Discovery and Problem Validation

Which customer problems are important, unresolved and actionable?

Product discovery is not collecting feature requests. It is building and challenging an explanation of a customer situation: the outcome sought, present behaviour, friction, alternatives, decision roles and conditions for change.

A problem becomes product-relevant when it is consequential, recurrent or strategically important; current alternatives are inadequate; the user or buyer can act; and the organisation has a plausible route to superior delivery.

1. Define the discovery decision

State what the work must decide: whether to enter a problem space, whom to prioritise, which outcome to design around or which assumption to test next. Define scope without embedding the proposed solution.

Weak question: “Would you use an AI assistant?” Strong question: “How is this task completed today, where does it fail, who bears the consequence and what would justify changing the workflow?”

2. Evidence sources

Combine:

observation and contextual inquiry;
interviews anchored in recent real episodes;
support, search, workflow and behavioural evidence;
lost-customer and non-consumer evidence;
commercial, operational and regulatory context;
experiments demonstrating commitment.

GOV.UK guidance recommends reviewing existing evidence, observing and interviewing actual or likely users, and continuing research through discovery, alpha, beta and live stages.¹

3. Map the problem system

For each situation capture:

trigger and context;
desired outcome and success criterion;
current journey and workaround;
functional, emotional and social consequences;
user, buyer, approver and affected parties;
frequency, severity and variability;
switching friction and trust;
ability and willingness to commit;
evidence for and against the hypothesis.

4. Sample for contrast

Recruit across outcomes, not only average customers: successful users, struggling users, switchers, rejecters, novices, experts and people excluded by the current solution. Contrast reveals mechanisms that a homogeneous sample hides.

Qualitative research explains how and why; it does not estimate population prevalence without an appropriate design. Quantitative work sizes patterns but may hide context. Behavioural tests assess action but require interpretation.

5. From insight to opportunity

An insight should include observation, mechanism, affected population, consequence and design implication. “Customers want simplicity” is not decision-ready. “First-time applicants abandon when asked for information they cannot access until an employer responds” identifies a situation, dependency and intervention point.

6. Successful solution patterns

Several mature organisations publish useful mechanisms:

Amazon describes working backwards from customer needs and avoiding solutions invented in isolation.²
Intuit’s Design for Delight connects empathy, broad solution exploration and rapid customer experiments.³
GOV.UK treats user research as continuous through the service lifecycle, not a front-loaded phase.¹

These are patterns, not universal recipes. Their value is the mechanism: direct evidence, divergent options and continuing validation.

7. AI-native discovery

AI can search large feedback collections, translate interviews and surface candidate themes. It must not create synthetic customer evidence, infer representativeness or erase minority experiences. Preserve original context, verify coding samples, disclose AI transformation and protect personal data.

8. Asia-Pacific and Hong Kong application

Conduct research in the language and channel in which the behaviour occurs. Regional averages can obscure payment, family, employer, platform, procurement and regulatory differences. Hong Kong discovery should consider local and cross-border journeys where relevant.

9. Exit criteria and QA

Discovery is sufficient when the decision owner understands the outcome, mechanism, affected population, alternatives, consequence, uncertainty and next test. Common failure modes include leading questions, convenience samples, treating complaints as prevalence, asking about hypothetical purchase and excluding non-users.