Skip to content
Allness
All case studies

A multi-client marketing operation

Weekly numbers that read themselves

Every platform pulled and reconciled, the recap written by Monday, and follow-ups answered against live data.

The figures on this page are placeholders. Plausible, drawn from the work, and not yet re-measured for publication. They get replaced with confirmed numbers or removed.

Industry
Marketing services
Size
14 accounts
Duration
6 weeks
Focus
Reporting & analysis
Weekly reporting cycle

9 hrsto45 min

Recap delivery

midweektoMonday

Platforms reconciled

5to5 automatically

Follow-up questions answered

next daytoinstantly

Stack

Claude · Mastra · Google Ads · GA4 · Klaviyo · Shopify

Before

Fourteen accounts, 5 platforms each, one report a week. Nine hours of somebody’s week went to exporting, pasting, reconciling figures that never quite agree, and writing the same four sentences fourteen times over. The recap landed Wednesday or Thursday, so the week was half gone before anyone knew how the last one went. Reconciliation ate the most time. Ads reports one conversion number, GA4 reports another, the store reports a third, and a human has to decide which one to quote and then explain the gap. Anomalies hid in plain sight: a 40% drop in one channel sitting in row nine of a table nobody scrolled. And every report produced follow-up questions the analyst could not answer until the next day, because answering meant another export. The reporting was accurate. It was just too slow to change anything.

What got built

The system pulls from every platform an account runs on, reconciles the figures against a written rule for which source wins, and drafts the recap: what moved, by how much, the most likely reason, and the evidence behind that reason. Anomalies lead. If a channel dropped 40%, that is the first line of the report, not row nine of a table. Then the part that decides whether a reporting project survives its first month. Anyone can ask a follow-up in plain language and get an answer from live data without waiting on an analyst. That removes the bottleneck that kills most of these builds, where the report is automated but every question about it still routes through one person.

What changed

The weekly cycle went from about 9 hours to 45 minutes across 14 accounts, and the 45 minutes is review rather than assembly. Recaps land Monday morning instead of midweek, so decisions get made against a week that still has four days left in it. All 5 platforms reconcile automatically against a stated rule for which source is authoritative, which ended the recurring argument about whose number to quote in front of a room. The change nobody predicted was the follow-up questions. They used to cost a day each and now get answered in the moment, in plain language, against live data. The number of questions asked went up sharply, which is the point: cheap questions get asked, and the ones that used to go unasked are often the ones worth asking. Anomalies now open the report, so a 40% drop gets seen on Monday instead of found in a table three weeks later.

Why it matters

Most reporting automation stops at the artifact. You get a dashboard, or a deck that builds itself, and a month later people glance at it and move on, because it answers only the questions it was built to answer. The part that matters is self-service. When anyone can interrogate live data in plain language, reporting stops being a weekly delivery and becomes something people actually use on a Tuesday afternoon. The analyst stops being a lookup service and gets to do analysis. It is also the honest test of whether a reporting system works. Not whether the numbers are right, which is the price of entry, but whether anybody changes what they do because of them. Monday delivery with instant follow-up changes behavior. A perfect Thursday PDF does not.

Different problem, same shape.

What keeps not getting done, and what is it costing you? Two sentences is enough to judge fit.

Start a project