Disclosure: Some links below are affiliate links. If you buy through them, this site earns a small commission at no extra cost. Editorial recommendations are never influenced by affiliate rates.

Your organisation has just given you access to Claude. Maybe it arrived via a company-wide rollout, a new seat in a Team plan, or IT quietly sent around a login. Either way, you've got a tab open and you're not quite sure what to do with it beyond "ask it things."

This guide covers what actually matters for day-to-day workplace use in 2026: how to get it talking to your CRM, how to turn it into a presentation, how to make sense of a large dataset, and how to do all of this without accidentally exposing client data or wasting half your afternoon on a hallucinated statistic.

No fluff. No "Claude will transform your workflow forever." Just what it does well, what it doesn't, and how to get started on the things most office users actually need.


What you're working with

Claude is a large language model made by Anthropic. In practical terms, it reads text (and files, and images) you give it, and responds. It doesn't browse the internet unless it says it is. It doesn't know what happened last week. It doesn't have access to your calendar or inbox unless you connect those tools. What it's good at is understanding complex requests, processing large documents, writing with nuance, and handling structured tasks like "rewrite this email in a more formal tone" or "pull the key numbers from this 80-page report into a summary table."

As of mid-2026, Claude runs in four main flavours at work:

Your company's IT or admin team chose a plan when they signed up. This matters because what you can do depends on it.

Free: limited messages, no connectors, no file creation. Pro (USD $20/month, roughly NZD 33) : filecreation, projects, connectors, extendedreasoning.Enoughformostindividuals. *  * Max *  * (100-$200 USD/month, NZD 165−330): much higher usage limits. Worth it if you're using Claude heavily every day. Team (from USD 20/seat/month, NZD 33): organisation-managed accounts, shared projects, admin controls. Premium seats at USD 100−125/seat add Claude Code access. Enterprise: custom pricing, usually 70+ seats. Includes HIPAA readiness, SAML SSO, data residency controls, and a 1 million token context window.

If your company gave you access and you're unsure which plan you're on, check Settings > Account in Claude.ai. It'll say.


Connecting Claude to HubSpot

HubSpot has a native connector for Claude that's worth setting up if your team uses it for CRM, marketing, or support. Anthropic and HubSpot built this together, and it lives directly inside Claude's chat interface. You don't need IT to wire anything up, and there's no API key management.

To connect:

  1. Open Claude.ai and go to Settings > Connectors.
  2. Click Browse Connectors and find HubSpot under the Web Connectors tab.
  3. Click Add to your team, then Connect. You'll be sent to HubSpot's portal selector to choose your account.
  4. In any new chat, enable HubSpot as a data source via the toolbar icon (it looks like a small plug icon next to the message bar).

Once connected, you can ask Claude things like:

Claude can also write records back to HubSpot: create contacts, update deal stages, log notes and tasks. The integration is read-write, not just read-only.

Requirements: you need a paid Claude subscription (Pro, Max, Team, or Enterprise) and an active HubSpot user account. It doesn't work on the free Claude tier.

What it's not: it's not a replacement for HubSpot's native reporting or automation tools. It's a natural-language interface into your existing CRM data. Think of it as being able to ask your HubSpot account questions in plain English rather than building a filter view every time.

One real use case: a sales rep asked Claude to compare all deals closed in Q1 vs Q4 of the previous year, broken down by rep and source, then format the output as a table and write a one-paragraph summary for the Monday standup. That took about 30 seconds. Without the connector, it would have taken three separate reports and manual copying.


Building presentations and slide decks

Most people try Claude for slides the wrong way: they ask for bullet points and paste them into PowerPoint themselves. You can do better.

There are three solid approaches depending on what you need:

Claude Design (claude.ai/design)

Anthropic's newest tool, available to Pro plan subscribers and above. You describe the deck you want, Claude generates a fully formatted, clickable HTML slide deck in a canvas on the right side of your screen. It accepts uploads (Word docs, PDFs, spreadsheets) as source material. Export to PPTX, PDF, or standalone HTML when you're done.

The standout feature: if you paste in your company website URL or upload a brand doc, Claude Design infers your colour palette, fonts, and layout conventions before it generates anything. First-draft consistency is noticeably better than working from a blank prompt.

Realistic timeline: simple decks (10-15 slides, one data source) typically take two to four back-and-forth exchanges. Complex decks with multiple data sources and conditional logic need more iteration.

Credit budget matters here. On a Pro plan, the Design weekly budget is small. Three or four complex decks and you'll hit the wall. If your team is using this regularly, Max 5x or a Premium Team seat is the practical minimum.

Claude for PowerPoint (Microsoft 365 add-in)

For teams already working inside Office, this is often the better choice. Install it from Microsoft AppSource (search "Claude for PowerPoint"), sign in with your Claude account, and Claude appears as a side panel inside PowerPoint.

You describe what you want, and Claude generates slides that conform to your existing template's fonts, colours, and layout masters. This means brand consistency without extra effort on your part. You can also highlight existing slides and ask Claude to rewrite, simplify, or expand specific sections.

Works well for: turning a research document into a formatted deck, rewriting dense text slides into cleaner bullet points, and generating speaker notes from slide content.

Using Artifacts (for quick or technical decks)

In any Claude.ai chat, ask Claude to build a "presentation as an artifact." It'll generate an HTML slide deck you can cycle through in the preview panel. Less polished than Claude Design, but faster and doesn't eat into your Design budget. Good for internal working decks where visual polish isn't the priority.


Working with PDFs and documents

Claude handles document work well. As of April 2026, it can both read and create PDF, Word, Excel, and PowerPoint files directly, without needing a third-party converter.

Reading and extracting from existing documents

Upload a PDF (or multiple) into any Claude chat. It reads the whole thing, including tables and multi-column layouts. Useful patterns:

File size limit: up to 30 MB per file, up to 20 files per chat session. For a 200-page PDF, that's fine. For bulk document processing across hundreds of files, you'd need the API or a tool like Lumin (which integrates Claude for collaborative PDF annotation and filling).

Creating documents from scratch

Type a request like "Create a formal quotation letter for a NZD $12,000 website project, 30-day payment terms, GST exclusive," and Claude will produce a formatted .docx file you can download and open directly in Word or Google Docs. Same for PDFs and spreadsheets.

This is especially useful for first drafts of standard documents: proposals, meeting agendas, briefs, SOWs. Claude gets you 80% there in seconds; you handle the last 20% with specific context it doesn't have.

Form completion

Upload a PDF form with fields, describe your details, and Claude fills it. The output is a completed document you can save or send. Not 100% reliable with very complex form logic or unusual layouts, but works well on standard business forms.


Analysing and reporting on large data

This is one of Claude's strongest real-world use cases right now, especially with the large context window on newer plans.

Uploading spreadsheets and CSVs

Drop a CSV or Excel file directly into the chat. Claude can handle files up to 30 MB and reads multi-sheet workbooks. On Enterprise plans (1M token context window), you can upload multiple large files in one session.

Claude for Excel (the Microsoft 365 add-in, released to all Pro users in January 2026) lets you work directly inside Excel without switching tabs. Open your spreadsheet, activate the Claude panel, and describe what you want:

Claude can write Excel formulas, pivot tables, and conditional formatting instructions back to you in plain English. You apply them; it doesn't modify your file directly (which is the safe behaviour, even if it occasionally feels slow).

Large report analysis

Paste a 100-page market report, an annual return, or a transcript from a long meeting, and Claude can summarise it, extract specific data, identify discrepancies, or reformat it as needed. The practical limit is the context window: on Pro and Max plans, you get a 200,000 token context (roughly 150,000 words). On Enterprise, it's 1 million tokens. Most business documents fit comfortably.

Where it gets unreliable: once the chat conversation itself grows very long (multiple hours of back-and-forth with large files), Claude can start losing track of earlier context. If you notice it contradicting itself or forgetting instructions you gave earlier, start a fresh chat and re-upload.

Producing reports automatically

Claude can take raw data and produce formatted, narrative reports. Try:

"Here's our Q1 sales CSV. Write an executive summary (3-4 paragraphs), a bullet-point highlights section for the board pack, and a table showing month-over-month change for the top five products by revenue."

The output is copy-paste-ready. You won't get a stunning data visualisation natively in the chat (though Claude Design can generate charts if you're using that tool), but you'll get structured, accurate narrative and tables based on the numbers you gave it.


Privacy and data: what you need to know before you paste anything

This is the most important section for workplace use, and the one most people skip.

What happens to what you type

Anthropic's data handling depends on which plan your organisation is on:

Free and Pro plans: conversations are not used for AI training by default (you can opt in if you want to). Data may be stored for up to 30 days if you've opted out of training. Human reviewers may occasionally read conversations for safety and quality purposes.

Team plans: conversations are not used for training. Admin controls let your IT team manage what's stored.

Enterprise: includes full data retention controls, SAML SSO, HIPAA readiness, and options for data residency. For organisations handling sensitive client data, regulated health information, or anything under a strict confidentiality obligation, Enterprise is where these controls live.

Zero Data Retention (ZDR): available via API arrangement for sensitive sectors. Conversations are processed for real-time safety only; nothing is stored server-side. Relevant for healthcare, legal, and finance teams handling regulated data.

Practical rules for everyday use

Don't paste real client data into a Free or Pro account. This includes customer names, email addresses, financial records, health information, or anything your clients would expect to stay private. Even though training opt-out is default, conversations still pass through Anthropic's servers and may be subject to review.

Don't paste passwords, API keys, or internal credentials. Claude doesn't need them for most tasks, and there's no good reason to take the risk.

Anonymise before uploading where possible. If you need Claude to analyse a spreadsheet of customer transactions, replace real names with "Customer A, Customer B" before uploading. Claude works just as well on anonymised data.

Check your organisation's AI policy first. Many companies now have a formal AI acceptable-use policy that specifies what data can go into which tools. If yours does, read it. If it doesn't, treat Claude as a non-NDA'd external contractor: share what you would share with a capable freelancer, not what lives in your ERP or client database.

Enterprise features worth asking IT about

If your organisation is running a Team or Enterprise plan and handles sensitive information, these features are worth confirming are enabled:


What Claude gets wrong (and how to catch it)

Claude is not infallible. Knowing the failure modes saves you embarrassment.

Hallucinated facts and citations. Claude will sometimes state a specific statistic or cite a study that doesn't exist. It sounds confident. It's wrong. If you're including a fact in something external-facing (a report, a client presentation, a public document), verify it against a real source. Don't trust Claude's citations without checking the URL yourself.

Outdated information. As of May 2026, Claude Opus 4.7 has a training cutoff of January 2026. Events, product releases, pricing changes, and regulatory updates after that date aren't in its knowledge. If you're writing about something recent (a new piece of legislation, a competitor's pricing, a product launch), always verify against live sources.

Stale context in long conversations. If you've been in the same chat for two hours and uploaded several files, Claude can lose track of earlier instructions. Symptoms: it starts contradicting earlier responses, ignores a rule you specified, or produces something that conflicts with a document you uploaded at the start. Fix: start a fresh chat and re-state the context.

Overconfident tone on uncertain things. Claude doesn't hedge much by default. An answer that reads authoritative may reflect genuine knowledge or may be a plausible-sounding guess. The fix is to explicitly ask: "How confident are you in this? Are there aspects you're uncertain about?" It will tell you.

Not a real-time data source. Claude can't check today's exchange rates, live share prices, current court filings, or live inventory unless web search is enabled and it actually uses it. If it gives you a current NZD/USD rate, it's a guess from training data. Get live rates from your bank.


Getting started: five things to try in your first week

If you've made it this far and want somewhere to start, try these:

1. Summarise a long document. Upload a PDF (a contract, a report, meeting notes from the last board session) and ask: "Summarise the key points in plain English, then flag anything that needs a decision or action."

2. Draft an email you've been putting off. Describe the situation and your desired outcome: "I need to push back on a client's request to extend our project without additional budget. I want to be firm but not damage the relationship. Here are the relevant emails: [paste them]." Revise the output as needed.

3. Connect HubSpot and ask one question. Once connected, ask Claude to pull the last 30 days of new contacts and summarise acquisition sources. See how the output compares to what you'd normally build in HubSpot's reporting view.

4. Upload a spreadsheet and ask for insights. Paste or upload your last month's sales data, a budget vs actual comparison, or a staff roster. Ask: "What's the most important pattern in this data that I should be paying attention to?" See what it surfaces.

5. Build a deck from a doc. Take a report or briefing document, upload it to Claude Design (or the PowerPoint add-in), and ask it to turn it into a 10-slide presentation. Adjust the output once you see it. It's faster than starting in PowerPoint from a blank slide.


The realistic verdict

Claude is a genuinely useful tool for the kinds of tasks most office workers deal with daily: processing documents, drafting communications, pulling structure from messy data, and building presentation material. The HubSpot connector and the Microsoft 365 add-ins make the integration into existing workflows more practical than it was 12 months ago.

The limits are real. It's not a replacement for a domain expert, a live data feed, or a tool with actual access to your internal systems (unless IT has built that). It hallucinates. It doesn't know what happened last month. It's not appropriate for processing sensitive client data unless you're on an Enterprise plan with proper controls in place.

For NZ teams: most of the friction comes from pricing in USD. Pro at $20/month is NZD $33 at current exchange rates, Team Premium seats are NZD 165−206/seat. The value case stacks up if you're replacing a few hours of manual work per week. If you're on a company-managed plan, the licensing decision isn't yours, which removes the barrier entirely.

Start with the five tasks above. Most people find the right use cases for their own work within a few days, once they've stopped trying to treat Claude like a search engine.

TD
Toby Downs is an independent tech writer based in New Zealand, covering SaaS, AI tools, and business software for tpdowns.com. No paid placements, no sponsored opinions — just research.