Decision-Oriented Summarization

Intermediate

Extract signal for a specific decision

01 The Concept

Generic summarization produces generic summaries. The trick: tell the AI who the summary is for, what decision it needs to support, and what to cut.

A summary for a CEO deciding on an acquisition is completely different from one for an engineer deciding how to implement — even from the same document.

02 Weak vs. Strong

EX 01Research paper → product decision

Summarize this UX research paper for a product manager deciding whether to redesign our onboarding flow. They have 3 minutes.

Structure as:

Core finding (1 sentence — what did this study actually prove, not just suggest?)
Methodology quality (1 sentence — sample size, study type, robust enough to act on?)
Direct implication for our onboarding (2–3 sentences — what specifically should we change? Be concrete, not theoretical.)
Biggest caveat (1 sentence — why might this NOT apply to our specific product?)
Verdict: Act on this / Gather more data / Ignore — with one-sentence rationale

[paste paper]

→ Why it works

Reader named (PM, 3 minutes). Decision named (redesign onboarding). Structure is decision-oriented.

EX 02Competitor earnings call → strategic intelligence

Here is the Salesforce Q1 FY2026 earnings call transcript. I'm VP of Sales at a competing CRM. I need to know: are enterprise customers pulling back on CRM spend, and is Salesforce winning or losing in the mid-market?

Summarize with this lens only:

Enterprise spending signals — what did management say (or avoid saying) about enterprise deal sizes and expansion revenue?
Mid-market commentary — any data about companies with 100–500 employees?
Analyst skepticism — what did analysts push back hardest on?
The thing Salesforce didn't say — what is management avoiding based on what analysts asked?
One-sentence intelligence takeaway for my CEO

[paste transcript]

→ Why it works

Reader perspective defined (VP of Sales, competing CRM). Two specific strategic questions.

03 Key Points

01Always specify: who reads this + what decision does it support

02'Summarize for someone who has 2 minutes' forces ruthless prioritization

03Structured summaries: finding / evidence / implication / recommendation

04'What's missing from this document that I'd expect?' — high-value follow-up

05'What are the 3 things I'd regret not knowing?' beats 'summarize'

04 Model-Specific Notes

Claude handles very long documents well — paste entire transcripts or reports. 200K context handles full earnings calls, research papers, or technical specs.

05 For Your Role

Never just say 'summarize'. Always add 'for someone who needs to [decide X]' and 'in [N] sentences'. These two constraints transform the output.

Next: Audience Translation →