Claude in Practice — Two Worked Examples

Move one

Getting content into the conversation

Claude can only reason over what it can see. Here are the ways to put material in front of it — pick by where the content already lives.

Paste it

Short text, a clause, a few rows.

Straight into the message

Upload a file

PDFs, Word docs, spreadsheets, CSVs.

“+” or drag into the chat

Project knowledge

Material you reuse across many chats.

Project → Files

A connector

Live pull from Drive, Box, and similar.

“+” → Connectors

Web search

Current facts, benchmarks, norms.

Ask Claude to search

Local files (desktop)

A folder on your machine.

Filesystem extension / Cowork

Move two

What a better prompt is made of

“Review this” gives Claude nothing to aim at. A strong prompt usually has four parts — give it a role, the task, what to focus on, and the shape of the answer you want.

You’re helping me review a vendor SaaS agreement before we sign. Go through it clause by clause and flag anything risky or unusual. Pay special attention to auto-renewal, termination, liability caps, and data-security obligations. For each flag, quote the section, say whether it’s standard, and put the highest-risk items first.

Role & context The task What to focus on Format of the answer

Reviewing a contract for callouts

A first-pass read to surface risks, deviations, and missing terms — before it reaches Legal.

Step 1 · Get the contract in

Upload the agreement as a PDF or Word file (the “+” button, or drag it in). If you review a lot of vendor contracts, keep your standard template and a checklist in a Project so every review starts from the same baseline.

Step 2 · Ask the sharper question

“Review this contract and tell me if it’s fine.”

Better

“You’re helping me review this vendor SaaS agreement before signing. Go clause by clause and flag: payment terms and price escalators, auto-renewal and notice periods, termination rights, liability caps and indemnification, data-security and breach obligations, IP ownership, and SLA commitments. For each flag, quote the clause and section number, say whether it’s standard or unusual, and note anything one-sided or missing that I’d expect to see. Put the highest-risk items first. I’m not a lawyer — call out where I should get legal review.”

Gives a roleNames what to checkAsks for citationsStandard vs unusualRanks by riskFlags for escalation

Step 3 · Bring in a reference to check against

Import from your documents or the web

Claude reviews far better when it has something to compare to. Attach the reference, then point Claude at it.

From a local document“Compare this against our standard MSA template (attached) and list every place the vendor’s terms deviate from ours.”

From the web“Search for typical SaaS auto-renewal notice periods, then tell me whether this 90-day window is unusual and why.”

Stay grounded“Base your answer only on the attached contract and template. If a term isn’t addressed, say it’s missing rather than assuming.”

What good output looks like

Clause	Section	Standard?	Risk	Note
Auto-renewal	§8.2	Unusual	High	Renews unless cancelled 90 days out — longer than typical 30.
Liability cap	§11.1	One-sided	Medium	Caps vendor liability; no mutual cap for us.
Data security	—	Missing	High	No breach-notification timeline specified.

Illustrative shape only — Claude fills this in from the actual document.

Hopper note: contracts are business-confidential, so they’re safe to work on here — unless an exhibit or attachment contains patient data (e.g. a member list). Strip those first. And this is a first pass, not legal sign-off: route anything material to Legal. Data-handling guide

Analyzing aggregate financial data

Turning a spend or budget spreadsheet into trends, drivers, and outliers you can act on.

Step 1 · Get the data in

Upload the spreadsheet (CSV or XLSX) or connect the source sheet through a connector. With code execution on, Claude can actually compute over the numbers and build charts and a clean workbook — it isn’t eyeballing. Make sure the file is aggregated: totals by department, month, or category — not row-per-person.

Step 2 · Ask the sharper question

“Analyze this data and tell me what stands out.”

Better

“You’re my FP&A analyst. This is our aggregated monthly departmental spend for FY26 — no employee- or patient-level detail. Walk through: total spend trend month over month, the three biggest drivers of the Q2 increase, any line items that look like outliers or data-entry errors, and every budget-vs-actual variance over 10%. Show the numbers behind each point, build a chart of monthly spend by department, and flag anything you can’t explain from the data alone.”

Gives a roleStates what the data isSpecific questions“Show the numbers”Asks for a chartFlags gaps

Step 3 · Bring in a baseline or benchmark

Import from your documents or the web

Comparisons make the analysis useful. Give Claude the baseline, or let it fetch context.

From a local document“Use last year’s actuals in the second attached file as the baseline and add a year-over-year column for each department.”

From the web“Look up the current medical-care inflation rate, then tell me how much of our cost growth is above or below it.”

Stay grounded“Only use figures from the attached files and the search result. If a number isn’t in the data, don’t estimate it — tell me it’s missing.”

What good output looks like

A short written read-out (trend, top drivers, outliers, variances), a chart of monthly spend by department, and — if you ask — a tidy spreadsheet with the YoY columns added. Then you can keep going: “Draft a 4-bullet summary for the leadership update.”

Hopper note: aggregate, de-identified financials are safe to analyze here. The moment the file drops to patient- or claim-level detail with identifiers, it’s PHI and stops at the door. When in doubt, aggregate first. Data-handling guide

Keep these handy

Prompt patterns that work everywhere

Ground it

“Answer only from the attached files and the search results. If it’s not covered, say so instead of guessing.”

Show the work

“Show the numbers and the section references behind each point so I can check them.”

Rank it

“Put the highest-risk / highest-impact items first, and tell me what you’d look at next.”

Name the gaps

“Flag anything missing or anything you can’t determine from what I gave you.”

Set the format

“Give me a table with these columns…” or “a 4-bullet summary for leadership.”

Iterate

“Now tighten that to one paragraph” — keep refining in the same chat instead of starting over.”