A product decision depends on what people are trying to do, what prevents them and which response is worth pursuing. AI can help examine evidence, compare explanations and prepare a test.
Investigate a loss of market share
In the hypothetical case from Thinking through a strategic decision with AI, a company is losing market share. The proposed response includes matching a rival’s price and adding features. First establish what the share measure counts and whether the company is actually losing customers. If departures are confirmed, the next question is where those customers went.
The same business signal could fit several explanations:
| Explanation to investigate | Evidence that could distinguish it | How the answer could change the product decision |
|---|---|---|
| Customers choose a competing product | Ask about the last purchase or switch and inspect relevant cancellation records. | Compare the specific alternatives and the reason for choosing one before testing a price or feature response. |
| Customers stop buying the category | Ask how their underlying task or need changed. | Reconsider the problem being served; copying a rival’s feature may not address it. |
| Customers cannot complete onboarding | Observe the attempted task and inspect the relevant failure or support records. | Test a change to that step and check whether people can complete it. |
These are constructed alternatives, not findings about a real customer group. A cancellation form that asks only about cost cannot settle all three. Record that limitation instead of treating the available field as the whole explanation.
Give the assistant evidence it can examine
Provide the records the assistant is allowed to use and ask it to separate observations, interpretations and missing evidence. Keep each proposed conclusion linked to the passage or observation supporting it. Have it look for a counterexample that would change the recommendation.
An assistant can help prepare interview questions, compare permitted notes or identify a gap in the cancellation form. Its simulated customer response is a hypothesis. The GOV.UK user-research guidance emphasizes learning from actual or likely users and treating other suggestions as assumptions to investigate. Its context is government services; the distinction between evidence and assumptions is useful in this product case too.
Decide what you need to learn next
Choose the uncertainty whose answer would most change the proposed investment. For onboarding failure, a useful next observation is whether a person can complete the step that stopped them. A broad satisfaction question would provide different evidence.
If the team tries a bounded change, decide beforehand what it will measure and what would justify changing or stopping the trial. In the article’s price-cut example, retention needs to be considered alongside support effort and exception requests. A short trial cannot settle every long-term consequence.
The person responsible for the product chooses the investment and the conditions of the test. Preserve the reason for that choice so later results can change it. Project records help retain the decision; design helps examine the interaction people will encounter.
Use and improve the method
The complete strategic-decision case and worksheet develop the example. The maintained thinking-framework instructions offer questions about assumptions, dependencies, uncertainty, analogies and user needs. Choose the questions relevant to the decision rather than completing every heading.
Rajiv Pant wrote the framework and worked case. The public project welcomes examples from non-coding work. A useful contribution explains the decision, the evidence that changed it and the limits of what was learned. Obtain permission before sharing customer material. Base user-research findings on observations of actual or likely users.