Almost a year after ChatGPT entered our lives, the conversation about artificial intelligence has moved from “wow” to “fine, but what do I do with it on Monday morning?”. As an advisor, I spend a lot of time talking to startup founders and small business owners, and the question I hear most often is not about the future of humanity. It is much simpler: where do I start?
Back in spring I wrote about how I use ChatGPT in business. Since then, the tools have multiplied and matured: ChatGPT has had custom instructions since July, DALL-E 3 landed directly inside ChatGPT in October, Anthropic released Claude 2, Google has Bard, and as of 1 November Microsoft 365 Copilot is generally available to enterprise customers. Canva and Notion have added AI features too. So I put together a practical list for an ordinary small business: seven things you can automate, at least partly, starting tomorrow.
1. Customer emails and FAQ replies
Every small company answers the same questions dozens of times: how much is shipping, how do I return something, when will my order arrive, can I get an invoice. What to do: collect your 20–30 most frequent questions and your best answers in one document, then ask the assistant to draft replies based on them, in your company’s tone. With ChatGPT’s custom instructions you tell it once who you are, how you talk to customers and what you never promise.
The tools: ChatGPT, Claude or Bard for drafts, plus Zapier or Make if you want incoming emails routed to the assistant automatically and the draft dropped into your inbox. Watch out: a human reads every draft before it goes out. AI can calmly invent a return policy you don’t have.
2. Content for social media, newsletters and product descriptions
This is probably the quickest win. Say you run an online store with a few hundred products and descriptions copied from the supplier. Give the assistant the spec sheet and ask for a unique description written for the customer. For social media, ask for ten post ideas a month and pick three. Canva now has AI features for text and images, and DALL-E 3 in ChatGPT can produce simple illustrations.
Watch out for two things: check every number and every technical claim about the product, and don’t let your brand voice turn generic. “Correct but boring” copy is the biggest risk. As for images, I covered that when I tested DALL-E 2 and Midjourney: great for ideas and illustrations, less so for your actual product.
3. Meeting notes and summaries
Raise your hand if you have ever left a meeting without knowing exactly what was decided. In practice: record the meeting (with everyone’s consent), get the transcript from your video-call tool or a transcription service, then ask the assistant for a summary in three parts: what was decided, who does what, by when. For larger companies living in Microsoft 365, Copilot promises exactly this inside Teams, but for a small business the manual route works fine.
Watch out: a meeting with a client or with sensitive data shouldn’t be uploaded just anywhere. Read the tool’s terms and find out whether your data is used for training.
4. Offers, contracts and documents
The first version of a sales offer, an internal procedure or a standard contract takes hours. With AI, it takes minutes. Give it the structure, the terms and an old example, and it hands you a clean draft. It also works the other way round: give it a contract you received and ask it to explain the clauses and flag anything unusual.
Watch out, in capital letters: AI is not a lawyer. Any important document that binds you legally or financially goes to a lawyer. And your partners’ identification details have no place in a free chat tool.
5. Data analysis in Excel and Google Sheets
Plenty of small business owners have the data but not the time or patience to analyse it. AI is surprisingly useful here: describe your table and ask for the formula that calculates margin per product, clean up a list where names are spelled five different ways, or suggest a simple monthly report. In ChatGPT Plus, with GPT-4, you can upload a file directly and ask it to run the analysis.
Watch out: check the results on a few rows by hand. And anonymise the data before uploading it: you don’t need customer names to find out which product sells best.
6. Recruitment: job ads and screening support
A good job ad is harder to write than it looks. AI gives you a clear first version, with realistic responsibilities and requirements, and helps you prepare interview questions. It can also summarise the CVs you receive.
Be more careful here than anywhere else: models can carry biases, and the decision to reject a candidate is not one you hand to a machine. CVs are personal data, so GDPR applies in full; if you’re unsure what you’re allowed to do, reread my practical guide to GDPR. European rules are coming too: recruitment is among the areas the upcoming EU regulation on AI treats as high-risk.
7. Internal knowledge and onboarding
In most small companies, procedures live in the heads of two or three people. When one of them goes on holiday, everything stalls. The fix is old; AI just makes it easier: write the procedures down once (how to issue an invoice, how to handle a return, how to answer a complaint), keep them in a shared place, for example Notion, which now has AI built in, and let a new hire ask the assistant instead of interrupting a busy colleague.
Watch out: the assistant is only as good as the documents. If a procedure is outdated, the answer will be wrong, but delivered with confidence. Someone has to keep them up to date.
How to start without breaking your company
Don’t do all seven at once. Here is what works, from what I see in the companies I advise:
- One process first. Pick the most repetitive, most boring one.
- Measure. Write down how long it took before and how long it takes now. Without your own number, the discussion stays at the level of impressions.
- Protect customer data. Set clear rules about what never goes into an AI tool.
- Train the team. One hour of training with real examples is worth more than ten articles.
- Make someone responsible. One person who tests, picks the tools and owns the result.
As I said about GPT-4 and the end of jobs: AI won’t replace your people, but people who know how to use it will outpace those who don’t. In small companies, that gap shows up faster than anywhere else.
What have you already automated in your business, and what disappointed you? Let me know, I’m curious which of these seven has given you the most time back.