Numerical Questions to Ask
Questions for the moment someone asserts a number without showing it: a status meeting, a vendor pitch, a research review, a board paper. They ask for the denominator, the baseline, the sample, the spread, what is counted and excluded, and the figure that would change the decision. Written to be said out loud without sounding like an interrogation.
The questions
Open any question for the note
What's the actual number behind that?
Why ask it
Words like most, significant and doubled survive because nobody asks. The answer is often smaller than the phrasing suggested, and occasionally there is no number at all, which is the single most useful thing you can learn in a meeting.
Out of how many?
Why ask it
The denominator is the most commonly omitted part of any claim. Forty percent of five customers and forty percent of five thousand read identically in a slide deck and mean nothing alike. Ask before you react to the percentage.
Is that the mean or the median?
Why ask it
Averages of salaries, response times and deal sizes are pulled upward by a few large values, so the mean often describes nobody. If the answer is that they are not sure, the figure was almost certainly copied from somewhere rather than calculated.
What does the spread look like: what are the best and worst cases in there?
Why ask it
A single number hides whether the situation is uniform or bimodal. An average handling time of four days can mean everything takes four days, or half take one and half take a week, which calls for two different responses.
What are we comparing this to?
Why ask it
Any change needs a baseline, and a poorly chosen one flatters the result. Watch for comparisons against the worst month on record, against a forecast rather than actual history, or against a period that has been redefined since.
How does it look against the same period last year?
Why ask it
Year on year strips out seasonality that makes month-on-month movement look like performance. If the comparison is unavailable, ask whether the data goes back that far, since a short history is itself an important limitation.
How big is the sample?
Why ask it
Small samples generate swings that look exactly like trends. Also ask how the sample was chosen, because a survey of the people who replied describes the people who reply, not the population.
Where did this number come from, and who put it together?
Why ask it
The chain of custody matters. A figure from the finance system, a figure from a vendor's marketing page and a figure someone estimated in a hurry deserve very different weight, and all three look the same on a slide.
What's counted in this, and what's left out?
Why ask it
Definitions do the quiet work. Active user, incident, complaint and churn are all defined by someone, and the definition often changed the last time the number needed to improve. Ask when the definition last changed.
When was this last measured?
Why ask it
Numbers get repeated long after they were true, especially in decks that are reused. Ask for the date of the underlying data rather than the date of the slide, and watch for figures that have been cited for years.
How often does this happen, per week or per thousand?
Why ask it
Converting a total into a rate makes it comparable and often deflates it. Three hundred errors sounds alarming until it is three hundred out of two million, and trivial until it is three hundred out of four hundred.
How much of this movement could just be noise?
Why ask it
Normal variation gets read as a result, and someone gets congratulated or blamed for it. If nobody knows how much the number bounces week to week, that variability is the first thing worth measuring.
If the trend holds, what's the number in six months?
Why ask it
Extrapolating out loud exposes assumptions people would not state directly. It also reveals absurdities quickly, since many growth claims imply reaching more customers than exist within two years.
What would this number have to be for us to decide differently?
Why ask it
This turns a debate about data into a decision about thresholds, and it is the fastest way to end a long meeting. If no value would change the decision, the analysis is decoration and can be stopped.
What's the cost per unit once the parts people forget are included?
Why ask it
The overlooked items are usually staff time, support load, migration, licensing and the cost of the thing being replaced twice. Ask who built the estimate and whether the operational team saw it.
What's the breakeven, and when do we reach it?
Why ask it
A payback period gives you the size of your margin for error. A breakeven that arrives in year four for a two-year technology cycle is a no, without needing an argument about whether the forecast is optimistic.
If we hit the target, what does that get us in money or hours?
Why ask it
Translating a metric back into the thing anyone cares about often shows the target is not worth the work. Answers that stay in the language of the metric usually mean nobody has done the conversion.
Where does this stop scaling?
Why ask it
Every process has a ceiling: a rate limit, a hiring pipeline, one person who knows the system, a supplier's capacity. Naming the binding constraint early prevents plans built on doubling something that cannot double.
What result would tell us we're wrong?
Why ask it
Asking for the disconfirming number tests whether this is analysis or advocacy. If no possible result would count against the plan, treat the numbers as illustration and make the decision on other grounds.
Who owns this number, and what happens if it's off by twenty percent?
Why ask it
Two answers together: whether one identifiable person is accountable, and how fragile the plan is. If a fifth of error breaks the case entirely, the case needs a better number before anything else is discussed.
Asking for numbers without derailing the room
Practical guidance for the conversation itself
How to ask so you get an answer
Ask for one number, not an audit
Six challenges in a row reads as an attack and the presenter stops volunteering anything. Pick the single figure the decision rests on and ask about that one properly.
Aim at the number, not the person
What is the denominator here lands very differently from how did you calculate that. Naming the figure rather than the author keeps the presenter working with you on it.
Accept I do not know as a good answer
If admitting a gap is treated as a failure, people fill gaps with guesses next time. Ask when they could find out, and write down who is checking.
Ask in the order that saves time
Denominator, then definition, then source, then baseline. Most weak numbers fail at the first two, and there is no point discussing a trend in a metric nobody has defined.
Where numbers usually go wrong
- Percentages of percentages. A rise from 2 percent to 3 percent is one point, and also a fifty percent increase. Both are true and one is chosen for effect.
- Shifting definitions. If active user or resolved ticket was redefined last quarter, the trend line spans two different measurements.
- Survivorship. Averages over customers who stayed, projects that finished or funds that still exist omit exactly the cases you were worried about.
- Seasonality. January against December tells you about the holidays, not about the business.
- Rounded certainty. A figure quoted to two decimal places from a sample of thirty implies precision the data cannot support.
- One number for two things. Cost savings and revenue growth combined into a single benefit figure hides which half is speculative.
What to do with the answers
- Write the number, its denominator, its date and its source in one line. A figure without those three attachments is not reusable.
- Do the arithmetic in the room. Dividing the total by the population takes ten seconds and settles most arguments.
- Record the threshold that would change the decision, then check against it later. That single note makes the next review much shorter.
- Note who agreed to find each missing number, and by when.
- Ask for the raw extract, not the chart, if the decision is large. Most surprises show up in the rows that were filtered out.
When no number exists
Sometimes the honest answer is that nobody has measured this. That is workable if it is said plainly: agree on a rough order of magnitude, note the assumption, and decide whether the decision can wait for a real measurement. The failure mode is a placeholder estimate that gets copied into a plan, then into a board paper, and returns six months later as an established fact with no source attached. If you invent a figure to keep moving, label it as an estimate in the document itself.