Do Not Start with AI in NPD. Start with the Decision.

Written By Escape Partner, Michael Sandfær

Many organisations begin their AI journey by asking where AI can be added to the product development process.

That is the wrong starting point.

The question may generate an impressive list of possibilities: market analysis, idea generation, forecasting, design optimisation, automated testing, risk identification and project reporting.

It may also generate numerous pilots without materially improving product development.

A more useful question is:

Which decisions repeatedly delay, weaken or derail our product development — and could AI improve the evidence behind them?

More information does not guarantee better decisions

AI can process information, identify patterns and generate alternatives faster than conventional approaches.

In the front end of innovation, it can analyse customer feedback, market developments and emerging technologies. During development, it can support design exploration, simulation and testing. At portfolio and gate level, it can consolidate evidence, test assumptions and highlight risks.

These are significant capabilities. But faster analysis does not automatically produce better decisions.

AI can also generate more noise, more alternatives and greater confidence in conclusions that remain uncertain. If an organisation has not defined the decision, the necessary evidence or the criteria for success, AI may simply accelerate an already weak process.

A weak business case does not become strong because it contains an AI-generated forecast. A gate review does not become more effective because the project team can produce more analysis. And a portfolio decision does not become objective simply because an algorithm provides a score.

Decision quality still depends on the underlying data, the assumptions being made and management’s willingness to act on the evidence.

The gap between experimentation and value

Research by Robert Cooper and Alexander Brem indicates that many companies are still struggling to move from experimenting with AI to using it systematically in new product development.

The constraint is often not access to technology. Their work points to familiar implementation challenges: unclear priorities, weak business cases, limited data readiness, insufficient management ownership, capability gaps and difficulty moving successful experiments into normal ways of working.

A systematic review by Vallé and colleagues provides a broader research perspective. It shows how AI can strengthen the way organisations collect, process and apply information throughout NPD. However, the outcomes depend on organisational and technical conditions, not on the technology alone.

The practical conclusion is less dramatic than many AI claims, but more useful: AI has considerable potential in NPD when it is integrated into better work systems and used to improve decisions that matter.

Begin with a recurring decision

Rather than launching a broad initiative to implement AI across NPD, identify one important decision that is currently slow, poorly informed or repeatedly disputed.

For example:

  • Do we have sufficient evidence that the customer problem justifies further development?

  • Which assumptions have the greatest influence on the business case?

  • Should the organisation continue, redirect or stop the initiative at the next gate?

These are not AI questions. They are product development and management questions.

Once the decision is clear, the organisation can assess whether AI can improve the evidence, reduce the analytical burden or identify something that people are currently overlooking.

In the front end, AI might help analyse a larger body of customer feedback. Management must still determine whether the identified need is strategically relevant.

In a business case, AI might test multiple demand, cost and timing scenarios. Management must still judge which assumptions are credible and which risks the organisation is prepared to accept.

At a gate review, AI might consolidate evidence and identify patterns from previous projects. The gatekeepers must still decide whether to continue, hold, redirect or stop the initiative.

AI can strengthen the basis for a decision. It cannot remove the need to make one.

Five questions before introducing AI

Before launching another AI pilot in NPD, leaders should be able to answer five questions:

  1. Decision: What specific decision or recurring workflow are we trying to improve? “Using AI in the front end” is too broad. “Improving how we validate customer demand before concept approval” is sufficiently specific.

  2. Evidence: What information does the decision require, and what is currently missing, slow or unreliable? This includes an honest assessment of data quality and uncertainty.

  3. Owner: Who owns the workflow, and who remains accountable for the decision? When the intended value lies in an NPD decision, ownership cannot sit solely with IT, a data team or an external supplier.

  4. Value: How will we know that the application has improved the work? Relevant measures might include reduced analysis time, earlier identification of critical risks, improved forecast accuracy or fewer late changes.

  5. Guardrails: Where are human judgement, review and approval required? This includes confidentiality, intellectual property, bias, traceability and responsibility for the consequences.

Together, these questions connect the technology to a real management problem. They also provide a basis for deciding whether an experiment deserves further investment.

AI makes governance more important

There is an understandable expectation that AI will make NPD processes lighter and faster. It probably will in some areas.

However, faster information flow increases the need for clarity about decision rights, criteria and accountability. When analysis becomes easier to produce, management must become better at distinguishing relevant evidence from plausible-looking output.

This is particularly important at gates and other major investment decisions.

The purpose of a gate is not to confirm that the project team has completed the required documents. It is to determine whether the organisation should continue committing resources under the current assumptions.

AI can prepare material, compare scenarios, test assumptions and expose inconsistencies. Over time, it may perform more of the analytical work currently completed by project teams and specialists.

But it cannot own the strategic choice or accept the organisational risk. That responsibility remains with management.

Let AI earn its place

The practical starting point is not an enterprise-wide ambition to implement AI across NPD.

Choose one recurring decision that matters. Define what better evidence would look like. Establish ownership and success measures. Test whether AI improves the quality or speed of the work. Increase the investment only as the value becomes demonstrable.

Some experiments should be scaled. Others should be stopped. That is not a failure of AI adoption; it is sound innovation governance.

AI should not be introduced into product development simply because it is available. It should earn its place by improving a decision that matters.

Sources and further reading

Escape Partner, Michael Sandfær
Next

A Project Delivers. An Organisation Learns.