Sidekick Orchestration
A worked example

From sales exports to a management decision

Website enquiries doubled while sales stayed flat. Before increasing spend again, check lead quality and response delays.

Fictional service businessFictional recordsManagement report you can read in fullNo client claim
Records in · finished work out

3 records.
4 automatic steps.
One human check.

What goes in

The records the workflow reads

3 ordinary records, nothing prepared specially for AI. Every fact in the finished work traces back to one of them.

X1 · Sales export

CRM export · August to September

  • Referrals: 10 enquiries / 5 sales → 8 enquiries / 4 sales.
  • Website: 20 enquiries / 4 sales → 40 enquiries / 4 sales.
  • Events: 6 enquiries / 2 sales → 8 enquiries / 2 sales.
  • September dashboard headline: 12 sales.
M1 · Marketing spend

Campaign invoices

  • Website campaign: 800 in August; 1,600 in September. Event fees: 300 each month. Referral channel cost was not tracked.
  • Figures are fictional Canadian dollars; these are channel costs, not total business costs.
N1 · Operating notes

Manager’s notes for the month

  • Median first response to website enquiries rose from 4 to 19 hours. A new campaign started in September.
  • No lead-quality review or source-attribution check has been completed. Management must decide whether to double next month’s website budget.

How the work runs

From records to a review-ready result

Set up once: the report layout, the calculations and the questions the owner always asks live in a saved project. Each month the three exports are dropped in and the same report comes back, so review time goes to the decision.

A saved project is a folder inside your AI tool that keeps its instructions and reference files between conversations, so nothing is re-explained.

  1. The three files are read together

    Sales export, spend and the manager’s notes are combined; the dashboard headline is treated as a claim to check, not a fact.

  2. Rates and costs are calculated

    Enquiry-to-sale rate per channel and website cost per sale are worked out from the rows, month by month.

  3. The total is reconciled

    Channel rows add to 10 sales; the dashboard says 12. The gap is reported, not smoothed over.

  4. A recommendation is written with its limits

    Hold the spend increase pending two checks. Cause is not claimed because the campaign and the response delay changed at the same time.

  5. A person reads, corrects and acts

    Check the figures against the export, decide the budget, assign the two follow-up checks.

What comes out

The finished artifact

This is the full review copy. Read it the way the owner would, with the records above open beside it.

Management reportFictional · prepared for review
Review copy

Prepared for review · fictional records

Channel performance and the website budget decision

Prepared for
Owner and manager
Built from
X1, M1, N1
Period
August to September
Decision
Double website spend: hold

Channel performance · August → September

ChannelEnquiriesSalesSale rate
Referrals10 → 85 → 450% → 50%
Website20 → 404 → 420% → 10%
Events6 → 82 → 233.3% → 25%

Budget decision

Hold the proposed doubling of website spend pending a short review. Recorded website cost per sale rose from CAD 200 to CAD 400 while sales stayed at four. More enquiries have not yet translated into more sales. This does not establish profitability or prove which factor caused the change.

Resolve before the next report

  • The September channel rows total 10 sales, while the dashboard says 12. Reconcile the two extra sales and their source before using the dashboard total.
  • Review September website enquiries by campaign: qualified, unqualified, no response and still open. Separate lead quality from leads that have not yet had time to close.
  • Review the increase in median first-response time from 4 to 19 hours. Assign enquiry coverage and compare response times and sales progression during the next reporting period.

What remains uncertain

The campaign and response delays both changed. These records do not isolate their effects. Referral costs are missing, and channel costs exclude other business expenses. A definitive channel-profit ranking would be unsupported.

Human review

What the owner checks before it goes out

Current rows total 10 sales, not 12. Website conversion halves and recorded cost per sale doubles. Neither the cause nor profit is established.

Try it

Run the same records through your own AI tool

The complete fictional pack and the instruction are below. Paste both into an AI chat you are allowed to use, then compare what comes back with the brief above.

Copy this fictional example

Copy the notes and instructions into an approved AI chat.

Fictional source material: X1 · CRM EXPORT · AUGUST TO SEPTEMBER Referrals: 10 enquiries / 5 sales → 8 enquiries / 4 sales. Website: 20 enquiries / 4 sales → 40 enquiries / 4 sales. Events: 6 enquiries / 2 sales → 8 enquiries / 2 sales. September dashboard headline: 12 sales. M1 · CAMPAIGN INVOICES Website campaign: 800 in August; 1,600 in September. Event fees: 300 each month. Referral channel cost was not tracked. Figures are fictional Canadian dollars; these are channel costs, not total business costs. N1 · MANAGER’S NOTES FOR THE MONTH Median first response to website enquiries rose from 4 to 19 hours. A new campaign started in September. No lead-quality review or source-attribution check has been completed. Management must decide whether to double next month’s website budget.
Your task: Prepare a concise management report with a comparison table, the decision you recommend, evidence, uncertainty and next actions. Reconcile totals. Calculate enquiry-to-sale rates and website cost per sale. Distinguish association from a proven cause; do not infer profit.

Check the result

Current rows total 10 sales, not 12. Website conversion halves and recorded cost per sale doubles. Neither the cause nor profit is established.

Apply it to your work

Before trying this on your firm's real files, use a business AI account and make sure someone checks every result before it is used.

Try a completed reporting period with the original exports, costs and operational notes. Check the report’s conclusions as well as its calculations.