New Feature: AI Summaries

AHP results are precise, but they aren't always easy to read at a glance. The ranking chart, the priority breakdowns and the sensitivity analysis each tell part of the story, and it's up to you to put them together. From today, decisionpoint.io does some of that work for you: the Results and Sensitivity pages now include a short summary, written in plain English.

What You Get

On the Results page, the summary sits just below the overall ranking. It tells you which option leads and by how much, which criteria drive that lead, and where the runner-up does better.

On the Sensitivity page, the summary focuses on the question that matters most: how robust is your top choice? If the top two options are effectively tied, it says so, and then lists the weight changes that would put a different option in first place, smallest first.

You can see both in action on the sample decision, Choose the best car for the Jones family.

Can You Trust It?

AI models are good at writing, but much less reliable at arithmetic. So the summaries are built around one simple rule: the AI never calculates anything. The app works out every number, and the AI's only job is to describe them. Before a summary is shown, every number in it is checked against your results. If the AI writes a number that didn't come from your results, even a plausible-looking one, the summary is rejected and written again.

A few other ground rules:

• Summaries are always labelled as AI-generated.
• The AI never tells you which option to choose. That decision is yours.
• When your results change, the summary is rewritten to match.
• Survey respondents never see AI-generated content.

Your Data

The summaries are written by Anthropic's Claude model, running on Amazon Bedrock in AWS's EU regions. Only what's needed to write the summary is sent: the names of the goal, criteria and options, and the computed results. Your account details are never sent, and Amazon Bedrock doesn't use your data to train models.

Availability

AI summaries follow the same rules as sensitivity analysis: they are free on public decisions, and available on private decisions with a subscription.

Try It Out

AI summaries are live now in the app. Open the results of any decision with some judgments entered, and the summary will be written on your first visit. As always, I'm very happy to receive feedback, especially if a summary reads oddly or gets something wrong. Just use the contact form or send me an email at philip.healy@decisionpoint.io.

— Philip