Assistants like Claude, Copilot and Gemini draft, summarize and analyze in seconds; Perplexity researches with citations; Granola and Otter turn meetings into notes nobody had to type. The gains are measurable. But every one works by reading your data — and most staff started before anyone decided which tools were approved or what may be pasted in.
The benefits — measured, not promised.
|
|
|
|
| saved per 40-hour week by the average generative-AI user. St. Louis Fed, 2025 | less time on professional writing tasks, with quality up 18%. Science, 2023 | more customer issues resolved per hour, most for newer staff. 5,000-agent field study |
The tools — what each is good for, and what to check first.
| Tool | Good for | Check before rollout |
|---|---|---|
| Claude | Drafting, analysis, long documents | Business plan, not personal accounts |
| Microsoft Copilot | Work inside Outlook, Word and Teams | SharePoint and Teams permissions |
| Google Gemini | Work inside Gmail, Docs and Drive | Drive sharing settings |
| Perplexity | Research with cited sources | Treat queries as public; verify citations |
| Granola | Meeting notes from your own device | Tell participants; review before sharing |
| Otter | Meeting transcription and summaries | Auto-join, consent and auto-share settings |
| Figure 1 — Share of data put into AI that is sensitive | Figure 2 — The AI oversight gap |
 |
 |
| Tripled in two years. Cyberhaven | Shadow AI added $670,000 per breach. IBM, 2025 |
Security concerns, and the challenges behind them
Data leakage |
Unwanted access |
Mistaken publishing |
|
Client names, contracts and code pasted into a prompt leave your control — worst on personal accounts. 39.7% of AI interactions involve sensitive data (Cyberhaven, 2026). |
Copilot and Gemini see everything the user can, so years of over-shared folders become answers. 99% of organizations have sensitive data AI can surface (Varonis, 2025). |
Notetakers can keep recording after you leave, then email the summary to every attendee. Alleged: Otter kept recording an investor call after the host left (2025). |
Shadow AI |
Confident errors |
Consent and compliance |
|
Staff use whatever helps, on whatever account. You cannot protect tools you do not know exist. 78% of AI users bring their own tools; 80% at smaller firms (Microsoft, 2024). |
AI writes wrong answers as fluently as right ones. Output needs checking by someone who knows. Developers were 19% slower with AI, believing they were 20% faster (METR, 2025). |
Recording and privacy law apply to AI transcripts. Many notetakers ask only the host. Four class actions against Otter were consolidated in 2025. |
What this means for us
- Approve a short list of tools, on business accounts only.
- Fix folder permissions before switching on Copilot or Gemini.
- Announce notetakers, and review summaries before sharing.
Adopt AI without the exposure.
info@mctechconsulting.com
mctechconsulting.com
