A path from the foundations to useful work

Turn business data into a checked decision brief

Learn to reconcile inputs, calculate reproducible metrics and separate observations from hypotheses.

16 connected lessons and a capstone
Inspect the capstone

Build the skills, then put them together

These lessons reuse the common curriculum. Open any lesson, and return to the capstone when you are ready to assemble the outputs.

Your saved progress loads in this browser.

Completion records your own checks. It is not a test score or a certification. You can open any lesson.

Start the first lesson
  1. Understand models, assistants, agents and automations
  2. Define the result before choosing the tool
  3. Decide what AI, software and people should do
  4. Map the work people actually do
  5. Record how the work performs today
  6. Find the constraints that change the plan
  7. Choose a platform from the work and constraints
  8. Give every agent a job and an owner
  9. Write a pilot you can actually evaluate
  10. Test ordinary work, messy work and when to stop
  11. Calculate first, then write the decision brief
  12. Review the output against evidence, not confidence
  13. Measure the whole job, including review
  14. Decide whether to adopt, adapt or stop
  15. Turn the accepted workflow into a playbook
  16. Keep the workflow useful after launch

Your capstone exercise

Produce a reproducible performance decision brief

Synthetic capstone: use the KPI workbench's labeled sample CSV. It contains four rows and three campaigns; repeated campaign names are valid contributions to the aggregate. The full sample totals are spend 500, impressions 47,000, clicks 770, conversions 58 and revenue 1,740. It has no prior-period comparison, so it cannot establish a trend. All figures are learning data, not real account performance.

Work with an approved public input or a clearly labeled fictional example. Completing the exercise does not authorize access to real systems or publication.

Prepare the inputs

  • Download the synthetic CSV from the KPI workbench and preserve the original.
  • Use the documented columns: campaign, spend, impressions, clicks, conversions and revenue.
  • Define the reporting grain and sample period as practice assumptions; record that currency and attribution are unspecified.
  • Prepare normal, missing-column and zero-denominator variants in separate copies, with expected behavior written before loading them.
  • Name the intended recipient and the question the brief should help them investigate.

Do the work

  1. Map source → validation → calculation → interpretation → review, assigning an owner for the calculation and the final brief.
  2. Inspect row counts, labels and required values. Do not remove the repeated campaign row merely because its name repeats.
  3. Load the sample into the local KPI workbench. Reconcile the displayed totals with the original and inspect the formulas.
  4. Verify aggregate CTR as 770/47,000, CPC as 500/770, click-based CVR as 58/770, CPA as 500/58 and ROAS as 1,740/500. Preserve the underlying values and label rounded display figures.
  5. Run the missing-column and zero-denominator cases. Expect an actionable error for missing required data and unavailable ratios for zero denominators.
  6. Write a concise brief with observations, hypotheses or questions, missing context and an owned next action. Ask AI to draft from the checked result only.
  7. Review every material number against the calculation. Record corrections, effort and limitations; write the pilot decision and a reusable reporting playbook.

Check what you made

  • All five totals reconcile to the preserved sample.
  • Rates are computed from summed values, not averages of row or campaign rates.
  • The missing-column case is rejected, and zero-denominator ratios remain unavailable.
  • The brief does not claim improvement, deterioration, profit or causality from evidence that does not establish them.
  • Currency, period, attribution and comparison limitations remain visible.
  • Another person can reproduce the calculation and locate the source, definitions and next-action owner.

Take these outputs with you

  • Preserved source CSV, validation notes and metric definitions.
  • Downloaded calculation report with formulas and limitations.
  • A decision brief separating observation, hypothesis and next evidence.
  • Evaluation results, pilot decision and reporting playbook.

Adapt the capstone to your context

Owner-led business or small team

Use the sample and personally reproduce at least one total and ratio before asking an assistant to explain them.

Department or growing team

Have an analyst own calculations and a reviewer check the narrative against them. Agree on the KPI definitions before comparing results.

Large or complex organization

Keep practice data separate from approved operational exports. For real work, use governed definitions, authorized access and existing data ownership; confirm the actual reporting consumer can access only intended data.

Take this capstone into your plan

Use the exercise as a starting point. Add a real owner, baseline, permissions and review process in My AI Plan.

Use the sample and personally reproduce at least one total and ratio before asking an assistant to explain them.

Open My AI Plan

Keep useful references nearby

Explore another learning path