A Day in the Life of an AI-Assisted Small Business

The business in this example

Most writing about AI at work is either a vendor demo or a warning. Neither tells you what an ordinary Tuesday looks like once the tools are simply part of the furniture, so here is one, told hour by hour.

The business is illustrative but typical of the firms we talk to: a 12-person commercial cleaning and facilities company with around 60 contract clients, two office staff, a sales lead, an operations manager, the owner and a team of supervisors in the field. They use Microsoft 365, Xero, a job management tool and a shared inbox that nobody loves.

They have not bought a grand AI platform. They pay for a business-grade AI assistant for the office staff, and they had a couple of small integrations built. That is the whole setup, and it is closer to what works for a firm this size than anything more ambitious.

7:30am: the morning briefing

Before the owner opens her inbox, a short briefing is waiting. It was assembled overnight from yesterday's completed jobs, missed visits flagged by supervisors, overdue invoices from Xero and anything in the shared inbox tagged urgent.

The AI's job here is narrow. The numbers come straight from the systems; the model writes three short paragraphs saying what changed and what needs a decision today. It does not calculate anything. That design choice matters more than any prompt, and we cover it properly in our post on building a morning business briefing with AI.

Reading it takes four minutes. Before, she spent the first half hour of every day opening five tabs to assemble the same picture in her head.

9:00am: the inbox and customer enquiries

The office manager works through the shared inbox. Overnight, each new email was categorised (new enquiry, complaint, schedule change, supplier, spam) and routine ones have a draft reply sitting underneath.

She edits most drafts lightly and rewrites perhaps one in five. The drafts are good at tone and structure and bad at anything they were not told, such as whether a particular site can actually take an extra Saturday clean. That is fine. The draft saves the typing; she supplies the judgement.

• Saves time: schedule-change confirmations, quote acknowledgements, chasing missing access details
• Saves a little: complaint replies, where the draft gives a calm starting point but always needs rewriting
• Saves nothing: the one furious email from a key account, which she phones about instead

11:00am: a meeting that produces actions

The weekly operations meeting runs for 40 minutes on Teams. The transcript goes through a summary step that pulls out decisions and action items with an owner and a date, then posts them into the team's task list for someone to confirm.

The operations manager spends two minutes correcting it. Twice this month it attributed an action to the wrong person, which is exactly why confirmation is a step and not an afterthought. Nobody writes minutes any more, and, more usefully, actions stop evaporating between meetings.

2:00pm: a quote and a proposal

The sales lead has a site survey from a new office building: photos, a floor plan and his own voice notes. He asks the assistant to turn his notes into the standard proposal structure, using two past proposals as a style reference.

The first draft is serviceable in ten minutes. The pricing he does himself in the costing spreadsheet, because nobody at this firm lets a language model near a number that goes on a contract. He spends another 40 minutes making the proposal specific to the client, which is the part that wins work. Previously the whole thing took most of an afternoon.

The AI writes the parts every proposal has. The person writes the parts that make this one worth signing.

4:00pm: admin, documents and the long tail

The afternoon is the long tail of small tasks where AI helps a little each time:

• Turning a supervisor's rambling voice note about a new site into a tidy checklist for the next visit
• Writing a job advert for a part-time evening cleaner, then rewriting it to sound less like every other advert
• Explaining a spreadsheet formula the previous office manager left behind
• Summarising a 30-page tender document into the six requirements that decide whether to bid
• Updating a standard operating procedure after a change to the key-holding process

None of these is dramatic. Added together they are where most of the saved time comes from, and they are also where most of the mistakes happen, because small tasks get less checking.

What the day adds up to, honestly

Here is a rough, illustrative tally for the two office staff and the sales lead. Your numbers will differ; the shape usually does not.

Roughly four hours saved across three people on a day that includes a proposal, and nearer two on a day that does not. That is worth having. It is not the replacement of a member of staff, and anyone who sells it to you as one is describing a different business.

What this business deliberately does not do

The restraint is as instructive as the adoption. They do not let AI send anything to a customer without a person reading it. They do not use it for pricing, payroll or anything in Xero beyond reading overdue invoices. They do not paste client contracts into consumer AI tools, and they have a one-page policy saying so.

When we help a firm like this, the first thing we check at SpiderHunts is where information is going and who approves what leaves the building. The integrations are small; our AI integration work for businesses this size is usually a few focused weeks, not a transformation programme. If you want the governance side first, our guide to writing an AI policy for your business is a sensible place to start.

Frequently asked questions