A role-by-role breakdown of where AI tools genuinely help — strategy, marketing, sales, operations, customer support, finance — and what to use for each.
Building on the previous post about general principles, here's the practical role-by-role breakdown. For each role, the genuinely-useful uses, the tools we'd recommend trying first, and what to keep doing the human way.
Strategy
**Useful for:** Generating options, stress-testing your assumptions, summarising long industry reports.
**Tools:** Claude for thinking partner conversations. Perplexity for synthesised research with citations. The Harvard Business Review hasn't been replaced; it's just easier to skim.
**Keep human:** The actual decision. Your gut, informed by the customer-discovery work we covered in this post, is the only thing that knows what fits your specific situation.
Marketing
**Useful for:** First drafts of blog posts, ad copy variants, image alt-text, social-media captions, A/B test hypotheses, SEO keyword research, draft press releases.
**Tools:** ChatGPT or Claude for copy. Canva's Magic Studio for visual mockups. Surfer SEO for content-fit analysis (with skepticism — see our SEO strategy resource for why keyword density isn't a strategy).
**Keep human:** Brand voice, the actual decision about what story to tell, photo selection. AI-written copy is recognisable and your customers can tell — see our piece on branding as a promise.
Sales
**Useful for:** Drafting follow-up emails, summarising call recordings, suggesting next-step questions, pre-call research on a prospect.
**Tools:** Gong for call analysis. Clay for outbound research. HubSpot's AI features if you're already on HubSpot.
**Keep human:** The actual conversation. The relationship. The judgement on whether this prospect is a fit. The negotiation.
Operations
**Useful for:** Project status summaries, action-item extraction from meeting notes, generating documentation from chat logs, fixing little scripts that automate repetitive work.
**Tools:** Notion AI for in-context document work. GitHub Copilot for engineering ops. Zapier AI Actions for connecting tools.
**Keep human:** Hiring decisions, performance reviews, anything HR-adjacent. The risk of bias plus the personal-data implications make this a no-go zone for current AI tools.
Customer support
**Useful for:** Suggested-reply triage, FAQ generation, sentiment analysis on incoming volume, language detection.
**Tools:** Intercom Fin, Zendesk AI, Cresta for call-centre work.
**Keep human:** The actual reply that goes to a frustrated customer. Always. The cost of an AI reply that misses the emotion is much higher than the cost of the human time. This is the ethical engagement line — using AI to triage is fine; using AI as the front-line voice is rarely OK.
Finance and admin
**Useful for:** Categorising transactions in your bookkeeping software, drafting invoices, parsing receipts.
**Tools:** Most modern bookkeeping platforms (Xero, QuickBooks, Procountor in Finland) now have built-in AI categorisation that's good enough for most small businesses.
**Keep human:** Tax decisions, signing off on the year-end accounts, anything where being wrong has financial regulatory consequences. Talk to a real accountant for those.
Engineering / web
**Useful for:** Boilerplate, refactoring, test generation, debugging hints, code review companion, translating between languages or frameworks.
**Tools:** Cursor, GitHub Copilot, Aider for terminal-driven workflows.
**Keep human:** Architecture decisions, security-sensitive code, anything where the cost of subtle bugs is high. The hard part of engineering isn't typing — it's deciding what to type. AI helps with the typing.
The cross-cutting principle
A useful frame from our take on AI for owners: AI accelerates the parts of work that are *generative* (producing first drafts, exploring options), but doesn't replace the parts that are *judgmental* (deciding which option is right). When a tool's marketing pitch promises both, it's overstating one of them.
The result: AI is best at giving a solo business owner the leverage of a small team — drafts, summaries, triage — without the management overhead. It's worst at the parts that made you start the business in the first place: relationships, judgement, vision.
Where to read next
- Building Effective Agents — Anthropic's engineering blog on what AI agents can and can't do well.
- Don Norman, The Design of Future Things — old book, completely relevant; how to design for human-AI collaboration without dropping responsibility on the human alone.
- The Stack's AI for Small Business newsletter — reasonably hype-resistant ongoing coverage.
If you want help wiring any of this into your business' actual systems — your website, your booking flow, your customer journey — that's where we come in. Get started.