How to Ask Your Own Data a Question

You want to know which clients have reduced their spend compared with last year and what they have in common. The answer is in your systems. Getting it requires somebody to pull three exports and spend a morning in a spreadsheet.

So the question does not get asked. Not because it does not matter, but because it costs a morning and you are not certain the answer will be interesting.

The cost of asking decided what you knew

This is the part worth understanding, because it explains why most businesses know so little about themselves.

When each question costs half a day, only questions somebody is already confident about get asked. Exploratory ones, the ones that begin with I wonder whether, never clear the bar. And those are exactly the questions that produce findings, because the ones you are confident about mostly confirm what you thought.

So reporting settles into a fixed set of numbers produced monthly, and everything outside that set is unknown, permanently.

What changed

AI is what changed it. Asking in plain language now works against your actual systems, because a model can translate a question into the query somebody used to have to write, and the cost of asking fell to roughly nothing.

That sounds like a convenience and it is a change in what a business can know. When a question costs four seconds, you ask the follow-up. And the follow-up is where the finding usually is: not which clients reduced spend, but what those clients have in common, and whether it started before or after the thing you changed in March.

Three or four follow-ups deep is territory no business ever reached with a monthly report, because nobody was going to spend two days on a hunch.

Ask better questions

The quality of the answer depends almost entirely on the question, and most people start too broad.

Ask about a group, not a total. Not what is our average project value, but how do projects over $50,000 differ from the rest in how long they take.

Ask for a comparison. Almost every useful business question is a comparison: this year against last, this branch against that, won deals against lost ones. A single number rarely tells you anything on its own.

Ask what changed rather than what is. The current state is usually less informative than the direction, and direction is where decisions live.

Then ask why. The first answer tells you something happened. The second and third tell you what to do, and that is the part that used to be unaffordable.

Check the answer before you act on it

An important caution, because a fluent answer is persuasive whether or not it is right.

Any AI system answering questions about your business should show you what it counted: which records, which dates, which definition it used. Revenue means different things depending on whether it is invoiced or collected, and an answer that does not say which is not an answer you can act on.

Insist on that. A number with its working shown can be checked in thirty seconds. A number on its own has to be trusted, and trust is the wrong basis for a decision about pricing or headcount.

Anything surprising deserves one manual verification before it changes anything. Usually it holds. Occasionally it reveals that a field means something other than everybody assumed, which is itself the most valuable finding available.

What AI does not fix

The one thing this does not fix.

If your business has never decided what a customer is, or when revenue is recognised, or which jobs count as complete, then asking the question faster produces a fast, confident, contested answer. Sales will dispute the figure and finance will dispute it differently, and everybody will be right.

Settle the handful of definitions that matter first. It is an hour in a room and it is what makes everything downstream trustworthy.

Where this gets used

• Before a decision, instead of proceeding on impression. What actually happens to jobs of this type, rather than what everybody remembers happening.

• After a change, to check whether it worked, without waiting for a quarterly review.

• When something feels wrong. The instinct that a client has gone quiet or a service line has slowed is usually right and is now checkable the same afternoon.

• In the meeting itself. Settling a disagreement about a number while everybody is still in the room changes how those meetings run.

Who should be allowed to ask

A question worth settling early, because the instinct is to restrict access and it is usually the wrong instinct.

The value of cheap questions comes from many people asking them, and the people closest to the work ask the best ones. A branch manager wondering why their quotes take longer than another branch’s will find something no head office report was ever going to.

Two sensible limits rather than a general restriction. Personal data and payroll stay out of general access, for obvious reasons. And anything that will be used to make a decision outside the room gets checked before it travels, so a rough answer does not become a quoted figure.

Keep the questions that turned out to matter

The habit that compounds, and almost nobody forms it.

When a question produces something useful, save it and have it run on a schedule. It has gone from a one-off enquiry to a monitor, and the second time it answers you it will be telling you something changed rather than something is.

Businesses that do this end up with fifteen or twenty things being watched continuously, none of which anybody had to specify in advance. That set is far better than any dashboard designed up front, because every item in it earned its place by having been genuinely worth asking once.

Start with the question you gave up on

Every owner has two or three questions they wanted answered a year ago, were told would take a week, and quietly dropped.

Those are the place to start, because you already know they matter and you already know nobody could answer them. Write them down as they were originally asked, in your own words. That list is a better specification for what to build than any requirements document.

AI Optimize connects the systems that hold the answer and shows the working behind every number, so a question costs seconds and can still be checked. That work sits under Reporting & Data.