AI at work — getting the most from Claude and Copilot

approved

Practical, safe use of Claude and Microsoft 365 Copilot: which tool for which job, prompt patterns that work, and the data boundaries every business needs before rolling AI out.

Audiences: customer, ae, tech · Tags: ai, claude, copilot, productivity · Last verified: 2026-07-16

Teach module

# Teach — Claude and Copilot in practice

## Core truths

- Claude and Microsoft 365 Copilot solve different jobs: Copilot works inside M365 apps with your tenant's data (email, documents, meetings); Claude excels at reasoning, drafting, analysis, and longer-form work you bring to it.
- Copilot's usefulness depends directly on tenant hygiene: it can only surface what permissions allow, so oversharing in SharePoint and Teams becomes an AI-shaped data exposure problem.
- Good prompting is context plus intent: state the role, the audience, the format you want, and give source material — quality in, quality out.
- Iteration beats perfection: treat the first response as a draft to steer, not a final answer to accept or reject.
- AI output requires human review before it leaves the business, especially anything with figures, legal implications, or client commitments.

## Common myths to correct

- "AI will replace the team" — in SMB practice it removes drafting and summarising drudgery; judgment, relationships, and accountability stay human.
- "Our data trains the public model" — business-tier offerings from both vendors state customer data is not used to train foundation models; the real risk is internal oversharing, not vendor training.
- "One tool does everything" — matching the tool to the job (in-tenant context versus deep reasoning) is where the productivity gain actually comes from.

## Pitfalls

- Rolling out Copilot before a permissions review: AI makes existing oversharing instantly discoverable.
- No acceptable-use policy: staff will use AI regardless; ungoverned use means unknown data flows.
- Measuring adoption by licence count instead of by workflows changed.
- Pasting client-confidential data into personal (non-business) AI accounts.

Apply module

# Apply — discovery, objections, and talk tracks

## Discovery questions

- Where does your team lose the most hours to writing, summarising, or re-keying information?
- Do you know which AI tools your staff already use, and on which accounts?
- If Copilot could see everything your most junior employee can see, would anything worry you?
- Do you have an acceptable-use policy for AI, and has anyone read it?
- Which single workflow, if it were twice as fast, would matter most to the business?

## Objections and responses

- "AI makes things up." — It can, which is why we pair it with grounding on your documents and a human review step; the process is the safety net, not blind trust.
- "We are too regulated for AI." — Regulated firms benefit most from governed AI; the alternative is staff using personal accounts with no controls at all.
- "We tried it and nobody used it." — Adoption fails without workflow targeting; we start with two or three named workflows per team, not a licence and good luck.

## Talk tracks

- "Right tool, right job, right guardrails" — the three-part frame for any AI conversation.
- Lead with the permissions story for business owners: AI readiness is mostly data hygiene, which is a service we can deliver.
- For staff training: show a bad prompt then a good prompt on the same task — the before/after sells better than any slide.

Convert module

# Convert — CTAs and next steps

## Approved CTAs

- Book an AI readiness assessment: permissions review, data hygiene check, and a shortlist of the workflows where AI will pay back first.
- Offer a staff prompt-skills workshop: half a day, hands-on, using the attendees' real work as exercises.
- Internal enablement: each team member brings one recurring task next week and pairs it with an AI-assisted version.

## Next-step framing

- Customer webinars: end on the readiness assessment, positioned as the safe first step before buying licences.
- AE sessions: the goal is booking assessment conversations, not selling Copilot seats directly.
- Never promise specific productivity percentages; anchor on the named workflows identified in discovery.

Demo steps

# Demo — Claude and Copilot side by side

1. Copilot in Outlook: summarise a long email thread and draft a reply in the sender's tone.
2. Copilot in Teams: recap a recorded meeting — decisions, actions, owners.
3. Claude: paste a messy process document and ask for a structured SOP with numbered steps and a review checklist.
4. Prompt upgrade, live: run a vague prompt ("write a proposal"), then the same task with role, audience, format, and source material — compare outputs side by side.
5. Boundary demo: show Copilot respecting permissions by searching for a document the demo account cannot access.
6. Close with the review habit: take one AI draft and mark it up as a human editor would before sending.

## Pre-demo checklist

- Demo tenant has Copilot licences active and sample content (email thread, meeting recording, shared documents) staged.
- No client-identifiable data anywhere in the demo tenant.
- Claude demo uses a business account, and the pasted documents are sanitised samples.

Claims policy

# Claims — Claude and Copilot

## Allowed claims

- Copilot works inside Microsoft 365 apps and respects existing tenant permissions.
- Claude is well suited to drafting, summarising, analysis, and long-document work.
- Business-tier offerings from both vendors state customer data is not used to train foundation models (cite current vendor documentation when asked).
- Effective prompting (role, audience, format, source material) measurably improves output quality.
- AI output should be human-reviewed before external use.

## Forbidden / needs-human-review claims

- Specific licence pricing for Copilot or Claude tiers — verify current pricing before quoting.
- Any guaranteed productivity gain percentage.
- Claims about specific model versions, context window sizes, or feature availability — these change; verify before stating.
- Legal or regulatory compliance assurances (GDPR, industry rules) — route to the compliance conversation, do not assert from the stage.
- Speculation about either vendor's roadmap.