I Asked AI to Run My Business for a Week — Here's What Worked and What Flopped

I spent one full working week using AI tools for every single task my business generates — responding to customer enquiries, writing proposals, creating content, analysing numbers, planning my schedule, and handling correspondence. I went in with genuine curiosity and zero obligation to produce a positive result. Here is the honest account of what saved me time, what embarrassed me, what surprised me, and what I will never do again.

The tools I used: Claude (Anthropic) as my primary writing and thinking assistant, ChatGPT as a secondary tool for comparison, and Google Sheets with a connected AI extension for data analysis. No special integrations, no enterprise tools, no setup beyond what any small business owner could do in an afternoon.

I am sharing this because most AI content is either uncritically enthusiastic or dismissively sceptical — neither of which is useful if you are a small business owner trying to make a practical decision about whether and how to use these tools. This is the honest middle ground.

What Worked Brilliantly: First-Draft Everything

The single most impactful thing AI did for me all week was eliminate the blank page for every piece of writing. Enquiry responses, follow-up emails, proposal outlines, social captions, FAQ updates — I gave Claude a two-sentence summary of what I needed and within ten seconds had a structurally correct, professionally toned first draft to work from.

Starting from a draft rather than a blank page cut my email time by approximately 60%. This is not because the AI drafts were perfect — they were not. They needed editing, personalisation, and tone adjustment. But editing is dramatically faster and cognitively lighter than writing from scratch. By Friday I had answered every email, sent three proposals, and written a week of social content — in approximately two-thirds of my normal time.

What made the drafts good was briefing specificity. On Monday I gave generic prompts: "write a professional reply to this enquiry." The output was generic. By Wednesday I had learned to brief in specifics: "write a reply to this enquiry from a plumber asking about our website build service — focus on our 6-week delivery timeline and reference our previous work with tradespeople. Tone: direct and confident, not corporate." The output was substantially better and required less editing.

What Worked Surprisingly Well: Brainstorming Options

Mid-week I had a genuine business dilemma: a client I liked wanted a project at a budget I was not sure about. I was stuck — going back and forth in my own head without making progress. On impulse I described the situation to Claude and asked: "what are ten ways I could handle this conversation?"

Three of the ten options were things I had already considered. Four were obvious but I had overlooked them in my deliberation. Three were approaches I had genuinely not thought of — including one that I ended up using, which involved separating the scope into phases so the client could start at their budget and expand. The conversation went well. AI did not make the decision — I did. But it expanded my option set in two minutes when I had been stuck for an hour.

I used the same brainstorming approach twice more during the week: once for a marketing problem and once for a difficult piece of feedback I needed to give a supplier. Both times, the AI-generated options list moved me from stuck to having a clear path in under five minutes. This is one of AI's most underrated uses — not to give you answers, but to generate options you would not have considered in your own internal loop.

What Flopped Badly: Skipping the Review

By Thursday I was getting overconfident. I had been editing drafts all week and they were consistently solid. When a proposal needed to go out quickly, I reviewed it more lightly than I should have. The client came back confused about two sections that referenced specific services and processes that sounded plausible but did not match anything I actually offer. Claude had extrapolated from my brief and invented two service components that had no basis in reality.

This is the failure mode everyone warns you about with AI, and it is real: AI does not know what it does not know. It will fill in gaps in your brief with plausible-sounding information that has no factual basis. The proposal would have been sent with inaccurate capability claims if I had not been asked about them. That is the kind of error that damages client relationships and, in regulated contexts, could create legal exposure.

The lesson is rigid: AI output that goes to a client must be reviewed with the same care you would give to anything drafted by a capable but unvetted junior. Fast does not mean unreviewed. The time saving of AI drafting is still significant even after thorough review — but the review cannot be treated as optional.

What Flopped Completely: AI for Decisions

On Wednesday I asked Claude a genuine business question: "Should I take on this project at £4,200 when my usual rate for this scope would be £5,500?" I provided full context: the client relationship, the strategic value, the cash flow position, the opportunity cost. Claude gave me a thoughtful, balanced response that laid out the arguments on both sides and concluded with something like "ultimately this depends on your priorities and risk tolerance."

It was useless. Genuinely useless. Not because the reasoning was wrong — the points were accurate — but because I already knew both sides of the argument. What I needed was a tiebreaker based on judgment about my specific situation. AI cannot provide that. It can give you information and options; it cannot make value judgments on your behalf. Business decisions — particularly those involving money, people, and strategy — are still entirely yours.

The Honest Summary After One Week

AI is a brilliant junior assistant that works at the speed of light, has no ego, and never gets tired. It also occasionally fabricates facts with complete confidence, cannot make judgment calls, and will happily write competent-sounding nonsense if your brief is vague enough. The right mental model is not "AI will handle this" — it is "AI will draft this and I will review it."

The time saving is real and substantial for the right categories of work. Writing first drafts, brainstorming options, summarising information, and formatting structured content — AI is faster and often better than me working alone for these tasks. Decision-making, client relationship judgment, and anything requiring specific knowledge about my business beyond what I briefed — still entirely human. The boundary between those two categories is where you need to be clear-eyed. For a broader look at where AI helps most at the small business level, see our realistic guide to AI for small businesses in 2025.

Dream Code Labs

Web Development & Automation Agency · 7+ years experience

Dream Code Labs is a remote-first development and automation agency specialising in custom websites, AI-powered tools, and workflow automation for marketing agencies and growing SMEs across the UK, US, Canada, and Australia. We have delivered 50+ projects that produce measurable, real-world results.