Inbox and document drafting
Turn rough notes into first drafts, meeting summaries, supplier replies and internal updates while keeping a human owner on every send.
AI training for SMEs
Teach the team what AI can touch, and what it must not.
Practical training for small teams using ChatGPT, Microsoft Copilot and Claude. The output is a working use-case list, review gates, Microsoft 365 guardrails and a short AI policy people can actually follow.
Safe use cases
Start with low-risk work people already understand.
The training starts with useful daily tasks, then adds boundaries around data, review and final decisions.
Policy and process cleanup
Convert existing procedures into clearer checklists, onboarding notes and support scripts without inventing decisions the business has not made.
Spreadsheet and research assistance
Use AI for first-pass categorisation, comparison tables and research notes, with source checks and limits around private data.
Risky tasks
The blocked list matters as much as the allowed list.
Small teams need plain rules for sensitive work, production changes and decisions that carry legal, financial or security impact.
- Sending customer, employee or supplier data into tools without approval.
- Letting AI answer legal, HR, finance or security questions without owner review.
- Copying generated PowerShell, Excel formulas or policy text into production without testing.
- Treating a polished answer as true when there is no source, log, screenshot or second check.
Tool workflows
ChatGPT, Copilot and Claude each get a lane.
The point is not a tool tour. Each tool gets a safe operating model, example prompts and clear stop conditions.
ChatGPT
Good for drafting, summarising and structured thinking. The training covers prompt shape, source checks, data boundaries and when to stop using it.
Microsoft Copilot
Useful when Microsoft 365 permissions are already sane. The workflow starts with SharePoint, OneDrive, Teams and sensitivity-label hygiene.
Claude
Strong for long documents, policies and review passes. The rule is the same: bounded context, clear owner, visible evidence, no silent decisions.
Review gates
Every AI-assisted output keeps a human owner.
Review gates keep AI useful without letting it become an invisible decision maker inside the business.
- Name the owner before using AI for business-facing output.
- Keep source material, prompt intent and final decision separate.
- Require a second human review for customers, contracts, HR, finance and security.
- Record the accepted use cases and blocked use cases in a one-page policy.
Microsoft 365 guardrails
Copilot only works safely when permissions are tidy.
- Check Microsoft 365 sharing before Copilot touches live business content.
- Separate public prompts from internal data, credentials and customer records.
- Use Conditional Access, MFA and device trust before expanding AI tool access.
- Keep logs and evidence for AI-assisted work that affects customers or suppliers.
Training format
Short workshop, practical examples, written rules.
The engagement is built for owner-led and manager-led teams that need useful training without weeks of workshops.
Step 1
Discovery
Map the tools in use, the risky data, the daily workflows and the people who approve changes.
Step 2
Workshop
Run practical exercises in ChatGPT, Copilot and Claude against safe sample work from the business.
Step 3
Policy
Leave a short policy, use-case list, risky-task list and review gate model the team can keep using.
Policy preview
One page beats a policy nobody reads.
The policy is deliberately short: what is allowed, what is blocked, and which work needs review before it leaves the business.
Draft shape
- Approved tasks: drafting, summarising, planning, first-pass research and internal rewrite work.
- Blocked tasks: secrets, credentials, sensitive customer data, legal decisions and unattended production changes.
- Review gates: named owner, source check, second review for high-impact work and evidence kept with the output.
Talk to Gyorgy
Send the tools your team already uses and the risk you are worried about.
I will reply with a sensible training shape, the guardrails to check first and the level of Microsoft 365 cleanup needed before wider AI rollout.