Use Many AI Models. Not One.

One model = blind spots. Many models = better work.

Simple view of AI

Conversational (LLMs)

Chatbots that read, write, and talk. They help you think and draft. Examples: ChatGPT and Claude.

Knowledge (Databases)

Spreadsheets and databases that store facts. They are exact but do not chat.

The real power comes when Conversational models read your Knowledge systems and report back in plain words.

Effective use of AI can save 10s of hours per person, every week. Increase in productivity is incredible.

Why many models win

  • •No single model is best at everything. Using two or more reduces mistakes.
  • •Cross‑check key work: ask both models and compare answers.
  • •Mix price tiers: cheap for simple tasks, premium for hard ones.
  • •Avoid lock‑in: you can switch fast as the market changes.

Why one model is wrong here: Different jobs need different strengths. A single model can fail from outages, policy changes, price swings, or bias. Using two models keeps work moving and keeps answers honest.

Two stars to start with

Primary

ChatGPT

  • •Great at quick thinking, analysis, and planning.
  • •Strong voice and image tools. Easy on phones.
  • •Wide app add‑ons to connect with your work tools.
Primary

Claude

  • •Calm, clear writing. Handles long docs well.
  • •Solid coding help and step‑by‑step reasoning.
  • •Pairs well with ChatGPT for double‑checks.

Why these two (ChatGPT + Claude): They are fast, stable, and widely used. ChatGPT is great at quick thinking and tools. Claude shines at long, careful writing. Together they catch each other's misses.

Also useful (secondary): Microsoft Copilot for Office files inside M365, and Perplexity for web research with sources.

Personal assistants for everyone

Each person gets an AI buddy on phone and laptop. It drafts emails, turns meetings into notes, pulls numbers from sheets, and reminds you of tasks. People do more in less time with fewer back‑and‑forths. Managers get faster updates and fewer status meetings.

Security: where your data goes

What happens under the hood

  • •Your prompts and files go to the provider's cloud servers to get an answer.
  • •Copies can be kept as logs for a time (for safety, abuse checks, and uptime).
  • •Unless on an enterprise plan with data controls, staff tools may view some logs.
  • •Attachments (PDFs, images, CSVs) are stored to process them, then kept for a period.
  • •Basic metadata is kept too (time, IP, file name). Choose Canada data options when needed.

Simple rule: do not paste secrets. For sensitive work, use an enterprise setup with data‑retention limits and model‑training opt‑out (available in settings on paid tiers).

By model (quick view)

  • ChatGPT (OpenAI): Prompts/files are processed in OpenAI's cloud. Consumer/Teams plans may retain logs for safety. Enterprise adds admin controls and limits on retention. Do not upload: client PII, contracts, unreleased financials, source code, or regulated data unless your enterprise controls are enabled.
  • Claude (Anthropic): Similar cloud processing with strong safety defaults. Enterprise plans offer stricter data controls. Do not upload sensitive or regulated data unless under enterprise protections.
  • Microsoft Copilot: Runs inside your Microsoft cloud and respects your Graph permissions. Good for Office files. Still avoid secrets unless your tenant policies allow and data classification is in place.
  • Perplexity: Built for web answers. Queries may reach the open web. Treat prompts as public; do not upload client files.

Hallucinations (and how to reduce them)

AI can "sound right" but be wrong. For example, When asking about a sheet, make it show its work:

  • •Ask for cell ranges used (e.g., "Sum B2:B19, exclude blanks").
  • •Ask for the exact formula it would place in a cell.
  • •Have it list checkpoints (totals match, counts match, no text in number columns).
  • •Never let it write to the sheet without a preview table you approve.
  • •Ask for sources or links when it cites facts.
  • •Cross‑check important answers with a second model.
  • •Spot‑check numbers and re-run on a small sample.

App integrations on your computer

Models can connect to your calendar, email, files, Sharepoint, Excel, Teams, and browsers via official add‑ons. Some tasks can drive desktop apps by safe automations. Always review the permission screen before turning on an add‑on.

How to use multimodal at work

  • •Voice → Notes: Dictate ideas; get a clean summary and tasks.
  • •Image → Data: Snap a whiteboard, receipt, or form; get a table.
  • •Docs → Answers: Drop a PDF/CSV; ask pointed questions.
  • •Screen → Steps: Record a short clip; get written how‑to steps.

Prompting: be fast

  • •Give Role, Goal, Source, Output, Limits (RG‑SOL).
  • •Paste a tiny example of the format you want.
  • •Ask for 3 options and a 5‑step checklist.
  • •Save good prompts in a shared note for reuse.

Example prompt (AirPro – CC295 fleet)

Role: You are an operations assistant for AirPro.

Goal: Create today's CC295 readiness report.

Sources: Excel "CC295_Fleet" — tabs Fleet (B2:G100), Defects (A2:F500), Maint (A2:H200); Ops_Log PDF at /AirPro/Logs/2025-08-01.pdf.

Output: For each tail, show: Location, Last flight date, Open defects (# and short text), Next maintenance, Crew notes. Then add a summary: aircraft ready today / total, top 3 risks.

Limits: If data is missing, write "Unknown" and list the exact cell ranges you checked. Do not guess. Show any formula used for totals.

Rapid custom app development

Small, useful tools can be built in days with no‑code. Examples: a form that writes to a sheet, a bot that summarizes PDFs, a dashboard that refreshes daily.

Timeline & quick access

1

Day 1

Create ChatGPT and Claude accounts; install mobile apps.

2

Week 1

Short training on prompts and voice. Pick 3 team tasks.

3

Weeks 2–3

Connect email, calendars, and sheets. Set safety rules.

4

Week 3

Ship one mini‑app (report bot or sheet helper).

5

Month 2

Scale to more teams; add second model checks.

Training matters

  • •Short, hands‑on sessions (60 minutes) beat long decks.
  • •Weekly drills on prompts and data checks build habits.
  • •Share a small prompt library; update it monthly.

Things you could do with AI today.

📧

Email

Triage inbox and draft replies.

Save 45–60 min/day
📝

Meetings

Turn recordings into action lists.

Save 20–30 min/meeting
📊

Data

Pull numbers from PDFs and sheets.

Save 2–4 hrs/report
✍️

Docs

First drafts of guides and proposals.

Save 1–3 hrs/doc
🔍

Research

Quick web scans with sources.

Save 3–6 hrs/project

AI agents (for AirPro)

Small programs that work for you, with your rules. Start simple and keep a human in the loop.

What to expect

  • •Works from your data: Excel, SharePoint, email, Teams.
  • •Reads and compiles; asks before sending or changing anything.
  • •Needs clear rules, named ranges, and test data.
  • •Mistakes happen if data is messy; review logs and improve.
  • •Runs on a schedule (hourly/daily) or a trigger (new row, email received).

Likely use cases

  • •Daily readiness bot: Pull Fleet/Defects/Maint tabs → build report → post to Teams and email ops.
  • •Defect triage: Flag high‑impact items, missing fields, or overdue actions; create Planner tasks.
  • •Document digester: Summarize new PDFs with key changes.
  • •Data QA: Spot anomalies in spreadsheets.

Agent timeline

Agents are new, but they will rapidly grow in importance and impact. Here's how AirPro might implement them.

1

Week 1-2

Pick one workflow (e.g., readiness). Create read‑only access and sample data.

2

Week 3-4

Build the bot in staging. Define the message format and approval step.

3

Month 2

Pilot with 3–5 users. Track errors; tune prompts and rules.

4

Month 3

Expand to defect triage and maintenance watch. Measure time saved.

Safety & access

  • •Start read‑only. Give least privilege and log every run.
  • •No secrets or regulated data in prompts unless under enterprise controls.
  • •Use a staging sheet first; promote after checks pass.
  • •Notify in Teams; require "Approve" to send email or write back.

Key points

  • •Do not lock into one model. Use two or more.
  • •Keep private data out unless on enterprise controls.
  • •Make AI show its work, especially with spreadsheets.
  • •Phones make AI handy: voice, photos, quick checks.
  • •Ship one mini‑app in 3 weeks; scale from there.

AI is a helper, not a boss. Pair strong Conversational models with your Knowledge systems, and use more than one model to stay sharp, safe, and fast.

This document was created by Lemonbrand for AirPro.

Lemonbrand Logo