Qelv

A prototype we built for ads-ops teams. September to December 2025.

The morning ads report, written before anyone logs in

The morning ads report is the same every day: yesterday against the day before, per campaign, what to act on at the top. This prototype builds it from the platform export before anyone logs in; the builder on this page runs the same arithmetic on a sample.

Paid media. Automate.

Open the demo (opens in a new tab)

Ads-ops agent in figures

  • 40 minutesof a person's morning, and a spreadsheet, before the prototype
  • 7measures per campaign, from spend to cost per acquisition
The ads-ops prototype's start screen: a drop zone for a CSV or JSON export, a start and end date, a demo mode button, and an empty report panel
A capture of the prototype, sample data.

The situation

The report is the same every morning: yesterday against the day before, per campaign, with the three or four things worth acting on pulled to the top. It takes a person forty minutes and a spreadsheet. We built the prototype to find out how much of that is pure arithmetic a machine can do from the export, and how much still needs a person.

What we built

  • An import step that accepts the CSV and JSON shapes the ad platforms export, and says which rows it could not read rather than dropping them
  • Per-campaign aggregation: spend, impressions, clicks, click-through rate, cost per click, conversions and cost per acquisition
  • A comparison against the previous day in the same export, where the export covers more than one day
  • A rules pass that pulls out what to look at first: spend with no conversions, cost per acquisition moving against the run rate, click-through rate under the floor
  • A demo mode with sample rows, so the surface can be shown without anybody's real campaign data on screen
  • A report builder on this page that runs the same arithmetic on a sample export, entirely in the browser

The screens

Sample data.Captures of the prototype on sample data.

The start screen

Where the morning starts: a drop zone for the platform export, the date range the report covers, and a demo mode that loads sample rows so the surface can be shown without anybody's real campaign data in it. Nothing is generated until an export is handed over.

The ads-ops prototype's start screen: a drop zone for a CSV or JSON export, a start and end date, a demo mode button, and an empty report panel
An export goes in here, or the sample rows do.

The start screen

What changed

Most of the report turned out to be arithmetic the machine does from the export in seconds. What is left for the person is the part that needs judgement: what to do about the three or four lines at the top.

Try it

Sample data.Two invented exports held in this page, with invented campaign names and figures. The report is assembled in your browser; a file you drop is read locally and never uploaded.

The export

Load one of the invented exports, paste your own, or drop a CSV file onto the box. Your own data is fine: it is read in this tab and goes nowhere.

  • Two days, six campaigns. The ordinary case: yesterday and the day before, in the platform's own column order. It trips all four rules and carries one row that will not parse.
  • Same day, columns moved. One day only, columns in a different order, a currency symbol in the spend column and a blank row. The report says it cannot compare rather than comparing against nothing.

A header row, then one row per campaign per day. Columns are matched by name: date, campaign, impressions, clicks and spend are required, conversions and channel are used when present.

What decides the flags

Four rules, printed here rather than hidden behind the output. A flag whose rule you cannot see is a flag you cannot argue with, and on a real account at least one of these is always wrong for you.

  • Spend of $100 or more with no conversions on the day.
  • Cost per acquisition more than 1.4 times the day before.
  • Click-through rate under 1 percent.
  • A campaign that reported the day before and reports nothing now.

The morning report

Sample data.

Nothing built yet. Load an export and press the button; the table, the comparison and the rows it could not read all appear here.

Read next

Elsewhere on the site

Where to next

Have a problem in this shape?

Say so in a few sentences. The reply comes from the person who built this one.

Send us the bottleneckThe Automate practiceEvery story

We reply within two business days. If it is a fit, you get a one-page note on what to do first, before any sales call.