No one gets fired for a superficial concept. And that's the problem.

Cost used to do the hard part

Before anyone used the word “product” to describe it, building software was expensive. That expense did several of management’s hardest jobs for it:

• It forced scrutiny: nobody spends a million pounds without awkward questions, and without someone putting their reputation on the line (even in 2021)
• It forced a decision: a budget had to be approved or refused
• It rationed the pipeline: there was only so much engineering capacity to go around
• It signalled commitment: significant spend usually meant someone senior believed an idea was worth backing

None of this was designed as an innovation method - it was a byproduct of things being hard to make. Spend, finish and conviction often rose together, so people could read a degree of organisational commitment from how finished an idea looked. That was never a reliable measure of whether the idea was good, but it was a signal people learned to trust.

Cost was a crude gatekeeper - sometimes it killed good ideas for the wrong reasons. But it also meant that not every idea could become a convincing proposition, a prototype and a funded initiative.

Two things de-risk an idea

Strip most innovation methods back and they are managing the tension between two forces. Fidelity is how finished something looks and behaves. Exposure is how much it has been tested against the reality it must survive: customers, users, operations, partners, regulation and the economics of delivery.

An idea becomes less risky when fidelity and exposure rise together. It becomes dangerous when fidelity runs ahead of exposure: when the thing looks ready before anyone has established that it should exist, that people will use it, or that the organisation can deliver what it promises.

The methods developed over the last thirty years tried to keep those forces in step:

• Design thinking brought users and their needs into the earliest stages.
• Lean Startup favoured small tests, learning and iteration before scaling.
• Dual-track Agile kept discovery running alongside delivery.
• Design sprints compressed a cycle of making and testing to expose weaknesses quickly.

These methods differ, but share a premise: learn from the world before investing too heavily in the answer. High-fidelity work used to be expensive enough that organisations had to be careful about when they made it. AI has weakened that natural constraint. The methods have not become irrelevant; the discipline they ask for has become easier to bypass.

What AI has changed

Generative AI can produce a clickable prototype, a forty-page deck and a financial model in days (sometimes hours) that might once have taken months and significant budget. Fidelity arrives early. Exposure still moves at the speed of real people, real systems and real decisions.

AI can help us imagine how a market might respond, but it cannot establish how that market will respond. It can generate plausible customer objections; it cannot make a customer change behaviour, share sensitive data, adopt a workflow or pay for a product. The gap between a convincing representation and a tested proposition has widened.

That creates four problems:

• Ideas get ahead of their evidence. Polish is persuasive. A committee can mistake the most convincing concept in the room for the most promising one, even when the concept has little evidence behind it.
• Scarcity stops choosing for us. When building was constrained, some ideas fell away because there was no capacity to pursue them. Now the organisation must choose deliberately what to pursue, what to refuse, what evidence would change its mind and what would make it stop.
• The risk changes kind. A failed prototype may now cost very little. But a polished promise reaching customers, partners or a regulator before it is true, safe, fair or deliverable can damage trust. Governance designed mainly to prevent overspend will not catch every one of these risks.
• The bottleneck moves. The scarce resources are increasingly customer access, attention, decision-making capacity and the organisation’s ability to absorb change—not the ability to produce another concept.

The response is not to abandon design thinking, Lean Startup or agile discovery. It is to make their core discipline explicit: the next investment should be earned by evidence, not justified by how finished the current artefact looks.

A new way: steps of exposure

The old lifecycle often moved through steps of fidelity: idea, concept, product, live. In practice, the cost of each step imposed a degree of consequence. The new lifecycle needs to create those consequences deliberately, before a team has built too much to change course.

Exposure should mean observable evidence about the assumptions that matter most—not simply that someone outside the team has seen a demo. Evidence should become stronger as commitment grows. Progress is earned, not automatic: at each step, proceed when the evidence supports the case, adapt when assumptions fail, or stop when the case no longer holds.

1. Find the pain.
Use a sketch, a short proposition or a conversation to establish who experiences the problem and what they do about it today. Look for concrete examples of behaviour, workarounds and consequences—not just expressions of interest.

2. Get commitment.
Put a rough proposition in front of real people and ask for a meaningful commitment: time, access to a real task or data, a change in workflow, or payment where appropriate. Treat stated enthusiasm as a clue, not proof.

3. Prove it works.
Test a working version on a real task with realistic conditions and data. Look for whether it does the job, whether people trust its output, and what fails outside a controlled demo.

4. Make it ours.
Test whether the organisation can operate and support the proposition at an acceptable cost and risk. Bring in the people responsible for delivery, security, legal and regulatory requirements, and ongoing support.

5. Earn the scale.
Expand only when evidence shows that the proposition creates value in practice. Track meaningful use, outcomes, retention where relevant, and the economics of delivery; continue watching for unintended consequences.

These are not universal stage gates or a requirement that every idea follow the same sequence. The right test depends on the proposition and its risks. A consumer subscription, an internal tool and a regulated service will need different evidence. The common rule is simpler: each increase in commitment should be supported by evidence proportionate to the risk.

At every step, define in advance what result would justify continuing, what would trigger a change and what would cause the team to stop. If there is no clear answer from the outside world, the default should not be to keep going. Momentum is not evidence, and neither is the quality of the deck.

A practical example

Imagine a company developing an AI assistant for procurement. Within two days, a team can produce a polished interface, a working demo, a financial model and a compelling presentation. Senior stakeholders are impressed. Yet the team may not know whether procurement staff trust the recommendations, whether the assistant works with real data, whether security will approve it or whether anyone will change an established workflow.

A steps-of-exposure approach would resist polishing the demo further just because it looks promising. First, the team would examine how the problem is handled today and whether it is costly or painful enough to matter. Next, it would test a rough version with real users and look for behaviour: will they use it on a meaningful task, challenge its output, or change how they work? Then it would test the data, security, operational and compliance assumptions before expanding the pilot.

The result might be a stronger proposition or an early stop. Both are useful outcomes. The point is not to slow down making the prototype. It is to avoid confusing the prototype with proof.

What it asks of leaders

• Ask for evidence from outside the team at every decision point, not just progress reports and demonstrations.
• Cap the number of live bets by the number of meaningful real-world tests the organisation can run - not by how many concepts AI can generate.
• Make stop conditions explicit before teams become attached to an idea.
• Treat an early, cheap stop as a result worth having: it returns money, time and attention to the organisation.
• Separate exploration from commitment. Let teams make and test things quickly, while requiring stronger evidence before funding scale, making public promises or exposing customers to risk.

Create the right friction

The answer is not more bureaucracy. Extra approvals, longer templates and more meetings can slow a team down without teaching it anything. The useful distinction is between friction that consumes effort and friction that creates learning.

Remove friction from making things. Add friction to committing to things.

Use AI to generate competing ideas, explore alternatives, build rough prototypes and prepare for sceptical questions. Use it to help design tests and interpret what you learn. But do not let a simulation stand in for the market. A simulation has nothing at stake; a real user does.

The most valuable tests often reveal details that were never captured in the brief: a workaround people rely on, a trust concern they struggle to articulate, a constraint hidden in an existing process. Those details - the texture of real behaviour - can expose the gap between a plausible concept and a worthwhile proposition.

Where this leaves us

No one gets fired for presenting a superficial concept. In many organisations, a polished concept is safer to champion than an incomplete experiment that might reveal an uncomfortable truth. It looks like progress, is easy to communicate and allows everyone to remain optimistic for a little longer.

AI has made that habit more consequential. A concept can now be finished, funded and shown to customers before anyone has established whether the proposition is wanted, viable or deliverable. When it fails, the damage may land on trust and reputation rather than just the budget.

The question to ask of any new idea is no longer how finished it is or how much it cost. It is: what has the outside world told us that we did not already believe?

If the honest answer is nothing, it remains a superficial concept, however convincing it looks.

Everyone is talking about removing friction. I think we need to create more of it.