Test the weight that flips your decision

TL;DR: Emily Higgins demonstrates asking Claude to build an interactive weighted decision matrix. I would use the table to find the assumption that changes the winner, rather than treating the highest total as a decision delivered by mathematics.

This is a useful share for someone comparing tools, proposals or project options with several competing criteria. The reel moves from naming a decision to adjusting criteria, weights and option scores, then asks the assistant to argue against the apparent winner. Original demonstration.

Around 0:29, Higgins enters the request. Around 0:38, the interface shows an adjustable comparison of Charlotte and New York; around 0:43, she introduces the challenge to the winning option. I reviewed the transcript and representative frames, including that table. The video does not establish that the method eliminates bias.

The useful number is the tipping point

Here is a checked, illustrative comparison I would use in an architecture review. Suppose two vendors are scored on reliability and ease of integration, with higher numbers better. Vendor A scores 9 and 6; vendor B scores 7 and 8. These are invented inputs for the example, not ratings of real products.

Give reliability 40% of the weight and integration 60%. A scores 7.2, while B scores 7.6. Change the weights to 60% reliability and 40% integration, and A wins with 7.8 against B's 7.4. At equal weights, they tie at 7.5.

The spreadsheet did not discover that A is objectively better. It exposed that the preference changes when reliability receives more than half the weight. That is the conversation I want the table to create.

I would then ask whether the scores deserve their precision. What evidence supports a reliability score of 9 rather than 7? Does the integration score include migration effort and ongoing maintenance, or just a pleasant first demo? A sensitivity check on guessed inputs can reveal fragility, but it cannot turn the guesses into measurements.

Keep requirements out of the popularity contest

Some conditions should be treated as prerequisites before scoring. If an option cannot satisfy a mandatory data-location requirement, a low price should not quietly compensate for that failure. I would filter for required conditions first, then compare the acceptable options.

This is my proposed use of the demonstration, not a claim that I have tested the creator's prompt across decisions. The AI can help build the interface and challenge assumptions, but the decision owner still has to define the criteria and supply defensible evidence.

Try changing the least certain weight and the least certain score in your next matrix. If a small adjustment reverses the ranking, the next action may be to gather information rather than approve the current winner. That is a more valuable output than another decimal place.

Use the matrix to locate the assumption you need to resolve before choosing.