The Question Is the Scarce Asset Now

For decades the answer was the expensive part. AI made it cheap — and moved the bottleneck somewhere most tools still aren’t looking.

Everyone who has worked with data learned the same quiet habit: don’t ask the expensive questions. Not because they didn’t matter — because each one cost a ticket, an analyst, a week of waiting, and the decision usually couldn’t wait that long. So the question went unasked, and the call got made on the most confident guess in the room. You learned the price of a question and stopped reaching for the ones you couldn’t afford.

That price is what changed. The cost of producing an answer — the analysis, the query, the synthesis a trained person used to perform — has collapsed toward zero. A model can now decompose a question, work the data, and return a reasoned result in the time it takes to read this paragraph. The thing that was scarce for the entire history of knowledge work, the answer, just became abundant.

When something abundant arrives, the interesting question is never the abundant thing. It’s what got scarce in its place. Economics has a tidy law for this: when a resource becomes plentiful, its value falls and the value migrates to its complement. Water is precious in a desert and worthless in a rainforest. Answers are becoming rainforest water. Their complement — the question — is where the scarcity, and the value, are moving.

The complement nobody priced

The observation isn’t new, and it isn’t ours. “A good question is worth a million good answers,” Kevin Kelly wrote in The Inevitable, his 2016 book on where technology was heading — because, he argued, answers were on their way to becoming cheap and plentiful while good questions only grew more valuable. He was describing a future. That future is now the present tense — the answers turned cheap, exactly as predicted, and the value is migrating to the half of the workflow almost no software was built to serve.

Consider what a question actually requires once answering is free. It requires knowing what to ask — which cut of the problem matters, which follow-up the last answer just earned, when a clean number is hiding a dirty assumption. None of that is the answer. All of it is the question, and it’s the part that doesn’t commoditize, because it depends on context a model doesn’t have: what this team is deciding, what it already tried, what would actually change its mind.

So the scarcity didn’t disappear when the answer got cheap. It moved one step upstream, from producing the answer to forming the question. And almost every tool in the analytics market is still pointed at the step that just stopped being hard.

Why the tools are aimed at the wrong half

This isn’t an oversight. It’s inheritance. For most of computing history the answer genuinely was the bottleneck, so every tool was built to compress it. The dashboard pre-computes answers so you don’t have to wait for them. The BI platform speeds up the analyst. The newest wave of “ask-AI” boxes generates an answer faster than ever. All of them optimize the same thing: the production of the answer. They were right to, for as long as that was the scarce part.

But a tool that ends at the answer hands you back the hardest remaining work. It assumes the answer was the finish line, when the finish line was always the decision — and you only reach a decision through a chain of questions, each one provoked by the last answer. The dashboard, the chart, the single-shot AI reply all stop exactly where the now-scarce work begins. They deliver the cheap thing beautifully and abandon you at the expensive one.

You can see the tell in how the market talks about itself. The pitch is almost always “answers, faster.” Faster answers were the right promise for the old scarcity. Against the new one, they optimize the half that’s already solved and stay silent on the half that isn’t.

What it looks like to serve the question

A system built for the new scarcity treats the answer as a step, not a destination. It expects the next question, because the next question is the point. Say you ask why one cohort of signups is converting worse than the last; it answers, and the answer is interesting precisely because it raises the next thing — was it the channel, the onboarding change, the pricing test that shipped that week? Each answer is cheap, so you keep going, down the thread, until you reach the thing you can actually decide on. The product isn’t the answer. The product is the ability to keep asking.

That is a different shape from a faster dashboard, and the difference isn’t speed — it’s where the system thinks the work ends. A reporting tool ends at display. An answer engine ends at the reply. A system built for the scarce asset ends where you do: at the decision, several questions later, none of which anyone pre-built a view for.

Picture that same question — why one cohort converts worse than the last — playing out under each model. Under reporting, you find the conversion line on a dashboard, see that it dipped, and stop — the board has no panel for “why,” so the why becomes a ticket and the meeting moves on without it. Under a single-shot answer engine, you ask “why did it convert worse,” get a competent paragraph naming three plausible causes, and stop — because now you have an answer, and the format quietly signals you’re done. Under a system built for the question, the three causes aren’t a conclusion; they’re a fork. You ask which one actually moved the number. It was the onboarding change. You ask which step. The third. You ask whether it hit every segment or one. One — the segment you’d just spent a quarter acquiring. Now you have something to decide. Same starting question, three different stopping points, and only the last one reaches the place a decision can be made. The first two weren’t wrong; they just ended early, at the answer, which is the cheap part.

What that scene shows is that “answer” and “decision” were never the same object — they only looked alike when answers were expensive enough to feel like the destination. Strip the cost away and the gap between them becomes the visible work. That gap is made of questions.

The pushback writes itself: but the AI gives me answers now — isn’t that the whole point? It was never the point. The answer was the cost, not the prize. The prize was the decision, and the decision lives at the end of a chain of questions a single answer can’t carry you through. Make answers free and you haven’t finished the job; you’ve revealed which part of it was always the real work.

A sharper objection runs underneath that one: the answer isn’t really cheap, because a trustworthy answer over a company’s own data — correct, grounded, properly attributed — is still hard. That’s true, and it’s the right thing to insist on; an answer you can’t trust isn’t cheap, it’s worthless, and a chain of questions built on bad answers just compounds the error. But it doesn’t move the scarce thing back to the answer. It raises the floor under both halves. Once trustworthy answering is in place — and that is exactly the engineering worth doing — the binding constraint on what a team actually learns is still which questions it thinks to ask. Reliable answers are the price of entry. The question is what you do with them, and it’s still the scarce part.

The market that follows from this

If the answer is cheap and the question is scarce, the next cycle of analytics tools won’t be won on better-looking answers. It’ll be won by whoever serves the question — who makes asking the next one frictionless, who treats curiosity as the thing to sustain rather than a cost to ration. Incumbents optimized for answer production over two decades of engineering. That’s exactly the muscle that’s now abundant, and exactly the wrong muscle to have over-built. You don’t catch up to a question-shaped competitor by generating answers faster. You’re sprinting toward the cheap end of the workflow.

And the disadvantage is structural, not a matter of effort. A tool built to render pre-decided views is organized, end to end, around the assumption that someone already knows the question — that’s what lets it be fast and reliable at display. Serving the unanticipated question is the opposite assumption: the system can’t know in advance what it will be asked, so it has to be built to reason from the question outward rather than retrieve toward a known view. Those are different foundations, not different feature sets. Bolting a chat box onto a reporting tool doesn’t change which assumption the foundation was poured around. That’s why the move tends to come from systems built for inquiry from the start — they were organized around the half that turned out to be scarce.

None of this means answers stop mattering — a question with no answer is just a wish. It means the binding constraint flipped. For a generation, the scarce, expensive, rationed thing was the answer, and we built an entire industry to make it cheaper. We succeeded. The scarce thing now is the question — and the tools that win next are the ones that finally point at it.

Key Takeaways

• For decades the answer was the expensive, rationed part of analytics. AI collapsed its cost — so the bottleneck moved, it didn’t disappear.
• When a resource becomes abundant, value migrates to its complement. Answers became abundant; the scarce asset is now the question.
• Forming the right question doesn’t commoditize the way producing the answer did — it depends on context a model doesn’t have: what you’re deciding and what would change your mind.
• Most analytics tools still optimize answer production — “answers, faster” — which is the half that just stopped being hard. A tool that ends at the answer hands you back the real work.
• The next cycle won’t be won on better answers but on serving the question: making the next one free to ask, all the way to the decision.