AI can draft, sort, summarise, compare, route, check, suggest, classify, and organise.
Below is a shopping list answering:
What kind of work is this?
What kind of work can AI do?
What could go wrong if AI got it wrong?
| Type of work | AI is good for | Be careful when | Example tools / also see |
|---|---|---|---|
| Writing first drafts | Emails, notes, proposal sections, short updates, internal messages | The message depends on tone, relationship history, tension, apology, persuasion, or nuance | ChatGPT, Claude, Gemini, Grammarly. Also see: Write Better Emails Faster |
| Client and stakeholder updates | Turning messy notes into clearer updates, shortening long explanations, structuring status messages | The update carries risk, bad news, accountability, or relationship sensitivity | ChatGPT, Claude, Gemini. Also see: AI Clinic - Clear Client & Stakeholder Updates Without the Back-and-Forth |
| Summarising and extracting information | Pulling key points from documents, calls, emails, reports, transcripts | The source is incomplete, legally sensitive, financially important, or nobody checks the summary | NotebookLM, Fathom, Fireflies, Otter |
| Marketing production | Repurposing content, drafting campaign variants, generating headlines, social posts, email drafts | AI starts setting positioning, inventing claims, or drifting away from the business’s actual point of view | ChatGPT, Jasper, Copy.ai, HubSpot AI. Also see: AI for Marketing That Runs Itself |
| Sales support | Follow-ups, call summaries, CRM notes, proposal skeletons, objection lists, next-step prompts | The situation depends on negotiation, pricing, relationship judgement, or reading the customer properly | HubSpot AI, Salesforce Einstein, Pipedrive AI, Gong. Also see: Sell Smarter with AI: A 3-Step Sales Guide for SMEs |
| CRM and pipeline discipline | Spotting stale deals, prompting follow-ups, turning records into next actions, summarising deal status | AI decorates a weak CRM instead of making it action-focused | HubSpot, Salesforce, Zoho, Pipedrive. Also see: Tune Your CRM for Actions, Not Storage |
| Customer support triage | Classifying tickets, drafting replies, suggesting help articles, spotting repeated questions | Customers are angry, confused, high-value, vulnerable, or dealing with an exception | Zendesk AI, Intercom, Freshdesk, Help Scout |
| Planning and budgeting | Surfacing assumptions, generating scenarios, preparing questions before the spreadsheet, stress-testing plans | AI replaces judgement or produces confident-looking guesses without enough evidence | ChatGPT, Claude, Gemini, Microsoft Copilot. Also see: Strengthen Your Planning with AI Before You Open the Spreadsheet |
| Process improvement | Mapping steps, finding duplication, creating checklists, SOPs, templates, handover notes | The business automates the existing mess instead of redesigning it first | Zapier, Make, Notion AI, ChatGPT, Miro AI |
| Internal knowledge capture | Turning repeated explanations into onboarding notes, FAQs, guides, training material, internal playbooks | The knowledge is sensitive, outdated, contested, or requires senior judgement | Notion AI, Confluence AI, Guru, Slite, NotebookLM |
| Finance/admin support | Categorising, preparing questions, summarising debtors, checking consistency, spotting anomalies | It touches cash, tax, compliance, credit control, or final decisions | Xero, QuickBooks, Dext, Excel Copilot |
| Decision process | Comparing options, surfacing assumptions, clarifying trade-offs, generating questions | AI is asked to choose, recommend, or carry responsibility for the decision | ChatGPT, Claude, Gemini, Perplexity. Also see: AI can compare two options. It can’t tell you what’s missing. |
| Comparison and benchmarking | Finding typical ranges, structuring comparisons, showing possible factors behind outcomes | The comparison itself may be wrong, incomplete, or based on the wrong benchmark | ChatGPT, Claude, Gemini, Perplexity. Also see: AI can give you a comparison. It can’t tell you if you’re comparing to the right thing. |
The pattern is simple.
AI is useful when the task has three features:
It becomes risky when one of those is missing.
Use AI to reduce unnecessary work.
Use it to draft, summarise, structure, classify, compare, prepare, and organise.
But keep people responsible for judgement, trust, exceptions, and outcomes.
AI can carry a lot of the work around the decision. It should not quietly become the person making it.
If this feels familiar, start here:
👉 Run the Second Look Decision Diagnostic to see what’s missing before you decide
👉See related business decision
👉 📖 Read more on AI Clinic blog
You can continue with making the decision afterwwards.