What you’ll make
One read-only decision brief with a checked cost per result and one question to investigate.
Worked example
A small campaign review for an online shop
One ad group spent $120 for 6 purchases and another spent $60 for 2 purchases. Together they spent $180 for 8 purchases, or $22.50 each. A separate ad spent $40 and recorded zero purchases. Include its spend if it belongs to the same report. In that case, $220 divided by 8 is $27.50 per purchase.
A change in clicks gives you something to investigate. If you think a different ad image affected sales, check which ad changed and which other settings stayed the same before planning a test.
Before you start
- Choose a Meta Ads account and a defined conversion event. Bring matching date ranges, your review threshold, currency, time zone, and the attribution setting displayed in your report.
- Use Manus’s official Meta Ads Manager connector for a read-only route if available in your workspace. An Ads Manager export works without giving the assistant account access.
Pull one report with a clear scope
Use a Meta Ads export for one date range and reporting level. Write down which result counts, the currency, and the settings you used. Bring the rows with spend and results or enter their totals above. Keep summary rows separate.
Ask for a brief from those rows
Paste the prompt into Manus or a chat that can read your export. Attach it or paste the rows, plus your scope and question if you didn’t fill in the boxes. Look for the exact spend and result counts it used.
I’d send $120 for 6 purchases and $60 for 2, with a note that I still need to check where a $40 ad belongs.
I’d expect $180 ÷ 8 = $22.50 for the two confirmed rows, and a note that adding the $40 would bring the figure to $27.50.
Turn this ad report into a decision brief
Write a read-only brief from one ad report. Use the details I provide in this chat or in my attachments. Restate the report period, currency, reporting level, conversion event, and filters I supplied. Ask for any missing setting that could change which rows or results belong together. Check that totals don’t include a summary row or duplicate data. Divide total in-scope spend by total results. Keep zero-result spend in scope when it belongs there, and show cost per result as undefined when the denominator is zero. Give me one short brief with the source totals, calculation, and one question to investigate. Identify any assumption that would change the result. Keep observations separate from causes I haven’t tested. Leave my account settings as they are.
Keep one number and one question
Save the date range, filters, and verified cost per result in the brief. Write down one thing to check next. Here I’d find out whether the $40 ad used the same filters before comparing the brief with my $25 target.
Reconcile it against the export
Open the export and trace the numbers yourself before you make a decision from the brief.
- Confirm the rows use the same dates, currency, and result definition.
- Add the spend and results, then divide once using those totals.
- Check the brief names the remaining question and marks any number you haven’t verified.
Optional: Compare a second period
Use a second export with matching definitions and date ranges of the same length. Trace the biggest change back to the source rows and write down one question to investigate.
Review my ad results from the source numbers
Use this with a read-only report or export to reconcile costs and find what’s worth investigating.
Review [read-only Meta Ads report or export] for [account, dates, time zone, currency, attribution window, conversion event, and reporting level]. My target cost and minimum evidence are [rules]. First confirm the available fields and filters. If you can’t access the report or a required setting is missing, tell me what to export or confirm and show which calculations are on hold. Make a table with IDs, spend, event counts, and cost per result. Treat zero results as undefined cost and flag low counts separately. Reconcile spend and events against the source totals. Calculate any combined cost as total spend divided by total events. Flag mixed currencies, event types, or overlapping rows before combining anything. Give me three observations backed by rows and dates, one possible explanation to investigate, and one experiment proposal for me to review. Work from read-only data throughout.
Compare two periods on the same terms
Use this with two comparable exports to see which rows drove the change in cost or results.
Here are [current and previous date ranges and reports]. Confirm they cover equal-length periods, the same level, currency, time zone, conversion event, and attribution settings. Reconcile spend and events in each and calculate blended cost from the totals. Use stable IDs to show which rows drove the biggest changes in spend or results. Call out new or missing rows and mark late conversions as a possibility if the recent data is incomplete. Keep suspected effects of creative or traffic mix tied to evidence I supplied. If the periods can’t be compared, list the exact filters I need to rerun.
Design one experiment I can review later
Use this after reconciling the report to turn one finding into a test brief for your review.
Start from [verified finding, affected ads, and business goal]. Write one proposed experiment changing [one variable]. Include the comparison setup, what stays fixed, the result and cost metric, how much evidence I want before deciding, and the review date. Explain what outcome would support the idea and what would leave it unresolved. Use [my budget and reporting limits] if provided, or mark the missing choices for me. End with the account, level, filters, and date rules to pull the same report again. Return the brief for review without making account changes.
Troubleshooting
Manus and Ads Manager show different results.
Check the ad account, date boundaries, time zone, attribution window, event, and aggregation level. Compare one known row first and use an export if the mismatch remains unexplained.
The review recommends pausing on a single bad day.
Expand the window, check conversion delays, and apply your evidence rule. Treat the recommendation as a candidate for human review.