Course-data workshop · 20–30 min practice
Build a report from the sample data
A manager asks whether sales increased. The file contains a duplicate, a draft proposal, a foreign currency and an adjustment. Prepare an answer you can verify. All companies and transactions are fictional; no CRM or email is needed.
Lessons 05, 06 and 16 use the same example. Keep your earlier result and build on it rather than starting over. Time is an estimate.
- Attach the CSV or copy the request and data below. A text answer is enough; a file or dashboard is optional.
- Use explicit counting rules. A good result includes a table, sources for the numbers and exclusions, not just “sales increased”.
- Calculate the totals independently with a calculator or spreadsheet. Check the duplicate, currency, status and adjustment sign before reviewing the commentary.
Request with rules and the complete data
Prepare a report for the sales manager: a two-period table, difference in PLN, percentage change, exclusions with IDs, and up to three sentences of commentary. Use only the data below. Choose available calculation tools; show the formula and documents included in each total. Do not invent causes of growth. Ask for missing data. Do not publish or send anything. Compare 1–7 with 8–14 October 2026 (inclusive). Sum posted invoices and net adjustments in PLN. Count identical duplicate documents once; flag conflicting duplicates for clarification. Exclude draft proposals. Report EUR separately without conversion. Include the negative adjustment. Change % = (new total − old total) / old total × 100%; do not calculate an ordinary growth percentage when the old total is zero. document_id,date,customer,document_type,status,currency,net_amount FV-001,2026-10-01,Company A,invoice,posted,PLN,1000 FV-002,2026-10-02,Company B,invoice,posted,PLN,1500 FV-003,2026-10-03,Company C,invoice,posted,PLN,500 FV-004,2026-10-08,Company A,invoice,posted,PLN,1200 FV-005,2026-10-09,Company B,invoice,posted,PLN,1800 FV-005,2026-10-09,Company B,invoice,posted,PLN,1800 FV-006,2026-10-10,Company D,invoice,posted,PLN,600 OF-007,2026-10-10,Company E,proposal,draft,PLN,900 FV-008,2026-10-11,Company F,invoice,posted,EUR,400 KOR-009,2026-10-11,Company A,adjustment,posted,PLN,-200
Find the error before sharing the report
“Sales rose from PLN 3000 to PLN 6500, or 116.67%. The marketing campaign worked, so increase its budget by 20%.”
A deliberately incorrect training example, not an observed output from a particular model. Which mistakes concern the inputs, and which are unsupported conclusions?
Check the answer and why it is correct
| Period | Documents | PLN |
|---|---|---|
| 1–7.10.2026 | FV-001, FV-002, FV-003 | 3000 |
| 8–14.10.2026 | FV-004, FV-005, FV-006, KOR-009 | 3400 |
Swipe the table sideways to see the amounts.
Difference: PLN 400. Change: 400 / 3000 × 100% = 13.33%.
6500 adds a duplicate 1800, draft 900 and EUR amount 400 to the correct 3400. A correct formula does not fix incorrect input selection. The −200 PLN adjustment is already included in 3400.
The file contains no campaign data. You can describe the change, but not its cause or the appropriate marketing budget.
Download answer and assessment rubricNow try without the ready-made prompt
Change FV-006 from PLN 600 to PLN 800. Predict the change, then request an update in your own words. Save your report and the mistakes you spotted.
Second-trial answer
Second-period total PLN 3600, difference PLN 600, growth 20%. Exclusions are unchanged. If you leave the amount blank instead, the known portion is PLN 2800, but the full total needs the missing data.
Lab
Four extensions - choose a useful variant
Start with the sales workshop above. Then choose an extension that fits your tools. The skill exercise can use your checked workshop report. The 30–50 minute estimates are practice time, additional to reading.
Lab 1. The same brief in Chat and in Work
See on your own task how the two modes differ: turnaround, autonomy, clarity and layout of the result, and whether you end up with a file.
Level: BasicTime: 30 minChatGPT ChatChatGPT Work
Lab 1. The same brief in Chat and in Work
Setup
- Pick one real task you do by hand that takes you at least an hour.
- Prepare the input: an export, a file or a list of sources. No sensitive data unless you are cleared for it.
- Write the task description in five fields: goal, audience, deliverable, constraints, acceptance criteria.
Steps
- 1Paste the same brief into a normal conversation. Note the start time.
- 2Bring the Chat answer to a usable state - count how many exchanges it took.
- 3Run the identical brief, unchanged, as a Work task. Before closing the app, check that the task runs in the cloud and does not depend on your local computer.
- 4Come back for the result. Compare both side by side against the four criteria below.
What to watch
- Time: how many minutes of your attention each mode consumed (not how long it ran).
- Autonomy: how often you had to add something the other mode figured out itself.
- Format: did you get structure, or text you still have to format.
- File: is there a file at the end that you could actually send on.
Exercise outcome
A half-page note: which of your tasks Chat is enough for, and from what point Work pays off.
Prompt to copy
GOAL: [what should exist and why] AUDIENCE: [who reads it, how much time, what they already know] DELIVERABLE: [what file, what sections, what length] CONSTRAINTS: [what is off limits, what not to guess, where to stop] ACCEPTANCE CRITERIA: [how I will know the result is good] At the end, list the sources you used and the places where data was missing.
Lab 2. Meeting prep from several tools at once
Check whether the agent can handle one vague command when the context is spread across calendar, mail, notes and drive.
Level: IntermediateTime: 40 minKalendarz / CalendarGmailNotatki ze spotkań / Meeting notesDrive
Lab 2. Meeting prep from several tools at once
Setup
- Connect calendar, mail, your meeting-notes tool and drive. Check the permissions - read-only if possible.
- Make sure tomorrow's calendar has at least two meetings with external people.
- Decide up front what the agent must NOT touch: NDA clients, HR matters, private threads.
Steps
- 1Issue one vague command: "get me ready for tomorrow", plus the deliverable and the constraints.
- 2Do not hint which tools to use. That is the thing being tested.
- 3Collect the deck and go slide by slide.
- 4Check each slide against the source: was that commitment really in the notes, does the date match the email.
What to watch
- Whether the agent listed the sources it used - and whether that list checks out.
- How often it wrote "no data" instead of filling a gap with a guess. That is a good sign, not a bad one.
- Whether any slide holds information you cannot trace to a source.
- Whether anything out of scope made it into the deck.
Exercise outcome
A deck with one slide per meeting plus your discrepancy list: what the agent invented and what it failed to find.
Prompt to copy
Get me ready for tomorrow. DELIVERABLE: a deck, one slide per meeting in tomorrow's calendar: company and attendees, contact history, open commitments with dates, 3 questions to ask, one risk. SOURCES: choose them yourself from my connected tools. CONSTRAINTS: skip internal and private meetings. Send nothing, change nothing in the calendar or in mail. When you cannot find something, write "no data" instead of guessing. AT THE END: list the sources you used for each slide.
Lab 3. A live command center
Build an interactive dashboard, publish it on a chatgpt.site address, configure access and variables, then deliberately share the link.
Level: IntermediateTime: 45 minChatGPT Work@sites
Lab 3. A live command center
Setup
- Prepare an export you are allowed to show - anonymized or demo data is ideal.
- Decide up front who should see the page: only you, named people, or anyone with the link.
- If the page will call an external API, prepare a scoped, rate-limited key - never the master key.
Steps
- 1Run a task with the @sites command and an outcome brief describing the sections and what should be clickable.
- 2Collect the link and open the page in a private window - see exactly what a recipient sees.
- 3Set access and the address. If the page uses a key, store it in server-side secret storage, never in prompts or configuration sent to the browser.
- 4Ask for two live edits (a different alert threshold, an extra column) and verify the page updated at the same address.
What to watch
- Whether the private window reveals data no outsider should see.
- Whether the secret stays on the server and never reaches browser code or network responses.
- Whether the page is readable on a phone - that is where you will usually open it.
- Who takes the page down when it is no longer needed, and when.
Exercise outcome
A working address you can open on your phone, plus a calendar entry: review this page in 30 days.
Prompt to copy
@sites GOAL: a command center for [TEAM] - one page opened every morning. DELIVERABLE: a page with these sections: 1) the week's key metrics with a chart, 2) a [WHAT] table with an alert colour when [CONDITION], 3) today's task list with checkboxes persisted in the browser, 4) a notes field. DATA: the uploaded file [NAME]. Add no records that are not in it. ACCESS: private, me only, until I say otherwise. Read API keys only on the server from secret storage. Never send them to the browser. LOOKS: readable on a phone, light and dark background.
Lab 4. Build a Skill and put it on a schedule
Turn a repeatable prompt chain into a Skill, test it on known data and schedule Monday runs. Check where the task executes: cloud work needs sources available there, while local files or apps may require the computer to stay on.
Level: AdvancedTime: 50 minChatGPT WorkSkillsDriveScheduled tasks
Lab 4. Build a Skill and put it on a schedule
Setup
- Pick a process you repeat: content repurposing, data analysis, ticket triage.
- Write down your definitions - what exactly each term in the output means.
- Find data from a period you already know. It becomes your reference during the test.
Steps
- 1Ask the agent to write your prompt chain up as a Skill: steps, output format, QA, hard limits.
- 2Run the Skill on known data and compare with the result you already know. Fix the Skill, not the prompt.
- 3Repeat on a different period - the Skill should produce the same structure, not the same numbers.
- 4Schedule it for Monday mornings, with the first four runs going to approval.
- 5After a month, check the results still make sense and that the task is not running on data nobody updates any more.
What to watch
- Whether the Skill stops when data is missing instead of estimating the gap.
- Whether results are comparable week to week - same structure, same section names.
- Whether someone other than you can run the Skill and get the same thing.
- Who gets notified when a scheduled run fails.
Exercise outcome
A saved Skill, two successful tests on known data, and a live recurring task whose first runs go to approval.
Prompt to copy
Write up what we did in this conversation as a reusable Skill. IT MUST CONTAIN: - a name and one sentence on "when to use this", - our definitions (ask me about the ones you do not know), - the steps in execution order, - the exact output format including section names, - a QA list: what to check before handing the result over, - a "do not" list, including the points where you must stop. Do not put this specific case's data into the Skill - it holds rules, not numbers. Finish by proposing one test case.
Availability depends on your plan, region, workspace and administrator configuration.