three responsibilities that meet all five criteria appear first and can be considered for the first tasks.
  2. 2Responding to supplier inquiries takes 22 hours per month, but no reviewer is available and the steps are unclear, so it meets only three criteria.
  3. 3Convert the entered duration of "2 hours" to 120 minutes.
  4. 4Two people entered the same task. Flag it as a possible duplicate, but do not merge the entries.

04Data Requiring Correction

EgentWrX results screen: The correction list identifies entries that only name a job function, omit frequency, or may be duplicates. It then shows the number of items that can be calculated and the total monthly work hours.
  1. 1Do not calculate work hours for the two entries that only name job functions. Ask the employees who completed the form to correct them.
  2. 2One entry omits the frequency, so its monthly work hours cannot be calculated.
  3. 3Calculate possible duplicates as usual and only flag them.
  4. 427 items can be calculated, totaling 223.42 hours per month.

This scenario has no client figures; the description above reflects typical practice.

03

FAQ

Why not ask employees directly what they want AI to do?
That approach requires employees to understand what AI can do and break their work down into tasks that can be assigned. Instead, ask what they do in their regular work. Everyone can start by listing tasks that occur daily, weekly, or monthly. Whether AI can perform those tasks can be determined in the next step.
What if employees worry that reported work hours will be used for performance evaluations?
Clearly explain the purpose and access permissions first. Use the inventory only to estimate task value and establish a baseline, not to rank individuals. Only department managers and the implementation team can access the raw data. Management meetings review department totals only. If the time required is difficult to estimate, start by recording frequency and steps.
What should we do after the inventory is complete and the first tasks are live?
Ask department managers and the implementation team to review implementation progress together. Confirm that the first tasks are operating reliably, have assigned maintainers, and have people responsible for handling exceptions. Then select the next tasks from the same inventory list. This allows each expansion step to use the experience gained from the previous group of tasks.

Where to go next

30 minutes to map out which work to hand to AI first

Want every employee to have their own AI teammate?

A consultant will be in touch shortly.