A good question to your agency data has four parts: a metric, a group, a time period and a sort order. "Which creator earned the least after their split in the last 30 days" contains all four. Questions that fail usually leave out the time period, which is the part that most changes the answer.
Being able to ask instead of clicking is the change that made agency management workable on a phone. It also introduces a new skill, which is knowing how to ask. Most people write a vague question, get a vague answer and conclude the tool is weak.
The four parts of a question that works
| Part | Weak | Strong |
|---|---|---|
| Metric | How are we doing | Net revenue after split |
| Group | Everyone | By creator |
| Time period | Recently | In the last 30 days |
| Sort order | Not stated | Lowest first |
Put together: "Show me net revenue after split by creator for the last 30 days, lowest first." That returns a table you can act on immediately.
Ten questions worth asking regularly
- Which creator has the worst margin after their split this month?
- Which subscribers are expiring in the next seven days?
- Which fans open messages but have never purchased?
- Which chatter has the highest revenue per conversation this week?
- Which creators have declined for three consecutive weeks?
- Who are our top twenty spenders in the last 90 days?
- What was our best performing message last month, by unlock rate?
- Which subscribers bought once and never again?
- How does this creator's week compare to their previous four weeks?
- Which segment has not been messaged in the last 14 days?
Those ten cover most of what an agency needs to know week to week. Saving them means the thinking is done once.
Why the time period matters most
Leaving out the time period is the single most common mistake. Without it the software has to guess, and the guess is usually a default that does not match your intent.
Worse, the answer still looks plausible. A revenue table with no stated period is not obviously wrong, so people act on it. Always state the window, and prefer short windows for operational decisions. Thirty days shows you what is true now. A year to date figure shows you history.
Follow up questions do the real work
The first question tells you where to look. The second tells you why. That pattern matters more than perfecting the first question.
For example: "Which creator declined most last month" gives you a name. "Show me that creator's revenue by day for the last 60 days" tells you whether it was a gradual fade or a specific event. "What did we send that creator's subscribers in that period" usually gives you the cause.
Three questions, thirty seconds, and you have moved from a number to an explanation. That is the workflow worth learning.
What to do when the answer looks wrong
- Check the period. Nine times out of ten the window was not what you assumed.
- Check whether the metric is gross or net. These differ a great deal once splits are applied.
- Check who is included. Cancelled, refunded and expired subscribers are handled differently by different definitions.
- Ask for the underlying rows. If a number surprises you, look at the records behind it before acting.
Write questions your team can reuse
If a question is worth asking twice, save it with a name that describes the intention. "Lapsing buyers" rather than "purchase in 30 no open in 7". Named questions make your reporting consistent across the team, which is what makes results comparable month to month.
Key takeaways
- A usable question has a metric, a group, a time period and a sort order.
- Missing time periods cause most wrong answers, and the answers still look plausible.
- Ask a follow up. The first question finds the problem, the second explains it.
- Prefer short windows for operational decisions.
- Save and name the questions worth repeating so the whole team asks them the same way.
Frequently asked questions
What makes a good question to ask your agency data?
Four parts: the metric, the grouping, the time period and the sort order. For example, net revenue after split, by creator, for the last 30 days, lowest first.
Why do I get an answer that looks wrong?
Most often the time period was assumed rather than stated, or the metric was gross where you expected net. Check both before questioning the data itself.
Should I ask one big question or several small ones?
Several small ones. The first question locates the problem, the second explains it, and the third confirms the cause. Complicated single questions are harder to verify.
