Scheduled tasks: the search report that writes itself

There is a recurring block on your team calendar called something like search review. It survived four weeks. Now two people decline it every Monday and the third opens the dashboard, scrolls, closes it again.

The thing that actually stuck in our beta was not a clever prompt. It was the report nobody had to ask for. A scheduled task runs the weekly check and delivers the written answer before Monday's first coffee.

Setup is one sentence. What changes is bigger: how your whole team relates to its own numbers, starting the first Monday it lands.

A scheduled task is a standing question

You phrase the ask once and it keeps answering.

From then on the assistant runs the same pre-built analysis on schedule and writes the result as a short answer with a direction word on it. Not a dashboard export. Freezing the question is what makes the answers comparable week to week, and a pile of ad-hoc checks never gets there no matter how many of them you run.

Where it lands matters as much as when. A result that lands where your team already talks gets opened. One that lands in a reports folder joins the graveyard.

The sentence is worth crafting, because every ambiguity you resolve at setup is an inconsistency your reports never develop. Four decisions, made once:

1. Name the check you want run.
2. Pin the scope, so the report cannot drift to a different segment.
3. Pick the schedule.
4. Choose where the answer lands.

Every run brings its own checks

A scheduled report inherits the discipline of whatever it runs, and it cannot shed it.

The weekly check arrives with its significance test armed, a statistical check on whether a change is bigger than the normal week-to-week wobble. It arrives with its scope stated and its as-of date printed, the date the data was last complete. Nobody relaxes the standard at eight on a Monday morning, because nobody is there to relax it.

That is the quiet advantage over the recurring report you build by hand, which sheds its caveats under time pressure every single time. The automated one physically cannot.

The as-of date earns double on a schedule, because the report speaks at a fixed hour about data that keeps its own clock. The card says which morning it means, so nobody mistakes Monday's answer for Monday's data.

Pinning the scope does the same job across weeks that it does across teammates. Your report cannot quietly drift to a different segment, because somebody wrote the segment down once and nobody re-decides it.

Why arriving unasked changes everything

Because asking is a performance, and busy weeks skip performances.

When somebody has to request the report, checking becomes a small demonstration of diligence, and the week you are underwater is the week it quietly does not happen. When it arrives on its own, your team consumes an answer instead of performing vigilance, which means the weeks nobody would have thought to check are exactly the weeks it still turns up in the channel.

The daily version is a standing check like the share comparison below, running whether or not anyone remembered to care that day. The workflow card is a worked example of one question answered end to end, with the clicks it replaces.

There is an archive effect too. A year of identical weekly answers is an honest history of your program, written as it happened, immune to the smoothing memory applies afterwards.

Leadership feels the difference inside a month. Nobody asks how's search doing downward any more, because the answer already arrived sideways, on schedule, with its evidence attached.

The workflow that does this: Daily market share vs competitors →

Escalation stays human

The scheduled task's job ends at the answer.

When the direction word comes back as slipping, a person decides whether to pull the deeper thread, and the chain is the one the Monday check article describes. Answer first, diagnosis second, in that order.

Automating the schedule without automating the judgment is the design, and the split is deliberate. The machine holds the schedule. You keep the decisions.

The report writes the handoff for you. A slipping answer names the next question worth asking, so your escalation starts one message deep instead of at a blank prompt.

See also: what the one weekly report should contain →

Start with one, resist the fleet

The failure mode of scheduled reporting is abundance. Ten automated reports nobody opens rebuild the dashboard graveyard with extra steps and a nicer font.

So start with the weekly check. Add a second only when somebody asks for its contents twice in a month. That threshold sounds arbitrary and it is roughly right, because a question asked twice is a question, and a question asked once is a mood.

A useful test for the second report: would its slipping answer change what anyone does that week? If not, it is a dashboard in an envelope. One report read every week beats five reports skimmed never.