How can product teams use AI to explore solution options without converging too early?

Generative AI can produce a page of product ideas in seconds. That volume can look like broad exploration, although many outputs may be variations of the same underlying solution.

The first convincing concept also creates an anchor. A team may spend its remaining time refining that example rather than asking whether another service mechanism, workflow or non-product intervention would address the need better.

Product teams need to learn how to manage the search itself. AI can widen the option space, but the problem, constraints, evidence and decision criteria remain with the team.

More generated options do not guarantee a wider search

A list of twenty concepts may contain one idea expressed in twenty forms. The wording, layout or feature set changes, but each assumes that the customer needs another interface and that the organisation should build it.

This matters because product discovery is concerned with finding an effective way to create customer and business value. The best response might involve changing a process, improving support, publishing clearer information, integrating an existing service or choosing not to build anything.

Generative systems can support divergent thinking by producing alternatives quickly. Research on generative AI and design also identifies a risk of fixation: exposure to generated examples can narrow later thinking or encourage premature commitment. The effect depends on the task and how the activity is designed, so neither benefit nor risk should be treated as automatic.

The practical question is whether AI helps the team consider materially different mechanisms, not how many cards it fills.

Product discovery already separates possibilities from commitments

Product teams have established ways to explore. Designers, engineers, researchers, operational specialists and commercial colleagues bring different perspectives. Teams sketch alternatives, investigate constraints, compare assumptions and decide which options deserve evidence.

No single participant holds the entire product context. Design, engineering, research and operations colleagues may each expose a different route.

AI can add another source of prompts and combinations. It does not have privileged knowledge of the customer need. Without relevant context, it may reproduce familiar patterns, ignore local constraints or invent evidence. It can also make weak concepts sound unusually complete.

Conventional discovery remains the source of customer contact, specialist judgement and accountable selection. AI changes the cost and speed of producing candidates, which makes discipline at the point of convergence more important.

Design a deliberate divergent-convergent cycle

First, freeze the frame for the exercise. State the user need, desired outcome, evidence already available and non-negotiable constraints. If those are unclear, return to discovery rather than using solution generation to hide the gap.

During divergence, ask for alternatives based on different mechanisms, such as prevention, self-service, assisted service, process change, partnership and no-build responses. Ask which assumption each option depends on.

Keep evaluation out of this stage. Early scoring pulls attention towards the easiest option to explain and may suppress incomplete but valuable alternatives. Team members should add ideas independently before reviewing the generated set, which reduces the chance that the first AI example defines the discussion.

Next, group duplicates and identify what is genuinely different. A change of colour, channel or wording may not constitute a new mechanism. Retain a manageable set that represents different assumptions and trade-offs.

Only then begin convergence. Agree criteria before scoring: strength of user evidence, likely contribution to the product goal, usability, feasibility, viability, accessibility, ethics, operational impact and risk. Not every criterion needs a number. The purpose is to make the reasoning inspectable.

Make convergence an evidence-led team decision

AI can organise a comparison or challenge missing criteria, but it should not decide which option wins. A ranking reflects the context, weights and evidence supplied to it. When those inputs are incomplete, a neat table creates false precision.

Ask specialists to challenge the shortlist. Identify which claims are observed, inferred or generated. Record why each option is selected, rejected or held for further evidence. If two options remain plausible, the next step may be a prototype, research session or technical investigation rather than more generation.

Inclusive exploration also needs attention. AI training data and prompts may favour familiar users and service patterns. Deliberately examine excluded users, assisted routes and accessibility constraints. Relevant human perspectives cannot be recovered simply by asking for more output.

Practical learning can ask teams to compare exploration strategies using the same evidence. The facilitator examines whether they widened the mechanisms, preserved the problem and made convergence criteria visible.

Example

A hypothetical product team wants to help small-business customers resolve failed invoice payments. It fixes the user need and constraints, then asks AI for options based on prevention, self-service recovery, assisted support and changes outside the digital product.

The team adds an operational option that AI missed and removes several cosmetic duplicates. A user researcher checks each option against customer evidence, while engineering and operations colleagues identify dependencies and service impacts.

The surviving concepts are compared against accessibility, likely customer value, cost, feasibility and product strategy. AI widened the initial search, but the team decides which options deserve prototyping and which need more evidence.

FAQs

• How many AI-generated product options are enough?

  There is no universal number. Stop counting outputs and ask whether the set contains genuinely different mechanisms, assumptions and trade-offs. More generation adds little once the next uncertainty is better answered by users, data or specialist investigation.

• Can AI help a team avoid design fixation?

  It can prompt alternatives when the team deliberately requests different mechanisms and viewpoints. Early generated examples can also create fixation. Separate human idea generation, AI exploration and later selection so the first fluent answer does not set the whole direction.

• When should a product team stop generating and start testing?

  Move to testing when the shortlist is materially diverse, criteria are clear and the important remaining question requires evidence. A prototype, customer conversation or technical investigation often creates more learning than another round of ideas.