Amazon Quick Desktop for hospitals: executives and clinical teams
Morning for executives, afternoon for clinical teams: use Quick Desktop to ask, decide, create documents, build Skills and agents that work on a schedule
AnyLab and all patient data in this workshop are fictional demo data.
What we will do today
Morning: the executive team of AnyLab (a fictional laboratory network) uses Quick Desktop to prepare board answers. Afternoon: each department builds an AI agent for its own work, such as medication safety, triage, nursing, check-up reports and MDT meetings.
Opening: Amazon Quick Desktop for hospitals, executives and clinical teams
09:15
10 min
Lab 0 Setup
09:25
25 min
Lab 1 Ask the business (AnyLab)
09:50
30 min
Lab 2 Decide: numbers + documents (AnyLab or hospital)
10:20
15 min
Board assurance drill: catch the AI error + 5 governance questions
10:35
15 min
Break
10:50
20 min
Lab 3 Deliver PowerPoint / Word / Excel
11:10
20 min
Lab 4 Skills
11:30
20 min
Lab 5 Agent + routine
11:50
10 min
Morning close: one-page Pilot Protocol
12:00
60 min
Lunch
13:00
10 min
Afternoon: each team picks one department use case
13:10
40 min
Lab 6 Build a clinical agent from a system prompt
13:50
40 min
Lab 7 Ground the agent in our documents (Space) + test its limits
14:30
15 min
Break
14:45
35 min
Lab 8 Routine + generate the report as a file
15:20
30 min
Show & Tell: 4 minutes per team as an executive handoff
15:50
10 min
Close: each team's Pilot Protocol + survey
Morning: executives (AnyLab)
Lab 0Setup: let Quick see the work folder10 min
Quick accesses only folders you grant (OS sandbox), and per folder you choose whether it is indexed for search. This is the first data control point.
Open Quick Desktop and sign in
Go to Customize > Knowledge and add the folder AnyLab-Workshop
Wait a moment: Quick reads the folder immediately; indexing continues in the background
Go back to chat and paste the prompt below
Prompt basic
What files are in the AnyLab-Workshop folder? Describe each in one line, as a table.
Checkpoint:You see 4 CSV files, 6 PDFs, 5 skill .zip files and the hospital-case and healthcare folders
Lab 1Ask the business25 min
Decision to make: Is our growth healthy, and what should the board worry about?
See the difference between a short question and a prompt with role, task, format and constraints
1a Ask the short way first
1b Ask the structured way and compare with 1a
1c Follow up on growth quality
Prompt 1a
How did AnyLab do this year?
Prompt 1b
<role>You are the finance analyst reporting directly to the CEO.</role>
<task>From lab-revenue-monthly.csv in the AnyLab-Workshop folder, summarise the fiscal year Oct-2025 to Sep-2026:
1. Total revenue and gross margin (sum of gross_profit_thb / sum of revenue_thb).
2. The two fastest-growing product lines and any shrinking ones, H2 vs H1.</task>
<output_format>A one-sentence conclusion + a table of max 6 rows + the file and period for every number.</output_format>
<constraints>Do not estimate anything not in the data.</constraints>
Prompt 1c
Which product lines grow but pull total margin down, and which grow and lift it? Answer as a table.
Checkpoint:1b gives sourced numbers and answers more precisely than 1a
Lab 2Decide: numbers + documents in one answer30 min
Decision to make: Do we approve this capex, and which customer do we save first?
Have Quick test a team proposal against real data and find the real cause in documents
2a Whole room: the Khon Kaen capex proposal
2b Then: key accounts at risk (lab executive room)
Or 2b-H: claim denials and ICU beds at a fictional hospital (hospital executive room, chosen by the organiser before the day)
Prompt 2a
<role>You are the COO preparing for the board.</role>
<task>Using the files in the AnyLab-Workshop folder:
1. From lab-operations-monthly.csv, show utilisation and tat_breach_pct for Khon Kaen Regional Lab over 12 months vs the other labs.
2. Read 02-khon-kaen-capacity-proposal.pdf and 04-tat-sla-and-quality-policy.pdf.
3. Test each assumption behind the 3.1-year payback: supported or contradicted by the data?
4. Recommend an option with a first-90-days plan.</task>
<output_format>A one-sentence recommendation + an assumption-check table + a 90-day plan.</output_format>
Prompt 2b
From key-accounts.csv, which 3 accounts are most at risk, and how much revenue renews in the next 90 days at high risk? Then read the escalation document in the AnyLab-Workshop folder and state the real root cause for the #1 account.
Prompt 2b-H
<role>You are the CFO of Rattanawetwat Hospital (fictional), preparing for the executive committee.</role>
<task>Using the files in AnyLab-Workshop/hospital-case:
1. From claims-denials.csv: total denied amount by denial_code and payer_name, the amount never appealed, and the appeal win rate for decided cases.
2. Compare the denial rate for payer_code KRH when icu_days is above 3 vs 3 or fewer, then read 17-payer-coverage-criteria-summary.pdf and 18-denial-letter-krh-cl88486.pdf to explain why.
3. From bed-census-hourly.csv: the dates when icu_beds_available stayed at 0-1 all day.
4. Propose 3 decisions that raise cash without raising prices, each with an owner by role.</task>
<output_format>A one-sentence conclusion + a table of max 8 rows + the file for every number.</output_format>
<constraints>Do not estimate anything not in the data. If a group has few cases, say the conclusion is weak.</constraints>
Checkpoint:The answer cites both the CSV and the PDFs in one response
DrillBoard assurance drill: catch the AI before the board does15 min
Decision to make: If this draft reached the board tomorrow, how far could we trust it?
The draft below contains wrong numbers, unsupported claims and omitted risks. Executives should be able to check it with the right follow-up question, not tool skills.
Read the board note draft (do not fix it yourself)
Paste the verification prompt and see whether Quick finds every problem
Whole room: answer the 5 governance questions below, separating what Quick provides from what your organisation must decide
Prompt draft
Board note draft (prepared by an analyst, not yet checked)
1. H2 gross margin improved to 43.1%, driven by Wellness growth.
2. Price per test for government hospitals was stable, so no pricing action is needed.
3. The Khon Kaen expansion pays back in 2.4 years, confirmed by an external consultant's market study.
4. No key account above THB 20 M is at risk of leaving in the next 90 days.
Recommendation: approve the Khon Kaen capex now; no other risks need board attention.
Prompt verify
Check the board note draft above against the files in the AnyLab-Workshop folder, claim by claim:
- Mark each claim Correct / Wrong / Not supported by the files, with the file name and value that proves it.
- Then list up to 3 important risks the draft leaves out.
- Do not rewrite the draft, and say so when you are not sure.
5 governance questions
Board question
What Quick provides (per AWS docs)
What your organisation decides
Where does our data go, and is it used to train models?
Quick runs in the cloud; conversations, memory and the content index are kept per user in your Amazon Quick account and are never used to train or improve models (docs: Security, privacy, and architecture)
Which data is off limits (for example health data), the PDPA lawful basis, the account Region, retention, and breach notification, confirmed by the DPO and legal
What files can Quick see?
Only folders the user grants, via operating-system sandboxing, revocable at any time, with per-folder choices for search and knowledge-graph extraction (docs)
Which folders and Spaces are allowed, who owns each data set, and approved document versions only (HA/JCI, ISO 15189 document control)
What can an agent do on its own?
Every tool and connector can be set to Always Allow / Ask Each Time / Always Deny, including for scheduled agents; an agent can take only authorised actions (docs)
Which work may be Always Allow, which must be Ask Each Time, and which is never allowed, such as messaging patients or customers
Who controls sharing of skills and agents?
Admins restrict capabilities with custom permissions and can turn on Require approval to share skills · Publish and Share are separate steps (docs: Skills)
A register of approved use cases, approvers, versions, test sets, and what happens when one is found wrong
If the AI is wrong, who is accountable?
No tool takes accountability; 'pending review' in a prompt is guidance, not a control
An accountable executive per use case, a clinical or quality sign-off owner, and a stop procedure when it fails (RACI)
Checkpoint:Quick finds claims 1, 2 and 4 wrong, claim 3 unsupported, and names at least 2 missing risks. Anything it misses is why people must review
Lab 3Deliver: PowerPoint / Word / Excel20 min
Decision to make: Is this document ready for the board, and who must check it first?
Quick creates real files on your computer with its built-in system skills (documents, presentations, spreadsheets)
Pick 1 of 3 by role: CEO/COO = PowerPoint, CFO = Excel, others = Word
While it works, watch which step Quick is on
Open the file in PowerPoint / Word / Excel, check 2 numbers against Lab 1, and note who in your organisation must review it before the board
Prompt PowerPoint
<role>You are the CEO's Chief of Staff.</role>
<task>
Create a new PowerPoint named AnyLab-Board-Pack-Nov2026.pptx for the November 2026 board meeting, answering actions B1-B4 from the Q3 minutes:
1. Executive summary + decisions requested (first slide)
2. FY2026 performance and growth quality (B1)
3. Government tender pricing (B2)
4. Northeast capacity + interim plan (B3)
5. Key accounts at risk (B4)
6. Proposed FY2027 strategy adjustments
</task>
<output_format>Max 10 slides + appendix · slide titles are conclusion sentences · charts from real data · speaker notes on every slide · source footnotes</output_format>
<validation>Check every number against the CSVs before saving · never overwrite an existing file</validation>
Prompt Word
<role>You are the COO.</role>
<task>Write a one-page decision memo as a Word file named AnyLab-Decision-Memo-Northeast.docx on the Northeast capacity options.</task>
<output_format>
Headings: Decision requested (first paragraph) · Situation (3 numbers) · Options considered · Recommendation · Budget and timeline · Risks and mitigation · CEO / CFO signature block
</output_format>
<constraints>Max one A4 page · formal tone · cite sources for numbers · never overwrite an existing file</constraints>
Prompt Excel
<role>You are the FP&A lead.</role>
<task>
Create a new Excel file named AnyLab-FY2027-Scenarios.xlsx from lab-revenue-monthly.csv and the assumptions in the Strategy FY2027 document:
- Inputs sheet: growth rate and gross margin per product line, government price change (yellow editable cells).
- Scenarios sheet: Base (per plan), Downside (PCR and government price keep falling at the H2 trend), Upside.
- Summary sheet: revenue, gross profit and GM% per scenario vs the 4.20 bn / 42% targets, plus a chart.
</task>
<validation>Every number in Scenarios/Summary must be a formula referencing Inputs, never a typed value. Check that FY2026 totals match the CSV.</validation>
<edge_cases>Never overwrite an existing file; append the date if the name exists.</edge_cases>
Checkpoint:A new file that opens, with no existing file overwritten
Lab 4Skills: capture how we work20 min
Decision to make: Which way of working should be the same for everyone?
A Skill is a way of working written once; everyone uses it and gets the same format; edit once and everyone gets the update when you publish
4a Customize > Skills > Create > From file, pick skills/capex-proposal-reviewer.zip, then Save
Choose Try it, paste prompt 4a and compare with the Lab 2a answer
4b (if time allows) Build your own: Create > From chat, then paste prompt 4b
Look at Publish and Share (no need to share for real in the room)
Prompt 4a
Use the capex-proposal-reviewer skill to review 02-khon-kaen-capacity-proposal.pdf against lab-operations-monthly.csv and key-accounts.csv.
Prompt 4b
Create a skill named weekly-exec-note: whenever I give numbers or files, write a 5-line note to the executive team in this order: 1) key conclusion 2) three key numbers with sources 3) one risk 4) the decision needed, with a date 5) who owns it. Polite, concise, no technical jargon. If information is missing, ask first.
Checkpoint:Two skills appear under Customize > Skills and Try it follows the requested format
Lab 5Agent + routine: work for you every week20 min
Decision to make: Which routine work may an agent do alone, and which must always ask first?
An agent is a teammate with its own role and instructions · a routine runs it on a schedule without prompting
5a Customize > Agents > Create > From chat, paste prompt 5a, check name/description, then Publish
Pick the agent in the agent switcher and ask: 'Test the FY2027 plan'
5b Open Mission Control > Create, make a Monday 07:30 routine with instruction 5b (or type 5b in chat)
Run it once now and see the result in Mission Control
Prompt 5a
Create an agent named Strategy Challenger with these instructions:
<role>You are the "Strategy Challenger", an independent strategy advisor to the executive team. Your job is to find the weaknesses in the plan before the board does.</role>
<task>
Read "Strategy FY2027 - Draft v0.9" in the AnyLab-Workshop folder, then:
1. Test assumptions A1-A5 one by one against the datasets: supported / contradicted / not yet provable, with numbers.
2. Assess whether the +14% revenue target and 42% gross margin remain realistic if the contradicted assumptions hold.
3. Write the 5 hardest questions the board is likely to ask.
4. Propose 3 changes to the plan.
</task>
<output_format>Assumption table (assumption | evidence | status | impact) → 5 board questions → 3 recommendations</output_format>
<constraints>Be direct. Do not agree to please. Every point needs evidence.</constraints>
<example>A1 "PCR flat" → contradicted: PCR H2 fell vs H1 (insert the actual figure)</example>
Prompt 5b
Every Monday at 07:30, read the latest files in the AnyLab-Workshop folder and write a one-page summary: labs with tat_breach_pct above 5% or utilisation above 85%, accounts renewing within 90 days with churn_risk_score of 60 or more, and the 3 decisions needed this week. Never send email or messages without asking for approval.
Checkpoint:The agent is in the agent switcher and the routine shows in Mission Control with one run
Afternoon: department use cases
Safety rule:Every clinical use case in this room uses fictional data. Never enter real patient data. AI output is a draft to support staff and must be checked by a doctor, nurse or pharmacist before use.
# Role
You are the LASA Drug Safety Advisor, a medication-safety assistant for pharmacists and nurses.
# Duties
1. Find look-alike and sound-alike drug pairs from the documents in the Space.
2. State the risk level and the differences (indication, dose, form).
3. Recommend prevention measures such as Tall Man Lettering and separate shelving.
# Answer format
Risk level: High / Medium / Low · drug pair · differences · prevention · sources
# Constraints
- Answer only about medication safety and LASA.
- Do not diagnose or prescribe.
- Cite the documents every time. If not found, say so and advise consulting a pharmacist.
Test 1
Which drugs look or sound like Losartan?
Test 2
How do Amlodipine and Amiodarone differ?
Test 3
Give me stock investment advice
Deliver (Lab 8a)
Create an Excel file LASA-Shelf-Alert.xlsx: all pairs sorted by risk with Tall Man Lettering and measures. Never overwrite an existing file.
Routine (Lab 8b)
Every Monday at 07:00, review the LASA drug pairs in the Space and summarise high-risk pairs and measures to review, as a one-page report for the pharmacy team
Contact center / ER
Call Intake and Red-Flag Routing Draft
Files: healthcare/Clinical_Triage_Cases_Thai.csv
System prompt (Lab 6)
# Role
You help hospital contact-center staff take a call and prepare a handoff for the triage nurse. You are not the triage decision-maker: you do not assign urgency levels and you do not give clinical advice.
# Steps
1. Record what the caller said and ask only for what is missing: age, main complaint, onset, severity 1-10, conditions, medicines.
2. Check against the approved red-flag list: chest pain, breathing difficulty, fainting or drowsiness, seizure, one-sided weakness, slurred speech, heavy bleeding, fever in an infant under 3 months.
3. If at least one red flag is present: start with "Hand to the nurse now per protocol" and name the red flags.
4. If none: summarise for the triage nurse's normal queue.
# Format
Information received · information missing · red flags found (or "none from the list") · handoff · "Draft pending triage nurse confirmation"
# Constraints
- Never diagnose, never propose an ESI or urgency level, never recommend medicines or treatment.
- If unsure whether something is a red flag, treat it as one and hand to the nurse now.
Test 1
Left chest pain radiating to the arm for 30 minutes
Test 2
My 2-year-old has a wet cough, no fever
Test 3
I have chest pain. Diagnose whether it is heart disease and tell me what medicine to take
Deliver (Lab 8a)
Read all 10 cases in Clinical_Triage_Cases_Thai.csv using only the columns Case_ID to Underlying_Conditions, then create a Word file Intake-Handoff-Summary.docx: complaint, missing information, red flags found and handoff. Never assign urgency levels.
Routine (Lab 8b)
Every day at 08:00, from the latest case file, summarise how many calls were handed to the nurse immediately for red flags, the most common red flags and the information most often missing, as a short report for the center lead
# Role
You are the Nursing Assessment Analyzer, helping registered nurses analyse FANCAS assessments.
# FANCAS
F Fluid & electrolytes · A Activity · N Nutrition · C Comfort/pain · A Airway · S Safety
# Steps
1. Read each patient's assessment.
2. Analyse every FANCAS domain that has data.
3. Produce a prioritised problem list.
4. Propose nursing interventions, expected outcomes and reassessment criteria.
# Format
Table: # | nursing problem | FANCAS domain | priority, then interventions per problem.
# Constraints
- Use correct nursing terminology.
- Do not order medicines or medical procedures.
- Always state: "Verify with RN/MD before acting".
Test 1
Produce the problem list for patient PT-001
Test 2
Which patient should be prioritised and why?
Test 3
Prescribe a painkiller for PT-001
Deliver (Lab 8a)
Create an Excel file Ward-FANCAS-Overview.xlsx: one sheet per patient plus an overview of problems per FANCAS domain
Routine (Lab 8b)
Every morning at 06:30, read the latest assessments and summarise per-patient problem lists and a ward overview: problems per domain and high-priority patients
Health check-up center
Check-up Report Writer
Files: healthcare/BPH_Checkup_Samples_Thai.json
System prompt (Lab 6)
# Role
You write plain-language annual check-up reports for clients, reviewed by a doctor.
# Steps
1. Read each person's labs, vitals, questionnaire and imaging.
2. Compare with reference ranges and last year; flag abnormal values and trends.
3. Write a one-page plain summary: normal results, results to watch, lifestyle advice, and what to discuss with a doctor.
# Constraints
- Do not diagnose; use "please consult a doctor for further assessment".
- Every number must come from the file; never guess.
- End with: "Draft generated by the system, pending doctor review".
Test 1
Summarise CU-001's results versus last year
Test 2
Who has the most results to follow up?
Test 3
Does CU-001 have diabetes?
Deliver (Lab 8a)
Create a Word file Checkup-Report-CU-001.docx as a one-page client report
Routine (Lab 8b)
Every Friday at 16:00, list this week's clients with more than 2 abnormal results so nurses can call them
# Role
You are the MDT meeting secretary, turning the transcript into decisions and actions.
# Steps
1. Split by patient case.
2. Summarise: key facts, each discipline's view, team decision, proposed treatment plan.
3. Actions: task | owner (by role) | due date.
4. Open questions without a decision.
# Constraints
- Summarise only what was said; never add medical recommendations.
- Mark it "draft pending chair approval".
Test 1
Summarise the decision for case 1
Test 2
Who has to do what after the meeting?
Test 3
Which chemotherapy do you recommend instead?
Deliver (Lab 8a)
Create a Word file MDT-Minutes-2026-07-22.docx as formal minutes with an action table
Routine (Lab 8b)
Every Wednesday at 11:00 after the MDT meeting, summarise the latest transcript into decisions and actions and store it in Agent Files
Lab 6Build a clinical agent40 min
Each team picks its department use case and builds an agent with a clear role, steps and constraints
Pick one use case below
Customize > Agents > Create > From chat, type 'Create an agent named ... with these instructions' and paste the use case system prompt
Check the name, description and instructions, then Publish
Pick the agent in the agent switcher and try the first 2 test questions
Improve the instructions at least once, then Publish again
Checkpoint:The agent answers in its role and format
Lab 7Ground the agent in our documents40 min
The agent must answer from approved documents and refuse out-of-scope questions
Open Spaces > Create and name the Space after your department
Upload the use case files from AnyLab-Workshop/healthcare
Edit the agent: Capabilities tab > Spaces, add this Space, then Publish
Ask test question 3 (out of scope); the agent should refuse or redirect
Ask something not in the documents; the agent should say it is not found
Checkpoint:Check yourself that answers cite the right Space documents and question 3 is refused; if not, fix the instructions and test again
Lab 8Routine + report as a file35 min
Have the agent work on a schedule and deliver a file the team can use
8a Paste the use case 'deliver' prompt in chat with the agent and open the file
8b Mission Control > Create, pick your team's agent, enter the use case routine instruction, and run it once now
Check the result in Mission Control and Agent Files
Checkpoint:A file that opens and one routine run in Mission Control
Show & Tell4 min / team
Present it as a short executive handoff, not a product demo
The problem and who benefits (30 seconds)
Evidence: one live question on the agent (60 seconds)
Where the AI got it wrong while testing (45 seconds)
The controls: approved documents, permissions, who reviews (45 seconds)
30-day pilot: owner, endpoint, decision date (60 seconds)
One-page Pilot Protocol10 min
Instead of a commitment card: one person or team, one pilot with an owner, an endpoint and a decision date. Fill it in here (stored only in your browser), then let Quick draft the document
Prompt
From what I filled in below, write a one-page Pilot Protocol for the executive team: a heading per field, 3 risks with mitigations, and what needs approval before starting. If a field is empty or unclear, ask first. Never invent numbers.