Run an AI session with your team.Run an AI session with your team.Run an AI session with your team.
Share this screen and follow one action at a time. Your team will check a campaign brief, record its reasoning and leave with a named next step.
Seven steps · About 60 minutes · Fictional data · No AI account required
Loading this browser’s session…
Step 1 of 7 · Assign roles0 stages recorded
Return to any recorded stage. Editing an earlier output reopens it and later stages; your text is retained.
Reset this team session
This clears the team session only. Export it first if you want to keep it. Your individual exercise, lessons and pilot remain in My AI Plan.
Fictional source pack 2026-11-16. This workspace makes no model or vendor-account calls.
Facilitator reference and offline kit
Leave with an artifact someone can check.
Each person or group produces an executive brief, a calculation and correction ledger, and a bounded follow-up with an owner. The goal is to explain why the final output is usable and where its evidence ends.
The pack includes a CSV, definitions, a working brief, a deliberately flawed draft, an answer-free learner checklist and full facilitator notes. Keep its reviewed reference and correction ledger as facilitator answer keys until the group records its own review. Its records and results are invented for practice. They do not describe Joseph Carroll’s experience or a real business.
Download the pack and open the CSV and source notes before the session. Share review-rubric.md as the answer-free learner checklist. Keep the reviewed reference and correction ledger aside until participants have completed and recorded their own first review.
Check that everyone can read the files. For the KPI workbench, use “Load a campaign CSV” to select campaign-performance.csv from this pack; its default sample is different data. Confirm the uploaded filename before calculating. A spreadsheet, calculator or printed source rows are valid alternatives.
Decide whether an approved assistant will be used. No model, API key or account is required; the supplied flawed draft supports the same review exercise.
Use only the fictional pack. A colleague may describe a real workflow, but should not introduce private records, credentials or live account actions during this exercise.
Arrange pairs or groups of three: lead/writer, analyst and reviewer. Choose the working arrangement below and one final editor. Provide written instructions, allow individual reading time and pair readers who need help with unfamiliar metrics.
Decide in advance whether effort is being measured. If it is, record preparation, execution, review and correction separately, along with case difficulty and missing measures.
Load this exercise’s data
Download campaign-performance.csv from this exercise. In the KPI workbench, use “Load a campaign CSV” to select that file. Its default sample is different practice data. Before using any totals, confirm the status says “Uploaded CSV” and names campaign-performance.csv. Download the calculated report before leaving the tool. Reloading returns to the default sample; check the filename again when you return.
The lead also writes the first draft unless a writer is explicitly assigned. The analyst returns a calculation ledger. The reviewer starts after a draft exists and checks it against the original sources. The lead owns the sources and integrates corrections into the final brief. With one assistant or no assistant, perform those roles in sequence.
A second agent can check evidence; its agreement does not certify the output. Separate apps need an explicit copy of the source pack, permissions and handoff. This kit does not create a shared workspace.
The lead/operator owns My AI Plan in that browser and enters the group’s brief, review and final artifact. Other participants inspect the sources and discuss corrections. Export the group’s Markdown artifact and retain it in an approved location. The Lab does not synchronize another participant’s browser.
Separate browsers
Each participant works in their own browser and exports Markdown. The lead collects those exports in an approved lead-owned document or folder and manually integrates the agreed corrections into one final version. Share only permitted artifacts. Do not import participant JSON backups to combine work: JSON restore replaces the receiving plan; it never merges plans.
A 60-minute session
Give participants the learner checklist. Withhold the reference brief and correction ledger until they have recorded their own review. Use the flawed draft when an assistant is unavailable or when the group needs the same artifact to inspect.
00–05 minutes
Frame the outcome and permissions
Show the fictional task and the finished-artifact criteria. Name a lead who also writes, an analyst and a reviewer; explicitly reassign writing if needed. Choose a shared-screen or separate-browser arrangement. Agree that only drafting and review are permitted.
Ask: What should the marketing lead be able to decide after reading this?
Keep: A shared outcome and draft-only boundary.
05–13 minutes
Inspect the sources
Read S1 and S2 before opening the reference answer. Account for the repeated Search brand label, period, currency and purchase definition. For the KPI workbench, download campaign-performance.csv from this exercise, then select it under “Load a campaign CSV”. The default sample is different. Confirm “Uploaded CSV” names campaign-performance.csv before using the totals, and download the report before leaving. A calculator is also valid.
Ask: What tells us to combine these two rows, and what would we need to verify in a real export?
Keep: A source check and reconciled input totals.
13–21 minutes
Write the brief and handoff
Adapt the supplied working brief. Name the analyst’s ledger, the lead’s draft (or an explicitly assigned writer) and the reviewer’s correction list. Specify dependencies, where the artifacts will live and one final editor. Share only these fictional inputs if using an approved assistant.
Ask: What must the reviewer receive to check the work independently?
Keep: A working brief and explicit ownership.
21–29 minutes
Produce a first draft
Use an approved assistant, draft manually, or use the supplied flawed draft. Record which route you used. Do not reveal the reference yet. Capture actual effort and any setup or waiting separately if measuring the session.
Ask: Which claims are observations, and which would need more evidence?
Keep: A draft to inspect, not an approved brief.
29–43 minutes
Review the evidence and correct the draft
Swap artifacts with another person or run a separate reviewer step. Use S1 and S2 with the answer-free learner checklist to check each material claim. Record the error, source, correction and remaining uncertainty. Reveal the facilitator review ledger only after the group has recorded its own review.
Ask: Which statement looked plausible but was not supported, and how did you prove that?
Keep: A correction ledger and revised calculations.
43–53 minutes
Finish a checked executive brief
One owner integrates the corrections. Compare the final artifact with the reference and rubric, allowing different defensible wording. Save or download the brief and notes. Keep unresolved checks explicit.
Ask: Does the next step follow from this evidence and stay inside permission?
Keep: A concise checked draft with traceable claims and a proposed next action.
53–60 minutes
Debrief and choose a real pilot
Discuss what changed between drafts. Separate observed session effort from any claimed business savings. Name one suitable real task, its owner, data permission and the evidence needed before a pilot. Use My AI Plan to record the follow-up.
Ask: What would you retain, change or stop before trying this on approved real work?
Keep: One bounded follow-up with an owner, not an organization-wide rollout.
Coach the check, not the answer.
Someone averages campaign rates.
Ask them to name the aggregate numerator and denominator. Compare total clicks divided by total impressions with an unweighted average.
Revenue minus campaign spend becomes profit.
Ask which costs, refunds and adjustments the sources actually contain. Correct arithmetic can still support the wrong interpretation.
A high ROAS becomes a budget instruction.
Ask what evidence establishes incremental impact or a return on additional spend. Bring the group back to the draft-only permission.
Two agents agree.
Ask for the source rows, calculation and claim-level review. Agreement is not a substitute for evidence or an accountable human.
If the group gets stuck
No approved assistant: use the supplied draft. No network: use printed sources and a calculator. No errors found: ask for the source behind the claimed trend. Reviewers disagree: reconcile the source and formula instead of voting. Time runs out: retain the draft and unresolved checks; do not mark it finished to satisfy the agenda.
Adapt the group, keep the evidence standard.
Owner-led business or small team
Use one operator and one reviewing partner if available. One person can perform the roles in sequence, but should return to the original sources during review. Choose a narrow recurring task, such as preparing a campaign summary. Identify a practical backup owner before making it a routine.
Department or growing team
Work in groups of three and exchange review ledgers across groups. Compare the same source pack and rubric, then agree on one output format and handoff. Have the real workflow owner review whether the team’s existing tools and permissions cover the proposed pilot.
Large or complex organization
Run the same bounded exercise inside existing tool and data rules. Include a process owner and an appropriate reporting or platform owner in the debrief. Record which exact use, configuration and sources need approval. A workshop result does not approve access or justify a wider deployment.
Headcount does not determine readiness or permission. Match the pilot to the actual data, actions, existing tools and operating support.
Debrief with the work in view
Which change improved the usefulness of the final brief, and what artifact shows that?
Which claim survived a second check? Which claim was removed, and why?
What could a second agent verify, and what still required source evidence or a human decision?
Where did time move: preparation, drafting, review or correction? What was actually measured?
What input, permission, owner or fallback is missing for a real version of this task?
Capture the artifact version, reviewers, unresolved checks, observed effort and proposed real task in the session record included in the facilitator notes. A completed fictional example shows practice; it does not establish business savings or production readiness.
Keep the checked artifact and correction ledger together. Record who reviewed which version and what remains unresolved.
Choose one real task and describe its sources without copying private records into the Lab. Confirm data, action and tool permissions with the accountable owner.
Observe a comparable current-work case before claiming improvement. Use the pilot comparison to record quality, total effort, elapsed time and cost separately; leave unmeasured values unknown.
Run a bounded pilot on representative cases, including missing or ambiguous input. Record errors and recovery as well as successful outputs.
Make an adopt, adapt or stop decision from those observations. Publish or operationalize only through the organization’s actual approval process, with an owner and fallback.
My AI Plan is local to this browser when storage is available. Export the work to retain it or move devices. Share only artifacts your team is permitted to share.