The Gaussian Challenge and AI

Why the leaders getting the most from AI are not doing things faster. They are doing different things.

Last week, I wrote about the leaders who use AI mostly for the work that matters least: inboxes, email polish, meeting summaries, and, scarcely, the strategy work that actually pays off. I called this the Candy Machine Trap. The question that followed was what the leaders who escape it are doing instead.

The answer, in my own working life, has been one specific question, learned fifteen years ago in an Executive MBA classroom, that has done more for my thinking about AI than any other technique. It came wrapped in two stories, an old one about a schoolboy in Brunswick, and a more familiar one about Canon and Xerox. I will tell both, then name the question.

A boy in Brunswick

In the 1780s, a schoolteacher in the German town of Brunswick set his class the task of adding all the whole numbers from 1 to 100. He was expecting it to keep them quiet for the rest of the lesson. A boy of about eight, Carl Friedrich Gauss, returned the answer almost immediately. He had noticed that the numbers could be paired from the outside in: 1 and 100, 2 and 99, 3 and 98, and so on. Each pair summed to 101, and there were 50 such pairs. The answer was 5,050.

He had worked it out faster than the rest of the class because he had not tried to work harder at the obvious approach. He had seen a different shape in the problem.

Canon’s question

Two centuries later, Canon entered the US photocopying market against Xerox, which owned the technology, the brand, the corporate accounts, and the dominant business model: large centralised machines, leased rather than sold, supported by a service contract that paid Xerox month after month and gave it a tightly defended revenue base.

The conventional move for a smaller competitor was to build a slightly better Xerox machine and try to sell it the same way. Canon did not do that. Instead, the company asked three questions about the market that, taken together, threw away the assumption that the big machine on the floor of the corporate mailroom was the product at all.

Who was the buyer? Not the corporate procurement office, on whom Xerox had built its sales force, but the departmental manager who ran a finance team or a marketing function and wanted a copier they did not have to share with the rest of the building.

What was the product? Not a leased industrial machine, but a smaller, simpler one, sold outright at a price a department manager could sign off on without escalation.

How would it reach them? Not through a direct sales force, but through a retail dealer network that Canon could build cheaply, whereas Xerox could not match without dismantling its own network.

The answer was a different copier, sold to a different buyer, through a different channel. Xerox could not respond easily because the obvious response would have cannibalised the leasing revenue that paid for everything else. Canon, like Gauss, had not improved at the conventional game. They had changed the shape of the problem. It is the move I keep coming back to, fifteen years on, whenever a senior team I am working with is about to use AI to improve their conventional approach.

The question, named

Where conventional AI speeds up conventional answers, the Gaussian Challenge changes the question AI is being asked. AI makes the conventional answer cheap. It does not make the choice of question cheap. That choice is where the value moves.

Before you optimise a solution with AI, ask whether a different approach makes optimisation unnecessary. That is the question I learned in an EMBA classroom long before generative AI existed, and it is the one most worth carrying into the working week.

The Gaussian Challenge: a three-panel before/move/after triptych.

Left: conventional AI use speeds up the process of arriving at the conventional answer.

Centre: the move reframes the question (Jobs To Be Done, invert the speed, unlimited-resource threat, swap the vantage point).

Right: AI is asked a different question, producing a Gaussian answer.

What it looks like at work

Consider a senior team I observed recently in an international executive offsite. They sat down on a Tuesday afternoon to do their quarterly strategy review. They had used AI to draft the executive summary, generate three scenario analyses, and produce a polished slide deck. The whole exercise took an afternoon; a year ago, the same review would have consumed the better part of a week. The output was fluent, well-structured and comprehensive, better by any reasonable measure than what they would have produced by hand.

So what was the problem? They produced the same document they always do. They asked the same questions, applied the same frameworks, and arrived at the same categories of answer. The strategic thinking behind it had not moved at all. They had got faster at the conventional game.

The Gaussian move would have been to step back and ask what decision the review was meant to inform, then to ask AI to help find the shortest path to that decision. Perhaps the answer was not a strategy document at all. Perhaps it was three specific questions that needed fresh data, a ten-minute call with a customer, and a one-page decision memo. The strategy deck was never the point; the decision was.

A second example. A head of learning at a large firm was redesigning the organisation’s management development programme. AI-generated session plans, learning outcomes, case studies and facilitator guides were produced at speed. Within a fortnight, the team had a revised curriculum that was polished, coherent, and structurally indistinguishable from the one it was replacing. Faster, cleaner, broadly the same.

The Gaussian move would have begun earlier and elsewhere. The question is not how to use AI to redesign a programme, but what the managers in question are failing to do, and what the shortest path is to them doing it. Put that way, the honest answer might be that another three-day workshop is unlikely to change behaviour, whereas a series of short, AI-supported coaching conversations within the flow of work would. The programme is not the goal; changed behaviour is. AI then becomes the tool that makes a different architecture feasible, rather than the tool for producing a glossier version of the old one.

Both examples point to the same pattern.

“Conventional thinking, accelerated, produces conventional outputs, accelerated.”

The Gaussian move is what changes the output itself.

If AI has commoditised the conventional, then seeing the problem in a different light is the only part of the work that is not on its way to becoming free. That is where the returns will concentrate, and the people who earn them are those who have made a habit of asking a different question.

Reframed questions

What does the practice look like in a working week? It is less a technique than a habit, and here are four reframings that tend to produce Gaussian moves, and they are simple enough to build in. The first, the Jobs-to-Be-Done question, is the scaffold the other three sharpen: ask it well, and the rest of your prompting changes shape.

Ask the Jobs-to-Be-Done question. What job is this prompt being hired to do? Strip the prompt back to the underlying job, and the space of acceptable solutions widens immediately. The seemingly innocuous ‘draft a board paper on our AI strategy’ becomes ‘give the board enough to decide whether to back a different strategy than the one we are currently running’. AI’s response changes shape, and so does the conversation it sets up in the room. The framing comes from Clayton Christensen (2003): customers do not buy products; they hire them to do a job. Applied to your prompt, the same move asks what job your output is being hired for, and the question of how to write it often becomes the easier question of whether to write it at all.

Invert the speed. If AI were making this work ten times slower rather than ten times faster, how would you approach it differently? Inversion is deliberately uncomfortable. It strips away the technology’s seductive appeal and forces the underlying judgement to surface. Habits that depend on AI’s speed fall away; those that would survive without it usually hold the strategic value sits, and therefore are where your time is most worth spending.

Run the unlimited-resource threat. Imagine a new entrant attacking your position with unlimited compute, no legacy systems, and no team to protect it. What would they build instead of what you are about to build? This is the incumbent’s most useful prompt. It surfaces the parts of your current approach that exist because of your team size, budget, or legacy systems, rather than because the problem itself demands them. The same prompt, when applied to a stronger competitor, a regulatory shift, or a customer leaving the category, does similar work; the unlimited-resource version removes the most common alibi.

Swap the vantage point. Restate the problem from four professional viewpoints: an analyst, a designer, a regulator, and an anthropologist. When two of those professionals describe the same situation in different terms, the gap between their descriptions is where the reframe usually lies. The analyst’s ‘efficiency problem’ is the designer’s ‘experience problem’, which is the regulator’s ‘accountability problem’. The Gaussian move starts by picking the framing that the team is not using.

Once a reframing yields an output that diverges from the conventional one, the leader compares the two side by side and chooses which to commit to. The reframing is the input; the advantage is captured in the choice between outputs.

Why does the default pull the other way?

This is harder than it sounds, and it is worth being honest about why. Three cognitive biases pull most of us towards AI-as-accelerator and away from AI-as-reframing-tool. Anchoring locks us into our first framing of a task, so the prompt we type is usually a description of the version we were already going to produce. Satisficing stops us from looking once we have an answer that is good enough, which AI provides in seconds. Availability makes the approaches we have used before more readily accessible than those we have not, and AI, trained on what has already been done, amplifies the effect rather than countering it.

Left to its own defaults, AI reinforces these biases more than it challenges them, because it is optimised to give you a fluent answer to the question you asked. It will not tell you that you are asking the wrong question unless you ask it to, and even then, only sometimes. The biases that work against the Gaussian move are the subject of the next article in this series. For now, the point is that the four reframings do not occur on their own. Something has to make them habitual.

When not to ask

The Gaussian move does not apply to every task. A great deal of work is maintenance: regulatory returns, status reports, weekly updates, and routine correspondence. Here, the conventional approach is the right one, and doing it faster with AI is a straightforward win. A significant share of the AI productivity gains organisations will see in the next two or three years will come from work of this kind, and that is welcome. It would be a mistake to ask the Gaussian question of every email.

The framework applies to the work that matters: the decisions that shape strategy, the programmes that change behaviour, the investments that set direction for years, and the hires that change a team. For those tasks, the risk of doing the conventional thing faster is locking in a choice that a different question would have unravelled. The Gaussian Challenge is a filter applied to the small share of work where framing matters, which is the work that most repays the time.

The discipline

The four reframings are the move. The discipline is making them habitual, the first thing you do with AI rather than the thing you do once a quarter. Three habits embed the reframing move.

The pre-prompt pause. Before you type, ask what decision or outcome the work in front of you is meant to produce. The task is almost always a means to something. You are not writing a report; you are helping someone decide whether to invest. You are not redesigning a programme; you are trying to change behaviour. Once the real outcome is on the table, the task you were about to type into AI often looks narrower or broader than it needs to be, and sometimes redundant altogether. Naming the real outcome is what triggers the reframings.

The reframe-before-refine rule. When the first AI output is good but unsurprising, the temptation is to iterate on it, tightening the language, sharpening the recommendation, and polishing the slides. Instead, re-prompt with a different reframing. Iteration sharpens; reframing changes shape. The output you are about to refine may not be the output worth having.

The Gaussian-gap log. After each session that mattered, capture in one line the conventional answer, the reframed answer, and the size of the gap. The log is not a record; it is a decision input: at strategic decision points, the leader re-prompts AI with the reframed question, compares the two outputs side by side, and commits to the reframed path or stays with the conventional one. Over a quarter, the log also surfaces the types of decisions where the gap is consistently large, which indicates where to invest the practice next.

Gauss did not work any harder than the other children, and Canon did not build a better Xerox. The leaders who get the most from AI are those who have learned to do the same in their own work, one prompt at a time, by being willing to discard their first answer.

Where this has been tested

The Gaussian Challenge has been tested in three-hour sessions with chairs and governors in further education, deep-tech founders at the Royal Academy of Engineering, and manufacturing leaders in the UK government’s Made Smarter programme; it forms a design backbone of our COO, CDO and CPO open programmes.

For teams: the Gaussian Gap exercise

Take a real business problem your team is currently working on. Something with a deadline, something that matters.

First, use AI to produce the best possible conventional answer. Give it context, refine the output, iterate until you have something polished. Set it aside.

Then start again from scratch. This time, do not ask AI to solve the same problem better. Ask it to help you find an approach that would make your first answer irrelevant. What job is this work being hired to do? If AI were making it ten times slower, what would you change? What would a new entrant with unlimited compute build instead? How would a colleague from an unrelated profession approach the problem?

Compare the two outputs. The distance between them is your Gaussian gap: the measure of how much value you are leaving on the table by optimising within your current frame rather than questioning the frame itself.

“The Gaussian Challenge changes which question AI is being asked, not how fast AI answers it.”

About the author

Viren Lall FRSA is Managing Director of ChangeSchool LDN, an executive education design and delivery company that has worked with over 6,900 leaders and entrepreneurs across 39 countries. His current work focuses on how senior leaders use AI to improve judgement, allocation and organisational learning.

Viren Lall, Managing Director, ChangeSchool LDN (2026). virenlall.com/gaussian-challenge-ai